A method and system for identifying the state of a power transmission line

Abstract

This invention discloses a method and system for identifying the state of transmission lines, applied in the field of transmission line technology. The method includes: acquiring multi-dimensional historical data of a target transmission line; analyzing the trend characteristics of each dimension of historical data and the correlation characteristics between the historical data of each dimension, processing the trend characteristics and correlation characteristics to obtain the risk result of the target transmission line, and determining high-risk points based on the risk result; acquiring multi-dimensional real-time electrical data of the high-risk points, performing correlation analysis on the multi-dimensional real-time electrical data to determine the initial fault state; determining the first state identification result of the high-risk points; inputting the acquired real-time multimodal image data of the target transmission line into a pre-constructed fault state identification model to obtain the second state identification result, thus obtaining the state identification result of the target transmission line. The method and system for identifying the state of transmission lines provided by this invention effectively improve the efficiency of state identification of transmission lines.

Description

Technical Field

[0001] This invention relates to the field of power transmission line technology, and in particular to a method and system for identifying the status of power transmission lines. Background Technology

[0002] Transmission lines are a critical component of the power system, and their operational status directly affects the safety, reliability, and power supply quality of the power grid. Accurate assessment of the condition of transmission lines is of great significance for ensuring the safe and stable operation of the power grid.

[0003] Existing technologies for assessing the condition of transmission lines rely on methods such as manual inspections to conduct comprehensive and unified assessments of the transmission lines. For example, maintenance personnel need to regularly inspect each tower and base along the line. However, such assessment methods are time-consuming and labor-intensive, with a long coverage period, resulting in the inefficient use of a large amount of human, material, and time resources. Consequently, the assessment efficiency is greatly reduced, making it difficult to support the high-frequency and refined condition assessment requirements of large-scale transmission networks. Summary of the Invention

[0004] This invention provides a method and system for identifying the status of transmission lines, in order to solve the technical problem of low efficiency in existing methods for identifying the status of transmission lines, thereby improving the efficiency of identifying the status of transmission lines.

[0005] To address the aforementioned technical problems, this invention provides a method and system for identifying the status of transmission lines, the method comprising: Obtain multi-dimensional historical data of the target transmission line; Analyze the trend characteristics of historical data for each dimension and the correlation characteristics between the historical data for each dimension, process the trend characteristics and the correlation characteristics to obtain the risk results of the target transmission line, and use the risk results to determine high-risk locations; Obtain multidimensional real-time electrical data of the high-risk locations, perform correlation analysis on the multidimensional real-time electrical data, and determine the initial fault state; The multidimensional real-time electrical data is analyzed based on the trend features and the correlation features to determine the fault development trend. Based on the fault development trend and the initial fault state, the first state identification result of the high-risk point is determined. The real-time multimodal image data of the target transmission line is obtained and input into the constructed fault state identification model to obtain the second state identification result. The first state identification result and the second state identification result are then integrated and processed to obtain the state identification result of the target transmission line.

[0006] Preferably, the step of analyzing the trend characteristics of historical data in each dimension and the correlation characteristics between the historical data in the dimensions, processing the trend characteristics and the correlation characteristics to obtain the risk result of the target transmission line, and determining high-risk locations based on the risk result includes: Analyze the trend characteristics of historical data for each dimension; By analyzing the correlation coefficients of any two trend features, the correlation features between the historical data of the dimension can be obtained; The trend characteristics and correlation characteristics are input into the constructed fusion analysis model to obtain the comprehensive risk value; Based on the comprehensive risk value, the target transmission line is divided into regions to determine the high-risk locations.

[0007] Preferably, the step of performing correlation analysis on the multidimensional real-time electrical data to determine the initial fault state includes: The multidimensional real-time electrical data is preprocessed to obtain a time-aligned observation sequence; Numerical analysis is performed on each of the observation sequences to determine the first anomalous feature; By performing collaborative analysis on any two of the observed sequences, a second anomaly feature can be determined. The initial fault state is determined based on the first abnormal feature and the second abnormal feature.

[0008] Preferably, the step of analyzing the multidimensional real-time electrical data based on the trend features and the correlation features to determine the fault development trend includes: The trend characteristics of the multidimensional real-time electrical data and the corresponding historical data of the same dimension are compared to determine the initial fault development trend; Based on the correlation features, the collaborative evolution among the multidimensional real-time electrical data is analyzed, and the analysis results are used to correct the initial fault development trend to obtain the fault development trend.

[0009] Preferably, the construction process of the fault state identification model includes: Using the trend features and the correlation features as enhancement criteria, the text data in the multidimensional historical data is converted into image data to generate a dataset to expand the sample. The neural network model is trained based on the original dataset and the expanded samples in the multidimensional historical data to obtain the fault state identification model.

[0010] Another aspect of the present invention provides a power transmission line status identification system, comprising: The acquisition module is used to acquire multidimensional historical data of the target transmission line; The high-risk module is used to analyze the trend characteristics of historical data in each dimension and the correlation characteristics between the historical data in the dimensions, process the trend characteristics and the correlation characteristics to obtain the risk results of the target transmission line, and determine the high-risk locations based on the risk results; The initial module is used to acquire multi-dimensional real-time electrical data of the high-risk locations, perform correlation analysis on the multi-dimensional real-time electrical data, and determine the initial fault state. The first state module is used to analyze the multidimensional real-time electrical data based on the trend features and the correlation features, determine the fault development trend, and determine the first state identification result of the high-risk point based on the fault development trend and the initial fault state. The integration module is used to input the acquired real-time multimodal image data of the target transmission line into the constructed fault state identification model to obtain a second state identification result, and to integrate the first state identification result and the second state identification result to obtain the state identification result of the target transmission line.

[0011] Preferably, the high-risk module includes: Trend unit, used to analyze the trend characteristics of historical data for each dimension; The correlation unit is used to analyze the correlation coefficient of any two trend features to obtain the correlation features between the historical data of the dimension. The model unit is used to input the trend features and the correlation features into the constructed fusion analysis model to obtain a comprehensive risk value; The division unit is used to divide the target transmission line into regions based on the comprehensive risk value and determine the high-risk locations.

[0012] Preferably, the step of performing correlation analysis on the multidimensional real-time electrical data to determine the initial fault state includes: A preprocessing unit is used to preprocess the multidimensional real-time electrical data to obtain a time-aligned observation sequence; The first unit is used to perform numerical analysis on each of the observation sequences to determine the first anomalous feature; The first unit is used to perform collaborative analysis on any two of the observation sequences to determine the first anomalous feature; The fault unit is used to determine the initial fault state based on the first abnormal feature and the second abnormal feature.

