Data-driven power transmission line icing disaster early warning method and system

Abstract

This invention relates to the field of power system operation and maintenance technology, specifically to a data-driven method and system for early warning of icing disasters on transmission lines. The method includes: constructing an icing growth model and a sag deformation model based on historical statistical data; acquiring real-time dual-light monitoring images of the transmission line and meteorological forecast data for a preset future time period; inputting the meteorological forecast data into the icing growth model and calculating the predicted icing value based on the future icing growth trend; inputting the output of the icing growth model into the sag deformation model and calculating the predicted sag value based on the future sag deformation trend; and cross-validating the two predicted values ​​to provide tiered early warning and initiate targeted pre-response measures. This achieves high-accuracy assessment and prediction of icing growth on transmission lines while being completed in real-time on resource-constrained edge computing devices, exhibiting excellent environmental adaptability and practicality, thus effectively realizing preventative operation and maintenance of transmission lines.

Description

Technical Field

[0001] This invention relates to the field of power system operation and maintenance technology, specifically to a data-driven method and system for early warning of icing disasters on transmission lines. Background Technology

[0002] Transmission lines are a crucial component of power systems. However, in harsh environments, icing can cause serious accidents such as wire breakage and even tower collapse. Therefore, icing monitoring technology for transmission lines is essential. Currently, existing technologies for monitoring icing on transmission lines mainly include monitoring methods based on complex physical models, monitoring methods based on deep learning models, vibration analysis methods, and weight sensors installed on the transmission lines. While these technologies can provide some information on icing monitoring of transmission lines, several unresolved technical problems remain. Specifically:

[0003] 1. Monitoring methods based on complex physical models require the establishment of precise mechanical and thermal equations, which involve complex solutions. Monitoring accuracy heavily relies on the precision of input parameters (such as stress and strain, conductor characteristics, ice density, and ultrasonic emissivity), but these parameters are difficult to obtain accurately in real-world environments. Complex physical models are computationally burdensome, making real-time operation difficult on resource-constrained edge computing environments. Furthermore, parameter calibration is challenging, hindering large-scale deployment. Fixed-parameter complex physical models struggle to adapt to varying terrains, climates, and line aging, and lack predictive capabilities. Consequently, their disaster prevention strategies are essentially "response-based," only initiating de-icing after icing reaches a risk threshold, failing to achieve "preventative" resource scheduling and intervention.

[0004] 2. Monitoring methods based on deep learning models output an abstract comprehensive risk score through complex neural networks (CNN, LSTM) and fusion algorithms (attention mechanisms, DS evidence theory). However, this score cannot directly correspond to specific physical states, has poor interpretability, and maintenance personnel cannot directly and accurately understand the on-site situation based on the score, so decision-making still relies on experience. Moreover, deep learning models are also computationally intensive, making it difficult to run in real time on resource-constrained edge devices (such as monitoring devices installed on towers).

[0005] 3. Vibration analysis relies on the fact that icing alters the natural frequency of transmission lines, thus affecting their vibration modes. However, this method requires a precise model to distinguish vibrations caused by icing from those caused by other factors (such as wind); otherwise, misjudgments are highly likely. Weight sensors installed on transmission lines directly measure the weight change due to increased icing, but these sensors are prone to damage from prolonged exposure to harsh environments. Furthermore, both vibration analysis and weight sensor monitoring methods typically provide only localized information, failing to comprehensively cover long-distance transmission lines. Their integration is also low, hindering remote automated monitoring and risk warning for transmission line operation and maintenance. Summary of the Invention

[0006] To address the aforementioned technical issues, this invention proposes a data-driven method and system for early warning of icing disasters on power transmission lines. This system aims to achieve highly accurate assessment and prediction of icing growth on power transmission lines, while simultaneously enabling real-time performance on resource-constrained edge computing devices. It possesses excellent environmental adaptability and practicality, thereby effectively realizing preventative maintenance of power transmission lines.

[0007] Firstly, this application provides a data-driven method for early warning of icing disasters on transmission lines, including the following steps:

[0008] Based on historical statistical data, an ice growth model is constructed that reflects the mapping relationship between meteorological data and the increase in ice thickness.

[0009] Construct a sag deformation model that reflects the mapping relationship between the increase in icing thickness and the sag deformation of the transmission line;

[0010] Real-time acquisition of dual-light monitoring images of transmission lines and meteorological forecast data for a preset time period;

[0011] Meteorological forecast data is input into the icing growth model, and based on real-time acquired dual-light monitoring images and the output of the icing growth model, the icing growth trend within a preset time period is calculated to obtain the icing prediction value.