[0013] Preferably, the step of analyzing the multidimensional real-time electrical data based on the trend features and the correlation features to determine the fault development trend includes: The determining unit is used to compare the trend characteristics of the multidimensional real-time electrical data and the corresponding historical data of the dimensions to determine the initial fault development trend; The correction unit is used to analyze the collaborative evolution among the multidimensional real-time electrical data based on the correlation features, and to correct the initial fault development trend based on the analysis results, thereby obtaining the fault development trend.

[0014] Preferably, the construction process of the fault state identification model includes: An expansion unit is used to convert text data in the multidimensional historical data into image data and generate dataset expansion samples by using the trend features and the correlation features as enhancement criteria. The training unit is used to train the neural network model based on the original dataset and the expanded samples of the multidimensional historical data to obtain the fault state recognition model.

[0015] Compared with the prior art, the beneficial effects of the present invention are at least one of the following: This invention acquires multi-dimensional historical data of a target transmission line; analyzes the trend characteristics of each dimension of historical data and the correlation characteristics between the historical data of the dimensions, processes the trend characteristics and correlation characteristics to obtain the risk results of the target transmission line, and uses the risk results to identify high-risk points, realizing the transformation from indiscriminate inspection of the entire line to precise focus based on risk; then, it acquires multi-dimensional real-time electrical data of the high-risk points, performs correlation analysis on the multi-dimensional real-time electrical data to determine the initial fault state; based on the trend characteristics and correlation characteristics, it analyzes the multi-dimensional real-time electrical data to determine the fault development trend, forming a dynamic evolution perception capability; simultaneously, it inputs the acquired real-time multimodal image data of the target transmission line into a pre-constructed fault state recognition model to obtain a second state recognition result, and integrates the first state recognition result and the second state recognition result to achieve the fusion of full coverage and key monitoring. This solution effectively improves the efficiency, accuracy, and foresight of the assessment, significantly reduces the consumption of manpower and material resources, and meets the technical requirements of high-frequency, refined, and intelligent operation and maintenance of large-scale power grids. Attached Figure Description

[0016] Figure 1 This is a flowchart illustrating the state identification method for power transmission lines in one embodiment of the present invention. Figure 2 This is a schematic diagram of the structure of a Transformer-based method for analyzing the health status of transmission line insulators in one embodiment of the present invention. Figure 3 This is a schematic diagram of the structure of a power transmission line status identification system in one embodiment of the present invention; Figure label: Among them, 11. Acquisition module; 12. High-risk module; 13. Initial module; 14. First state module; 15. Integration module. Detailed Implementation

[0017] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. The purpose of providing these embodiments is to make the disclosure of the present invention more thorough and comprehensive. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0018] In the description of this invention, the terms "first," "second," "third," etc., are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Thus, a feature defined with "first," "second," "third," etc., may explicitly or implicitly include one or more of that feature. In the description of this invention, unless otherwise stated, "a plurality of" means two or more.

[0019] In the description of this invention, it should be noted that, unless otherwise explicitly specified and limited, the terms "installation," "connection," and "linking" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; and they can refer to the internal communication between two components. The term "and / or" as used herein includes any and all combinations of one or more of the associated listed items. Those skilled in the art will understand the specific meaning of the above terms in this invention based on the specific circumstances.

[0020] In the description of this invention, it should be noted that, unless otherwise defined, all technical and scientific terms used in this invention have the same meaning as commonly understood by one of ordinary skill in the art. The terminology used in this specification is for the purpose of describing specific embodiments only and is not intended to limit the invention. Those skilled in the art can understand the specific meaning of the above terms in this invention based on the specific circumstances.

[0021] Transmission lines are crucial to the power grid, and their condition directly affects the safety and quality of power supply. Traditional manual inspection methods require checking each tower and each base, which is time-consuming, labor-intensive, has a long coverage period, and is inefficient. It not only consumes a lot of manpower and resources, but also cannot meet the urgent needs of large-scale transmission networks for high-frequency and refined condition assessment.

[0022] One embodiment of the present invention provides a method for identifying the status of a power transmission line. For details, please refer to [link to relevant documentation]. Figure 1 , Figure 1 The diagram shown is a flowchart illustrating a power transmission line status identification method according to one embodiment of the present invention, including: S1. Obtain multi-dimensional historical data of the target transmission line; S2. Analyze the trend characteristics of historical data for each dimension and the correlation characteristics between historical data of each dimension. Process the trend characteristics and correlation characteristics to obtain the risk results of the target transmission line, and use the risk results to determine high-risk locations. S3. Obtain multi-dimensional real-time electrical data of high-risk locations, perform correlation analysis on the multi-dimensional real-time electrical data, and determine the initial fault state; S4. Analyze multidimensional real-time electrical data based on trend features and correlation features to determine the fault development trend, and determine the first state identification result of high-risk points based on the fault development trend and the initial fault state. S5. Input the acquired real-time multimodal image data of the target transmission line into the constructed fault state identification model to obtain the second state identification result, and integrate the first state identification result and the second state identification result to obtain the state identification result of the target transmission line.

[0023] First, acquire multidimensional historical data of the target transmission line. Acquiring this data is the fundamental prerequisite for the entire condition identification scheme. Its core is collecting historical operational data from different dimensions throughout the line's entire lifecycle, providing data support for subsequent risk analysis and condition assessment. Specifically, multidimensional historical data includes electrical and non-electrical quantity data. Electrical quantity data covers line current, voltage, power, frequency, and phase angle, while non-electrical quantity data includes environmental data, line physical condition data, and operation and maintenance data. Environmental data involves temperature, humidity, wind speed, precipitation, and icing thickness along the line. Line physical condition data includes historical monitoring data for conductors, insulators, hardware, and towers. Operation and maintenance data includes fault records, maintenance records, and test data. The specific implementation method requires combining multi-source data acquisition equipment and storage systems. Real-time historical data is collected through fault recording devices, intelligent sensors, and online monitoring terminals deployed along transmission lines. Simultaneously, it connects to the power company's Production Management System (PMS), Energy Management System (EMS), and equipment lifecycle management platform to extract historical operation and maintenance data. The collected multi-source heterogeneous data undergoes cleaning, deduplication, standardization, and integration processing. The processed data is then stored in a time-series database to ensure data integrity, consistency, and availability. This approach provides a high-quality data foundation for subsequent trend and correlation analysis, ensuring the accuracy of risk identification and status assessment. Furthermore, the accumulation of historical data allows for the discovery of line operation patterns, providing a reliable basis for fault prediction and status evaluation.

[0024] Secondly, the trend characteristics of historical data for each dimension and the correlation characteristics between historical data of each dimension are analyzed. These trend characteristics and correlation characteristics are processed to obtain the risk results for the target transmission line, and high-risk locations are identified based on these results. The analysis involves analyzing the trend characteristics of historical data for each dimension; analyzing the correlation coefficients of any two trend characteristics to obtain the correlation characteristics between historical data of each dimension; inputting the trend characteristics and correlation characteristics into the constructed fusion analysis model to obtain a comprehensive risk value; and dividing the target transmission line into regions based on the comprehensive risk value to determine high-risk locations. The core of analyzing the trend characteristics of historical data for each dimension and the correlation characteristics between historical data of each dimension is to extract the line operation patterns through data mining, and combine this with feature fusion to achieve risk quantification and high-risk location positioning, providing precise targeting for subsequent fault identification.

[0025] Trend characteristics refer to the changing patterns of data in various dimensions over time, such as the fluctuation trends of current and voltage, the increasing trend of icing thickness, and the decreasing trend of equipment status parameters. Time series analysis methods are used in the analysis, specifically including trend fitting and smoothing, and anomaly detection. Linear and nonlinear regression fitting algorithms are used to capture long-term data trends, moving average and exponential smoothing methods are used to eliminate data noise, and the 3σ and Grubbs criteria are used to identify abnormal fluctuation points, accurately extracting the rising, falling, and stable trend characteristics of each dimension of data. Correlation characteristics refer to the mutual influence relationships between historical data in different dimensions, such as the positive correlation between ambient temperature and line resistance, and the correlation between precipitation and insulator flashover risk. Correlation coefficients are calculated for any two trend characteristics. Specifically, Pearson correlation coefficients are used to analyze the degree of linear correlation, Spearman's rank correlation coefficients to analyze the degree of nonlinear correlation, and Kendall's concordance coefficients to verify correlation stability. Mutual information methods are combined to mine potential non-explicit correlation characteristics, forming a complete set of correlation characteristics. The fusion analysis model employs a machine learning-based ensemble model, integrating algorithms such as random forest gradient boosting trees and support vector machines. Extracted trend and correlation features are used as model input, and multi-feature fusion is achieved through feature weight allocation. Before model training, cross-validation is used to optimize hyperparameters, improving the model's generalization ability. The model outputs a comprehensive risk value for each region, reflecting the probability and impact of a fault in that region. Based on the comprehensive risk value, clustering analysis is used to divide the transmission line into regions. Regions with comprehensive risk values ​​exceeding a set threshold are designated as high-risk locations. Simultaneously, the location determination results are optimized by considering the line topology, equipment aging, and operational environment complexity, ensuring the accuracy and targeted nature of high-risk location identification. This approach leverages data feature mining to achieve quantitative risk assessment, accurately pinpointing high-fault areas and providing a focused direction for subsequent real-time monitoring and maintenance decisions, thereby improving the efficiency and reliability of line status identification.

[0026] Next, multidimensional real-time electrical data of high-risk locations is acquired, and correlation analysis is performed on this data to determine the initial fault state. The trend characteristics of the multidimensional real-time electrical data are compared with those of the corresponding historical data to determine the initial fault development trend. Based on the correlation characteristics, the co-evolution among the multidimensional real-time electrical data is analyzed, and the analysis results are used to correct the initial fault development trend, thus obtaining the final fault development trend. The core of acquiring multidimensional real-time electrical data of high-risk locations and conducting correlation analysis to determine the initial fault state is to focus on key areas to capture real-time operational anomalies and combine historical characteristics to correct trend judgments, ensuring the timeliness and accuracy of fault analysis.

[0027] Multidimensional real-time electrical data includes real-time current, voltage, power, phase angle, and leakage current at high-risk locations. This data is collected in real-time by RTUs (Remote Terminal Units) and FTUs (Feeder Terminal Units) deployed at critical locations such as pole insulators, and transmitted to the data processing center via 5G or fiber optic communication, ensuring low latency and high reliability. When performing correlation analysis on the multidimensional real-time electrical data, time-series association rule mining algorithms and causal analysis methods are employed, combined with preset fault electrical quantity thresholds, to identify abnormal correlation patterns between data, such as the coordinated anomaly of sudden current increases and voltage drops. This allows for the determination of initial fault states, such as short-circuit grounding insulation damage. The initial fault state refers to the preliminary judgment of the fault type and severity based on real-time data. The initial fault development trend refers to the direction and rate of change from fault occurrence to propagation. This is achieved by comparing the trend characteristics of multidimensional real-time electrical data with historical data of the corresponding dimensions. A dynamic time warping algorithm is used to calculate the similarity between real-time data and historical trend curves. Combined with the sliding window method, data changes are tracked in real-time to capture early signs of the fault trend and determine whether the fault is continuously deteriorating and slowly mitigating or maintaining a stable initial trend. Co-evolutionary analysis refers to the mutual influence patterns of real-time electrical data across different dimensions over time. Based on previously mined correlation features, a coupling coordination degree model is used to analyze the degree of coordinated change in various electrical quantities. A Bayesian network algorithm is then combined to quantify the influence weights of correlation features on real-time data evolution. This corrects the initial fault development trend, eliminates the bias of single-data-trend analysis, and yields a more realistic fault development trend. This approach accurately captures the real-time fault status of high-risk locations, dynamically corrects fault development predictions, and provides precise data for subsequent status identification and maintenance, thereby enhancing the initiative and effectiveness of line fault prevention and control.

[0028] Then, based on trend and correlation features, multidimensional real-time electrical data is analyzed to determine the fault development trend. Based on the fault development trend and initial fault state, the first state identification result of high-risk points is determined. Trend and correlation features are used as enhancement criteria to convert text data in multidimensional historical data into image data, generating expanded dataset samples. The neural network model is trained based on the original dataset and expanded dataset samples from multidimensional historical data to obtain the fault state identification model. The core of this method is to rely on historical features to enhance the interpretation of real-time data, achieving accurate preliminary judgment of the fault state.

[0029] The first-state identification result refers to a comprehensive judgment of the severity and development trend of the fault type based on electrical data dimensions. The analysis employs a real-time data parsing method that integrates historical features. Trend features are used to determine whether changes in real-time electrical data conform to historical fault evolution patterns. Correlation features are used to verify whether the coordinated changes in real-time data across various dimensions match fault correlation patterns. A weighted fusion algorithm integrates the analysis results of the two types of features, correcting biases in judgments based on single real-time data, clarifying fault development trends, and then combining the initial fault state with a multi-criteria decision-making method to comprehensively determine the first-state identification result, ensuring the reliability of the result. Text data refers to unstructured data such as fault records, maintenance reports, and test instructions from multi-dimensional historical data. Dataset augmentation samples are new training data generated using text-to-image technology. Trend and correlation features are used as enhancement criteria to ensure that the generated samples conform to the inherent patterns of the data and avoid sample distortion. Specifically, word embedding technology is used to convert text data into vector form, and a convolutional neural network is used to extract key information from the text. Generative adversarial networks generate image data with a distribution consistent with the original data. Feature constraints are introduced to ensure that the generated samples carry trend and correlation feature information, forming a complete training set together with the original dataset. The neural network model employs a hybrid approach combining GCN (Graph Convolutional Network) and LSTM (Long Short-Term Memory). GCN extracts spatial features from the data, while LSTM captures time-series features. During training, the original dataset and its augmented samples are proportionally divided into training, validation, and test sets. An adaptive moment estimation optimization algorithm is used to adjust the model parameters, and the prediction error is calculated using the cross-entropy loss function. The model performance is iteratively optimized, and after multiple training and validation iterations, a fault state recognition model is obtained. This model can accurately identify the feature patterns corresponding to different faults. This approach enhances the accuracy of real-time data analysis through historical features, improves the model's generalization ability by leveraging dataset augmentation, lays the foundation for subsequent two-dimensional state recognition, and improves overall recognition accuracy.