[0012] The output of the ice growth model is input into the sag deformation model, and based on the real-time acquired dual-light monitoring images and the output of the sag deformation model, the trend of sag deformation in the future preset time period is calculated to obtain the sag prediction value.

[0013] Cross-validate the predicted values ​​of icing and sag, and issue graded early warnings and initiate targeted pre-response measures based on the cross-validation results.

[0014] In some embodiments, an ice growth model reflecting the mapping relationship between meteorological data and ice thickness increments is constructed based on historical statistical data, including:

[0015] Obtain a reference image of the transmission line and extract the reference width of the transmission line from the reference image;

[0016] Acquire historical dual-light monitoring images of transmission lines and historical meteorological data for the corresponding areas;

[0017] Based on historical dual-light monitoring images, the historical visual width of several transmission line segments corresponding to each monitoring node was calculated.

[0018] The historical icing thickness of the transmission line is obtained by directly calculating the difference between the historical visual width and the baseline width.

[0019] The difference between historical icing thicknesses in adjacent time periods is calculated to obtain the historical icing thickness increment of the transmission line;

[0020] Based on historical ice thickness increments and corresponding regional historical meteorological data, a lookup table-based ice growth model is constructed to reflect the mapping relationship between meteorological data and ice thickness increments.

[0021] In some embodiments, a lookup-based icing growth model reflecting the mapping relationship between meteorological data and icing thickness increments is constructed based on historical icing thickness increments and corresponding regional historical meteorological data, including:

[0022] Historical meteorological statistical features and topographic features of each monitoring node are extracted from the historical meteorological data of the region and fused to form a comprehensive feature vector of each monitoring node;

[0023] Clustering algorithms were used to analyze the comprehensive feature vectors of all monitoring nodes, resulting in several cluster groups.

[0024] Based on the historical ice thickness increments and historical meteorological statistical characteristics of all monitoring nodes in each cluster, a shared lookup table-based ice growth model is constructed for each cluster.

[0025] In some embodiments, historical meteorological statistical features include average temperature, average humidity, and average wind speed over each time period. Based on the historical icing thickness increments and historical meteorological statistical features of all monitoring nodes in each cluster, a shared lookup-based icing growth model is constructed for each cluster, including:

[0026] The average wind speed is used to group the data into intervals. Within each wind speed interval group, the average temperature and average humidity are used as the joint query keys, and the historical ice thickness increment obtained under the corresponding meteorological statistical characteristics is used as the query result. A multidimensional query table reflecting the mapping relationship between wind speed-temperature and humidity data and ice thickness increment is established, which is denoted as the ice growth model.

[0027] In some embodiments, a sag deformation model reflecting the mapping relationship between the icing thickness increment and the sag deformation of the transmission line arc is constructed, including:

[0028] Acquire a reference image of the transmission line and extract the reference horizontal position of the transmission line from the reference image;

[0029] Acquire historical dual-light monitoring images of transmission lines;

[0030] Based on historical dual-light monitoring images, the historical horizontal positions of several transmission line segments corresponding to each monitoring node were calculated.

[0031] The historical sag amplitude of the transmission line is obtained by directly calculating the difference between the historical horizontal position and the benchmark horizontal position;

[0032] The historical sag deformation of the transmission line is obtained by calculating the difference between the historical sag amplitudes of adjacent time periods.

[0033] Based on historical sag variables and corresponding historical icing thickness increments over time periods, a lookup table-based sag deformation model is constructed to reflect the mapping relationship between icing thickness increments and transmission line sag variables.

[0034] In some embodiments, a lookup-based sag deformation model reflecting the mapping relationship between the icing thickness increment and the transmission line sag variable is constructed based on historical sag deformation variables and historical icing thickness increments for corresponding time periods, including:

[0035] Based on historical sag variables and corresponding historical icing thickness increments over the same time period, time-synchronized data pairs (icing thickness increment, sag variables) are obtained.

[0036] The data pairs (ice thickness increment, sag variable) are grouped into intervals according to the ice thickness increment. Within each ice thickness increment interval group, the statistical value of the sag variable in each data pair is calculated to obtain the typical sag variable value.