[0030] Finally, the acquired real-time multimodal image data of the target transmission line is input into the constructed fault state recognition model to obtain the second state recognition result. The first and second state recognition results are then integrated to obtain the state recognition result of the target transmission line. Real-time multimodal image data refers to image and video data capturing the state of the target transmission line from different dimensions, specifically including visible light images, infrared thermal images, and ultraviolet images. Visible light images are used to identify conductor breakage, insulator damage, and tower structural defects; infrared thermal images are used to detect abnormal equipment heating, such as overheating of joints and aging of insulators; and ultraviolet images are used to capture insulator corona discharge phenomena. Data is collected in real-time through UAV inspections and fixed-point monitoring equipment. After image preprocessing algorithms such as noise reduction, enhancement, and segmentation to remove redundant information and improve image quality, the data is input into the constructed fault state recognition model. The fault state recognition model extracts spatial and semantic features from the multimodal images to accurately match fault modes and outputs the second state recognition result. The second state recognition result indicates the fault type, location, and severity determined based on the image dimension, compensating for the limitations of single electrical data dimension recognition. The first and second state identification results are integrated and processed using an evidence-based fusion algorithm to quantify the credibility weights of the two types of results. Consistent identification results are strengthened, while conflict identification results are further verified by incorporating trend and correlation features. Simultaneously, line topology and historical maintenance data are introduced to optimize the integration logic, ultimately yielding the state identification result of the target transmission line. This result balances the real-time nature of electrical data with the intuitiveness of image data, achieving a comprehensive and accurate determination of fault status. This approach compensates for the shortcomings of a single data source through dual-dimensional result fusion, improving the accuracy and reliability of state identification and providing a comprehensive and scientific decision-making basis for transmission line operation and maintenance.

[0031] Another embodiment of the present invention provides a method for analyzing the health status of transmission line insulators based on Transformer. Please refer to [link to relevant documentation]. Figure 2 , Figure 2The diagram illustrates a Transformer-based method for analyzing the health status of transmission line insulators in one embodiment of the present invention. First, a UAV periodically collects visible light, infrared, and ultraviolet multimodal data from all insulators to establish a historical health database containing temporal characteristics. Then, based on the historical data, a risk quantification assessment is performed on each insulator to identify high-risk locations, and sensors are deployed at these locations for real-time monitoring. The sensors continuously collect real-time data such as leakage current, temperature, and strain, and transmit this data to a backend server. The deployed Transformer model analyzes and determines the fault type, location, and severity. Simultaneously, for low-risk insulators without deployed sensors, the UAV still periodically inspects and collects multimodal data, using a simplified airborne Transformer model for preliminary judgment. If a suspected fault is found, it is uploaded to the backend for in-depth analysis. The entire system also supports dynamic updates to model parameters and risk thresholds based on year-round operational data, achieving continuous optimization and thus constructing an intelligent operation and maintenance system of "historical modeling—risk identification—key monitoring—real-time diagnosis—edge screening—closed-loop optimization." Specifically: 1. Establish a historical health database, which contains time-series characteristic data of all insulators on the transmission line. The time-series characteristic data is collected by UAVs and includes visible light time-series characteristics, infrared time-series characteristics, and ultraviolet time-series characteristics.

[0032] Specifically, a drone equipped with a visible light camera, an infrared thermal imager, and an ultraviolet imager was used. The visible light camera had a resolution of 4K, the infrared thermal imager had a temperature measurement range of 20℃ to 150℃ with an accuracy of ±0.5℃, and the ultraviolet imager had a minimum detectable discharge level of 5pC. The drones inspected the transmission line at 5m intervals and 10m altitudes, collecting three types of raw data from the insulators on each transmission tower. Feature extraction was performed on the collected raw data to construct a historical health dataset H for the insulators. This dataset consists of three types of time-series features from multiple time points, including visible light features, infrared features, and ultraviolet features. Specifically, visible light features include appearance defect indicators such as crack length, contamination coverage area, and degree of corrosion of metal fittings, extracted from visible light images; infrared features include thermal parameters such as maximum temperature, temperature standard deviation, and hot spot area, derived from the infrared temperature distribution matrix; ultraviolet features include local discharge indicators such as maximum discharge intensity, number of discharges, and discharge area, derived from the ultraviolet discharge intensity matrix.

[0033] The entire historical health dataset is updated every three months, chronologically, for a total of four collections, forming a complete annual monitoring sequence. Each insulator in the dataset is assigned a unique number, and the data for all insulators is uniformly organized into a structured database for subsequent risk modeling and intelligent analysis. This dataset not only includes the current state of the insulators but also records their evolutionary trends over time, providing ample evidence for assessing their health status.

[0034] 2. Perform risk quantification assessment on the time-series characteristic data of each insulator in the historical health database to obtain the risk value of each insulator.

[0035] When performing risk quantification assessment on time-series characteristic data, the risk value of each insulator can be quantified from three dimensions: defect dimension, environmental risk dimension, and service life dimension. During the quantification process, a long short-term memory network is used to analyze the data change trend of each dimension, and a graph convolutional network is used to analyze the influence relationship between multiple dimensions. After splicing the trend features and correlation features obtained by these two methods, the risk value of each insulator required by this application can be obtained.

[0036] Specifically, the above method includes the following steps: 2.1 Multi-dimensional historical time series data preprocessing Extract textual or numerical time-series data for each insulator i from the historical health database, considering the dimensions of defect, environmental risk, and service life, to form the input sequence.

[0037] 2.2 Text-to-Image Mapping Conversion A time-series-grayscale mapping method is used to convert a one-dimensional text data matrix into a two-dimensional grayscale image to preserve time-series trend features.

[0038] 2.3 Image Reconstruction and Sample Augmentation The image is reconstructed and expanded based on single-dimensional historical time-series trend constraints and multi-dimensional correlation influence constraints, ensuring that the reconstructed image retains the original data patterns while conforming to the physical logic of insulator degradation. This step includes the following sub-steps.