[0037] Using the ice thickness increment range as the query key and the typical sag deformation value as the query result, a two-dimensional query table reflecting the mapping relationship between the ice thickness increment and the sag deformation of the transmission line is established, denoted as the sag deformation model.

[0038] In some embodiments, meteorological forecast data is input into an icing growth model, and based on real-time acquired dual-light monitoring images and the output of the icing growth model, the icing growth trend within a preset time period is calculated to obtain icing prediction values, including:

[0039] Meteorological statistical characteristics of monitoring nodes for each unit time period within a preset future time period are extracted from meteorological forecast data;

[0040] By inputting meteorological statistical characteristics into the icing growth model, the output is the increase in icing thickness of the monitoring node in each unit time period within a preset future time period.

[0041] Based on real-time acquired dual-light monitoring images, the real-time visual width of several transmission line segments corresponding to the monitoring nodes is calculated.

[0042] The real-time icing thickness of the transmission line is obtained by directly calculating the difference between the real-time visual width and the reference width.

[0043] The real-time icing thickness of each transmission line segment corresponding to the monitoring node is accumulated with the icing thickness increment for each unit time period within the future preset time period to obtain the predicted icing value for each transmission line segment corresponding to the monitoring node for each unit time period within the future preset time period.

[0044] In some embodiments, the output of the icing growth model is input into the sag deformation model, and based on the real-time acquired dual-light monitoring images and the output of the sag deformation model, the trend of sag deformation within a preset time period is estimated to obtain the sag prediction value, including:

[0045] The output of the icing growth model is input into the sag deformation model, and the output is the typical sag deformation value of the monitoring node in each unit time period within a preset time period in the future.

[0046] Based on real-time acquired dual-light monitoring images, the real-time horizontal position of several transmission line segments corresponding to the monitoring nodes is calculated.

[0047] The real-time sag amplitude of the transmission line is obtained by directly calculating the difference between the real-time horizontal position and the reference horizontal position.

[0048] The real-time sag amplitude of each transmission line segment corresponding to the monitoring node is accumulated with the typical sag deformation value of each unit time period within the future preset time period to obtain the predicted sag value of each transmission line segment corresponding to the monitoring node within each unit time period within the future preset time period.

[0049] In some embodiments, cross-validation is performed on the icing prediction value and the sag prediction value, and based on the cross-validation results, a graded early warning is issued and targeted pre-response measures are initiated, including:

[0050] With ice thickness as the horizontal axis and sag amplitude as the vertical axis, a cross-risk matrix is ​​defined. Each cell in the cross-risk matrix corresponds to a comprehensive early warning level and a targeted pre-treatment measure.

[0051] The calculated icing and sag prediction values ​​are used as coordinate points to locate the cross-risk matrix, and the movement trajectory of the coordinate points in the cross-risk matrix within a preset time period is obtained to determine the evolution of the future warning level and the targeted pre-response measures to be initiated.

[0052] Secondly, this application provides a data-driven early warning system for icing disasters on transmission lines, including:

[0053] The model building module is used to build an ice growth model that reflects the mapping relationship between meteorological data and ice thickness increment based on historical statistical data, and to build an arc sag deformation model that reflects the mapping relationship between ice thickness increment and transmission line arc sag variable.

[0054] The data acquisition module is used to acquire real-time dual-light monitoring images of the transmission line and meteorological forecast data for a preset time period.

[0055] The icing prediction module is used to input meteorological forecast data into the icing growth model, and based on the real-time acquired dual-light monitoring images and the output of the icing growth model, to estimate the icing growth trend within a preset time period and obtain the icing prediction value.

[0056] The sag prediction module is used to input the output of the icing growth model into the sag deformation model, and based on the real-time acquired dual-light monitoring images and the output of the sag deformation model, to estimate the trend of sag deformation within a preset time period and obtain the sag prediction value.

[0057] The disaster early warning module is used to cross-validate the icing prediction value and the sag prediction value, and to issue graded early warnings and initiate targeted pre-response measures based on the cross-validation results.