[0039] 2.3.1 Constraint Information Extraction (1) Extraction of single-dimensional historical time series trend constraints In the extraction of single-dimensional historical time-series trend constraints, the system calculates two key indicators for each insulator's raw time-series data across three dimensions: defects (D), environmental risk (E), and service life (A): trend slope and fluctuation amplitude. The trend slope is obtained by averaging the changes in values ​​at adjacent time points; it reflects the overall direction of change of this dimension's indicator over time—a result greater than zero indicates continued deterioration or an increase in value; zero indicates stability; and a result less than zero indicates improvement or a decline. Fluctuation amplitude is the average of the absolute values ​​of changes at adjacent time points, used to measure the degree of oscillation or instability of the data in this dimension during the observation period.

[0040] (2) Extraction of multi-dimensional correlation influence constraints Regarding the extraction of constraints on multi-dimensional correlations, the system further analyzes the interrelationships between different dimensions. Specifically, it calculates the Pearson correlation coefficient between data from any two different dimensions (e.g., defects and environment, environment and aging). This coefficient is obtained by dividing the covariance by the product of the standard deviations of the two dimensions, and its value ranges from negative one to positive one. A positive coefficient indicates a positive correlation between the two dimensions, meaning that deterioration in one dimension often accompanies deterioration in the other; a negative coefficient indicates a negative correlation; and a coefficient close to zero indicates that the two are essentially unrelated. This correlation information is used as a coupling constraint that must be followed during subsequent data augmentation.

[0041] 2.3.2 Calculation of constraint-type enhancement parameters.

[0042] During sample augmentation, the system calculates a set of adjustment parameters for each newly generated augmented sample, conforming to the two types of constraints mentioned above. For a grayscale image in a specific dimension, the adjustment of pixel values ​​consists of two parts: one is a linear scaling term based on the original trend slope of that dimension, i.e., the trend slope multiplied by a learnable weight coefficient α, and then multiplied by the maximum grayscale value of 255, used to preserve the original evolution direction; the other part is a bias term β, used to introduce reasonable random perturbations. The entire adjustment process ensures that the generated new samples not only continue a reasonable time trend in a single dimension but also maintain a physically reliable correlation structure across multiple dimensions, thereby avoiding the generation of unrealistic false data.

[0043] 2.3.3 Constrained Image Reconstruction.

[0044] During the image reconstruction stage, the system adjusts each pixel in the original grayscale image based on the previously calculated enhancement parameters to generate a new image. The adjustment considers the pixel's position within the entire observation sequence and assigns it a weight accordingly. If the original data shows an overall deteriorating trend, pixels corresponding to later time points (i.e., more recent data) are given higher weights; if the trend is positive or improving, earlier pixels have higher weights; if the data is relatively stable with no significant changes, the weights for all time points remain consistent.

[0045] 2.4 Image-to-text inverse conversion.

[0046] The reconstructed image is restored to text / numerical time-series data to obtain the expanded time-series sequence.

[0047] 2.5 Expanding and integrating the sample set.

[0048] For the three dimensions of defects, environmental risks, and service life, the system will generate multiple reconstructed images. These new images must meet two types of physical rationality constraints: First, in a single dimension, the direction of change reflected by the reconstructed time series (e.g., whether it is continuous deterioration or tending to stabilize) must be consistent with the original data, and its fluctuation level cannot exceed twice that of the original fluctuation; second, among multiple dimensions, the correlation between them (e.g., whether the aggravation of defects is accompanied by an increase in temperature) must also remain unchanged—positive correlations must remain positive, negative correlations must remain negative, and the change in the strength of the correlation cannot be too large.

[0049] After image reconstruction, the system converts these two-dimensional images back into one-dimensional time-series data. This step is equivalent to "translating" the images back into numerical form, allowing the enhanced data to be directly used by subsequent analysis models, just like processing the original acquired data.

[0050] Finally, the system merges the original samples and all new samples generated through the above methods to form an expanded complete dataset. This dataset contains the original state as well as a variety of reasonable variations that conform to physical laws, preserving real evolutionary characteristics while increasing sample diversity, thus effectively supporting the training and accurate evaluation of subsequent models under small sample conditions.

[0051] In the scenario of insulator defect detection, this application adopts a "text-image-text" processing method in the above steps, which solves the core technical bottleneck of the scenario. It is closely aligned with the insulator defect detection of this application in terms of data characteristic adaptation, breakthrough of sample scarcity, model input connection, and detection reliability assurance, and ultimately achieves the effect of improving the accuracy of insulator risk value calculation.

[0052] First, this method is adapted to the continuous temporal characteristics of insulator degradation. Insulator defects, such as crack propagation and accelerated aging, are gradual processes. The original one-dimensional text / numerical time-series data easily loses trend details, such as the continuity of the deterioration rate and precursors of abrupt changes. By mapping it to a two-dimensional grayscale image, the temporal trend can be transformed into spatially continuous features. For example, an increase in grayscale corresponds to the continuous deterioration of defects, allowing LSTM to accurately capture long-term evolution patterns and avoid the distortion of trend features caused by directly processing one-dimensional data.

[0053] Secondly, it addresses the industry pain point of scarce insulator fault samples. Insulator fault rates are low, and historical databases contain very few labeled samples, making direct training of LSTM and GCN prone to overfitting. By using images as the medium, samples can be reconstructed under the physical constraints of single-dimensional time-series trends and multi-dimensional correlations, ensuring that newly added samples conform to degradation logic and avoiding unreasonable data generated by traditional text / numerical augmentation, thus supporting the model's generalization ability.

[0054] Furthermore, it bridges the input requirements of LSTM and GCN. LSTM requires one-dimensional time series to extract trend features, while GCN requires numerical correlation coefficients to model dimensional interactions. The aforementioned text-image-text process forms a closed loop, where text→image achieves constrained expansion, and image→text restores the data form to conform to the input of both models. If this step is skipped, the two models cannot fuse features due to input mismatch, and the risk value calculation fails.

[0055] Finally, it ensures high reliability in transmission line testing. Insulator risk assessment is directly related to power grid safety and requires a low false / false positive rate. This method avoids the false positives or false negatives caused by traditional methods, such as single-model or unconstrained expansion, by preserving temporal details through image visualization, supplementing samples with constraints, and fusing features from dual models, thus meeting the rigid requirements of the scenario.

[0056] 2.6. LSTM module extracts single-dimensional historical trend features. The time series expanded with three dimensions are modeled using LSTM networks to capture the long-term evolution trend of each dimension.

[0057] This step specifically includes the following sub-steps: 2.6.1 LSTM Network Structure and Output In the model structure design, each evaluation dimension (such as defect, environmental risk, and service life) corresponds to an independent LSTM network unit, and the units do not share parameters. Taking the defect dimension as an example, LSTM processes temporal data through three control mechanisms: input gate, forget gate, and output gate. The input gate determines how much new information is incorporated into memory at the current time step; the forget gate determines how much information from past memories is discarded; and the output gate controls the output of the current memory state to the next time step. These gating mechanisms are calculated using weight matrices and bias vectors, where the dimension of the weight matrix is ​​related to the size of the hidden layer, and the bias vector is set accordingly. All operations are based on element-wise multiplication and activation functions.

[0058] 2.6.2 Trend Feature Output The internal state of an LSTM consists of cell states and hidden states. Cell states are used for long-term memory, while hidden states are the information output after gating. At each time step, the model updates both cell states and hidden states based on the current input and the hidden states from the previous time step. Finally, the hidden state of each LSTM network at its last time step is extracted as the historical trend feature for that dimension. For example, the long-term trend feature for the defect dimension is generated by its corresponding LSTM output; similarly, the environmental risk and service life dimensions also generate their own long-term trend features.

[0059] 2.7 Modeling the Inter-Dimensional Relationships Using Graph Convolutional Networks. A "dimensional relationship graph" is constructed, and graph convolutions are used to learn the mutual influence weights of different dimensions on risk. This step includes the following sub-steps: 2.7.1 Construction of Dimensional Relationship Graph The system constructs a "dimensional correlation graph" to model the interrelationships among the three assessment dimensions: defects, environmental risks, and service life. This graph consists of a set of nodes and a set of edges: nodes represent data from the three dimensions, and each node's feature is the historical trend characteristic of the corresponding LSTM network output; edges represent the correlations between different dimensions. Elements in the initial adjacency matrix are greater than zero, indicating a positive influence of one dimension on another. These initial weights are statistically initialized by analyzing the co-occurrence patterns of various factors in historical fault data.

[0060] 2.7.2 Calculating Correlated Features in Graph Convolutional Layers The system uses a two-layer Graph Convolutional Network (GCN) to update node features, thereby fusing the mutual influence between different dimensions. During graph convolution, each node aggregates information from its neighbors, including its own features and those of the other two dimensions. Specifically, each node's new feature is the weighted sum of its neighbors' features, with the weights determined by the influence coefficients in the initial adjacency matrix. This process is calculated using the graph convolution weight matrix and bias vector, ultimately generating the updated associated features.

[0061] After graph convolution processing, the system outputs updated features in three dimensions: the comprehensive features of the defect dimension after being affected by environmental risks and service life, the features of the environmental risk dimension after being affected by the other two dimensions, and the features of the service life dimension after interacting with other dimensions. These features not only retain the historical evolution information of their respective dimensions but also incorporate cross-dimensional coupling effects, enabling a more comprehensive reflection of the complex change mechanisms of insulator health status.

[0062] 2.8 Feature Fusion and Risk Value Calculation. The "historical trend features" of LSTM and the "dimensional correlation features" of GCN are fused, and the final risk value is output through a fully connected layer.

[0063] This step includes the following sub-steps: 2.8.1 Feature Concatenation and Dimensional Compression The system concatenates the long-term trend features of each dimension with its associated features learned through a graph convolutional network. For example, the trend features of the defect dimension are merged with its associated features after being affected by environmental risks and service life, forming a more comprehensive feature vector. This process integrates single-dimensional temporal evolution information with the interaction relationships between multiple dimensions.

[0064] 2.8.2 Final Risk Value Calculation The concatenated feature vectors are input into a fully connected layer, where they are compressed and calculated using a weight matrix and bias terms to obtain a comprehensive score for that dimension. This score reflects the overall health assessment after considering its own evolutionary trends and the influence of external factors. This process is repeated for the three dimensions of defects, environmental risks, and service life, generating their respective comprehensive scores.

[0065] The system utilizes the inter-dimensional influence weights learned through graph convolutional networks to perform a weighted summation of the comprehensive scores across the three dimensions, yielding the final risk value. This weight represents the degree to which a particular dimension contributes to the overall risk; for example, environmental risks may have a significant impact on insulator degradation, thus warranting a higher weight. In this way, the system can quantify the importance of each factor, generating a unified and comparable risk assessment result for subsequent hierarchical management and decision support.

[0066] 2.9 Model Training and Optimization. The model training process uses labeled insulator samples from historical data as the training set. Samples that have experienced a fault are labeled as 1, and samples that are operating normally are labeled as 0. This binary labeling guides the model in learning how to distinguish between healthy and faulty states.

[0067] During the training process, the cross-entropy loss function is adopted to measure the difference between the model prediction results and the true labels. This loss function calculates the overall error value by comparing the predicted risk values of each sample with the actual labels. The goal is to minimize this error through an optimization algorithm to improve the prediction accuracy of the model.

[0068] The Adam algorithm is selected as the optimizer, and its learning rate is set to 0.001. During the training process, the system continuously iteratively updates all the parameters of the LSTM and GCN networks until the loss value converges, that is, the error no longer significantly decreases. This process enables the model to fully learn the laws of the evolution of the insulator health state from historical data and finally possess the ability to accurately evaluate risks.

[0069] 3. Determine the high-risk points among all insulators according to the risk values.

[0070] Exemplarily, in the embodiments of this application, the risk threshold Rth = 0.6 is set. If Ri ≥ Rth, it is determined as a high-risk point; if Ri < Rth, it is determined as a low-risk point.

[0071] 4. Deploy sensors at high-risk points.

[0072] Exemplarily, for high-risk points, a three-in-one sensor is deployed on the insulator string at this point, including a leakage current sensor, a temperature sensor, and a strain sensor. No sensor is deployed for low-risk points.

[0073] Furthermore, after the sensors are deployed, the data collected by the sensors is calibrated.

[0074] When calibrating, it is required that the calibration error of the leakage current sensor ≤ ±1 μA, the calibration error of the temperature sensor ≤ ±0.3 °C, and the calibration error of the strain sensor ≤ ±1 με.

[0075] 5. The sensors collect the real-time data of the insulators at the corresponding high-risk points according to the set frequency. The real-time data includes real-time leakage current, real-time temperature, and real-time strain values.

[0076] Exemplarily, the sensors at high-risk points continuously collect real-time data at a sampling frequency of 1 Hz and transmit it to the background server in real time via a 4G / 5G private network.

[0077] 6. Transmit the real-time data to the background server. The background server deploys a Transformer model, and the Transformer model analyzes the real-time data to determine the fault location and fault degree of the insulators.

[0078] As an example, this Transformer model contains a 6-layer encoder, a 3-layer decoder, and 8 self-attention heads. The model processes input data through a multi-head self-attention mechanism, enabling it to capture complex dependencies between different time points and spatial locations, thus achieving high-precision fault localization.

[0079] The model outputs two key results: first, a probability distribution matrix of the fault location, with dimensions consistent with the resolution of the visible light image. In this matrix, the value at each location represents the probability of a fault occurring in that area; a higher value indicates a higher probability of a fault occurring at that location. Second, a fault severity index, ranging from 0 to 10, used to quantify the severity level of the fault. For example, 0 to 3 represents a minor fault, 3 to 6 a moderate fault, and 6 to 10 a severe fault.

[0080] The model's input can be raw data directly collected by sensors or preprocessed feature vectors uploaded by the UAV. After processing, the input data generates a probability distribution of fault locations using the weight matrix and bias term of the location layer. Simultaneously, the system extracts multiple fault features (such as leakage current exceedance rate, abnormal temperature values, crack length ratio, discharge intensity, and strain changes) and standardizes them. Each feature is assigned a different weight based on its importance, and the weighted sum is used to obtain a fault severity index, achieving a comprehensive assessment of the fault state. Furthermore, when the backend server determines that an insulator is faulty based on real-time data, it sends instructions to the UAV to collect multi-modal data from multiple insulators located outside the high-risk insulator at the high-risk location, and determines whether the insulators whose multi-modal data has been collected are faulty.

[0081] 7. Use drones to periodically collect multimodal data on insulators outside of high-risk locations. The multimodal data includes visible light data, infrared data, and ultraviolet data.

[0082] For low-risk insulators without deployed sensors, the system uses drones for regular inspections to collect multimodal data. This data includes visible light images, infrared images, and ultraviolet discharge images, used to comprehensively assess the health status of the insulators.

[0083] During the inspection, the drones fly along the parameter paths set when the historical health database was established, ensuring the comparability and consistency of the collected data. The raw multimodal data collected undergoes preprocessing, including noise reduction, alignment, and standardization, for subsequent analysis.

[0084] The processed data is integrated into a feature vector containing visible light, infrared, and ultraviolet information of the insulator at the current moment. Each type of data is standardized to uniformly normalize its numerical range to between 0 and 1, facilitating model input and comparison. This standardization method employs Min-Max normalization technology, which can eliminate dimensional differences between different sensors, improving data consistency and model stability.

[0085] 8. A simplified Transformer model deployed on the UAV is used to perform preliminary analysis of the multimodal data to determine whether there is a fault in the insulator. If a fault is found, the backend server uses the Transformer model for in-depth processing to determine the location and severity of the fault in the insulator.

[0086] For example, the simplified Transformer model includes a 2-layer encoder and a 4-head self-attention. This simplified Transformer model can be simplified based on a Transformer model deployed in the backend server, or it can be simplified from a model different from the Transformer model deployed in the backend server; this application does not impose any specific limitations.

[0087] This simplified model receives feature vectors uploaded by the UAV as input and outputs the fault probability value of each insulator at the current moment after internal calculation, ranging from 0 to 1. The higher the value, the higher the probability that the insulator is faulty. To improve judgment efficiency, the system sets an initial fault threshold (e.g., 0.7). If the initially judged fault probability is greater than this threshold, the UAV will immediately upload the raw data to the backend server for further in-depth analysis by the full version of the Transformer model; if the probability is lower than the threshold, the insulator is considered to be in normal condition, and the data is only stored locally, and subsequently summarized and uploaded periodically, thereby reducing the communication burden.

[0088] After running for a period of time, the system will use accumulated historical data to update the parameters of the simplified Transformer model on the drone and the complete Transformer model on the backend server. This process is based on actual operating data throughout the year and continuously optimizes model performance through an online learning mechanism.

[0089] Specifically, the model's weight matrix and risk assessment threshold are iteratively adjusted based on the difference between the predicted results and the actual fault labels. During updates, the cross-entropy loss function is used to measure the prediction error, and the adjustment magnitude is controlled by the learning rate. Simultaneously, the risk threshold is dynamically calibrated based on the average risk value of insulators that actually experience faults throughout the year, ensuring it more closely reflects reality. Through continuous iteration, the model's fault identification accuracy and risk point location precision gradually improve, while the false positive and false negative rates continuously decrease, achieving system self-optimization and continuous evolution.

[0090] Another embodiment of the present invention provides a status identification system for transmission lines. For details, please refer to [link to relevant documentation]. Figure 3 , Figure 3 The diagram shown illustrates the structure of a power transmission line status identification system according to one embodiment of the present invention, comprising: Module 11 is used to acquire multidimensional historical data of the target transmission line; The high-risk module 12 is used to analyze the trend characteristics of historical data in each dimension and the correlation characteristics between historical data in each dimension. The trend characteristics and correlation characteristics are processed to obtain the risk results of the target transmission line, and the high-risk points are determined based on the risk results. Initial module 13 is used to acquire multi-dimensional real-time electrical data of high-risk locations, perform correlation analysis on the multi-dimensional real-time electrical data, and determine the initial fault state; The first state module 14 is used to analyze multidimensional real-time electrical data based on trend features and correlation features, determine the fault development trend, and determine the first state identification result of high-risk points based on the fault development trend and the initial fault state. The integration module 15 is used to input the acquired real-time multimodal image data of the target transmission line into the constructed fault state recognition model to obtain the second state recognition result, and to integrate the first state recognition result and the second state recognition result to obtain the state recognition result of the target transmission line.

[0091] Preferably, the high-risk module 12 includes: Trend units are used to analyze the trend characteristics of historical data for each dimension. The correlation unit is used to analyze the correlation coefficient of any two trend features to obtain the correlation features between historical data of the dimension; The model unit is used to input trend features and correlation features into the constructed fusion analysis model to obtain a comprehensive risk value; The division unit is used to divide the target transmission line into regions based on the comprehensive risk value and identify high-risk locations.

[0092] Preferably, correlation analysis is performed on multidimensional real-time electrical data to determine the initial fault state, including: The preprocessing unit is used to preprocess multidimensional real-time electrical data to obtain time-aligned observation sequences; The first unit is used to perform numerical analysis on each observation sequence to determine the first anomalous feature; The first unit is used to perform collaborative analysis on any two observation sequences to determine the first anomalous feature; The fault unit is used to determine the initial fault state based on the first abnormal feature and the second abnormal feature.

[0093] Preferably, multidimensional real-time electrical data is analyzed based on trend and correlation characteristics to determine fault development trends, including: The determination unit is used to compare the trend characteristics of multidimensional real-time electrical data with the corresponding dimensional historical data to determine the initial fault development trend; The correction unit is used to analyze the collaborative evolution of multidimensional real-time electrical data based on correlation characteristics, and to correct the initial fault development trend based on the analysis results, thereby obtaining the fault development trend.

[0094] Preferably, the process of constructing the fault state identification model includes: The augmentation unit is used to convert text data in multidimensional historical data into image data and generate augmented dataset samples by using trend features and correlation features as the basis for enhancement. The training unit is used to train the neural network model based on the original dataset and augmented samples in the multidimensional historical data to obtain the fault state recognition model.

[0095] Those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. The program can be stored in a computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. The storage medium can be a magnetic disk, optical disk, read-only memory (ROM), or random access memory (RAM), etc.

[0096] Accordingly, embodiments of the present invention provide a computer-readable storage medium, the computer-readable storage medium including a stored computer program, wherein, when the computer program is executed, it controls the device where the computer-readable storage medium is located to perform steps in the power transmission line state identification method of the above embodiments, for example... Figure 1 Steps S1 to S5 as described above.

[0097] This invention breaks through the traditional passive inspection mode with its uniform cycle and comprehensive coverage. Its core lies in using multi-dimensional historical data as a foundation, quantifying the trend characteristics of each dimension and their interrelationships to generate scientific risk results, and dynamically delineating high-risk locations accordingly for precise resource allocation. Building upon this, it integrates multi-dimensional real-time electrical data from high-risk locations with real-time multimodal image data from the entire domain, using physical model-driven fault evolution analysis and data-driven intelligent identification models respectively to output first and second state identification results, which are then fused for decision-making. This not only solves the problems of low efficiency and long cycles of manual inspections but also significantly improves the timeliness, precision, and intelligence of power grid condition assessment.

[0098] The embodiments described above are merely illustrative of several implementations of the present invention, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of the present invention. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the present invention, and these modifications and improvements all fall within the scope of protection of the present invention. Therefore, the scope of protection of this patent should be determined by the appended claims.

Claims

1. A method for identifying the state of a power transmission line, characterized in that, include: Obtain multi-dimensional historical data of the target transmission line; Analyze the trend characteristics of historical data for each dimension and the correlation characteristics between the historical data for each dimension, process the trend characteristics and the correlation characteristics to obtain the risk results of the target transmission line, and use the risk results to determine high-risk locations; Obtain multidimensional real-time electrical data of the high-risk locations, perform correlation analysis on the multidimensional real-time electrical data, and determine the initial fault state; The multidimensional real-time electrical data is analyzed based on the trend features and the correlation features to determine the fault development trend. Based on the fault development trend and the initial fault state, the first state identification result of the high-risk point is determined. The real-time multimodal image data of the target transmission line is obtained and input into the constructed fault state identification model to obtain the second state identification result. The first state identification result and the second state identification result are then integrated and processed to obtain the state identification result of the target transmission line.

2. The method for identifying the state of a transmission line as described in claim 1, characterized in that, The analysis of the trend characteristics of historical data in each dimension and the correlation characteristics between the historical data in each dimension, processing the trend characteristics and the correlation characteristics to obtain the risk result of the target transmission line, and determining high-risk locations based on the risk result, includes: Analyze the trend characteristics of historical data for each dimension; By analyzing the correlation coefficients of any two trend features, the correlation features between the historical data of the dimension can be obtained; The trend characteristics and correlation characteristics are input into the constructed fusion analysis model to obtain the comprehensive risk value; Based on the comprehensive risk value, the target transmission line is divided into regions to determine the high-risk locations.

3. The method for identifying the status of transmission lines as described in claim 1, characterized in that, The step of performing correlation analysis on the multidimensional real-time electrical data to determine the initial fault state includes: The multidimensional real-time electrical data is preprocessed to obtain a time-aligned observation sequence; Numerical analysis is performed on each of the observation sequences to determine the first anomalous feature; By performing collaborative analysis on any two of the observed sequences, a second anomaly feature can be determined. The initial fault state is determined based on the first abnormal feature and the second abnormal feature.

4. The method for identifying the status of transmission lines as described in claim 1, characterized in that, The analysis of the multidimensional real-time electrical data based on the trend features and the correlation features to determine the fault development trend includes: The trend characteristics of the multidimensional real-time electrical data and the corresponding historical data of the same dimension are compared to determine the initial fault development trend; Based on the correlation features, the collaborative evolution among the multidimensional real-time electrical data is analyzed, and the analysis results are used to correct the initial fault development trend to obtain the fault development trend.

5. The method for identifying the status of transmission lines as described in claim 1, characterized in that, The construction process of the fault state identification model includes: Using the trend features and the correlation features as enhancement criteria, the text data in the multidimensional historical data is converted into image data to generate a dataset to expand the sample. The neural network model is trained based on the original dataset and the expanded samples in the multidimensional historical data to obtain the fault state identification model.

6. A status identification system for transmission lines, characterized in that, include: The acquisition module is used to acquire multidimensional historical data of the target transmission line; The high-risk module is used to analyze the trend characteristics of historical data in each dimension and the correlation characteristics between the historical data in the dimensions, process the trend characteristics and the correlation characteristics to obtain the risk results of the target transmission line, and determine the high-risk locations based on the risk results; The initial module is used to acquire multi-dimensional real-time electrical data of the high-risk locations, perform correlation analysis on the multi-dimensional real-time electrical data, and determine the initial fault state. The first state module is used to analyze the multidimensional real-time electrical data based on the trend features and the correlation features, determine the fault development trend, and determine the first state identification result of the high-risk point based on the fault development trend and the initial fault state. The integration module is used to input the acquired real-time multimodal image data of the target transmission line into the constructed fault state identification model to obtain a second state identification result, and to integrate the first state identification result and the second state identification result to obtain the state identification result of the target transmission line.

7. The transmission line status identification system as described in claim 6, characterized in that, The high-risk module includes: Trend unit, used to analyze the trend characteristics of historical data for each dimension; The correlation unit is used to analyze the correlation coefficient of any two trend features to obtain the correlation features between the historical data of the dimension. The model unit is used to input the trend features and the correlation features into the constructed fusion analysis model to obtain a comprehensive risk value; The division unit is used to divide the target transmission line into regions based on the comprehensive risk value and determine the high-risk locations.

8. The transmission line status identification system as described in claim 6, characterized in that, The step of performing correlation analysis on the multidimensional real-time electrical data to determine the initial fault state includes: A preprocessing unit is used to preprocess the multidimensional real-time electrical data to obtain a time-aligned observation sequence; The first unit is used to perform numerical analysis on each of the observation sequences to determine the first anomalous feature; The first unit is used to perform collaborative analysis on any two of the observation sequences to determine the first anomalous feature; The fault unit is used to determine the initial fault state based on the first abnormal feature and the second abnormal feature.

9. The transmission line status identification system as described in claim 6, characterized in that, The analysis of the multidimensional real-time electrical data based on the trend features and the correlation features to determine the fault development trend includes: The determining unit is used to compare the trend characteristics of the multidimensional real-time electrical data and the corresponding historical data of the dimensions to determine the initial fault development trend; The correction unit is used to analyze the collaborative evolution among the multidimensional real-time electrical data based on the correlation features, and to correct the initial fault development trend based on the analysis results, thereby obtaining the fault development trend.

10. The transmission line status identification system as described in claim 6, characterized in that, The construction process of the fault state identification model includes: An expansion unit is used to convert text data in the multidimensional historical data into image data and generate dataset expansion samples by using the trend features and the correlation features as enhancement criteria. The training unit is used to train the neural network model based on the original dataset and the expanded samples of the multidimensional historical data to obtain the fault state recognition model.

Images