[0058] The beneficial technical effects of the present invention include at least the following:

[0059] 1. This paper presents a data-driven early warning method and system for transmission line icing disasters. Through a collaborative technical closed loop of "reliable perception → data modeling → iterative prediction → precise intervention," it creatively provides a comprehensive solution for transmission line icing disaster prevention that is more adaptable to complex real-world environments, possesses forward-looking early warning capabilities, and supports preventative precision operations. Specifically, by coordinating dual-light monitoring images and meteorological data to ensure continuous perception, it addresses the reliability issues of existing technologies for monitoring long-distance transmission lines under severe weather conditions. The construction of icing growth models and sag deformation models avoids the practicality problems caused by extreme dependence on complex physical parameters and pure signals, enabling more stable prediction results that more closely reflect the historical performance of transmission lines even in actual noise environments and under uncertain model parameters. Addressing the lack of forward-looking early warning capabilities in existing technologies, this application, through collaborative perception, data-driven model construction, and weather forecasting, is the first to integrate a quantitative prediction function for future icing trends at the solution level, enabling early deployment of disaster prevention work and achieving a leap from "passive response" to "proactive prediction."

[0060] 2. By creating an icing monitoring and prediction system that does not rely on complex theoretical physics formulas and precise conductor parameters, the complex mechanical and acoustic measurement problems are transformed into more intuitive and stable image recognition and measurement problems, thus completely solving the parameter acquisition problem. Furthermore, a lookup model based on historical statistical data is used to characterize the complex physicochemical process of icing growth, replacing theory-driven calculations with data-driven approaches. By leveraging the statistical regularity between meteorological conditions and icing thickness increments, the accuracy, lightweight nature, and generalization ability of the icing growth model are effectively improved. Since the lookup operation is a low-cost memory access, high-frequency, low-latency icing growth assessment and prediction can run in real-time on resource-constrained edge computing devices. Simultaneously, by combining the meteorological-icing relationship of each specific micro-region, the system can fully consider the local climate effects caused by topography (wind gaps, valleys), altitude, and the impact of line aging (periodic updates to the reference width), thereby achieving strong environmental adaptability, high robustness, high practicality, and forward-looking prediction capabilities in harsh engineering environments.

[0061] 3. Adopting a data-driven, practical approach, it abandons explicit modeling of the internal mechanisms of physical processes, instead directly learning and utilizing the external input-output correlation patterns exhibited by the process on a specific object (this transmission line). In sag assessment, it does not concern itself with "how icing causes sag changes through mechanical principles," but only with "how much the mechanical state (sag) of the transmission line typically changes when icing thickness increases by a certain amount on this transmission line." Its output naturally includes all the real and complex comprehensive characteristics of this specific line segment (such as conductor creep, span differences, tower foundation settlement, and other factors that are difficult to quantify precisely in theoretical models), thereby achieving rapid, stable, and engineering-practical risk assessment of transmission lines.

[0062] 4. A creative dual-dimensional perception and joint decision-making mechanism for icing load (ice thickness) and mechanical response (sag) is introduced. In this concept, the sag deformation model acts as the quantifier of mechanical response. Its collaborative logic is as follows: the icing growth model predicts / monitors the input (incremental icing thickness), while the sag deformation model evaluates the system state response (sag deformation value) in real time. The early warning system then fuses, compares, and cross-validates these two types of information to trigger the early warning level with the highest degree of matching with the actual risk and targeted pre-treatment measures, forming a synergy of "prediction-guided precise intervention." This avoids the practicality problems of uncertainty in existing complex models and the lack of forward-looking early warning capabilities, greatly improving de-icing efficiency and the safety of transmission line operation and maintenance.

[0063] Other features and advantages of the present invention will be disclosed in detail in the following detailed description and accompanying drawings. Attached Figure Description

[0064] The invention will be further described below with reference to the accompanying drawings:

[0065] Figure 1 This is a flowchart of a data-driven early warning method for icing disasters on power transmission lines, according to an embodiment of the present invention.

[0066] Figure 2 This is a schematic diagram of the structure of a data-driven transmission line icing disaster early warning system according to an embodiment of the present invention. Detailed Implementation

[0067] The technical solutions of the embodiments of the present invention will be explained and described below with reference to the accompanying drawings. However, the following embodiments are only preferred embodiments of the present invention and not all of them. Other embodiments obtained by those skilled in the art based on the embodiments in the implementation methods without creative effort are all within the protection scope of the present invention.

[0068] In the following description, terms such as “inner,” “outer,” “upper,” “lower,” “left,” and “right” are used only to facilitate the description of embodiments and simplify the description, and are not intended to indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation, and therefore should not be construed as a limitation of the present invention.

[0069] Example 1:

[0070] Please see the appendix Figure 1 , Figure 1 A flowchart illustrating a data-driven early warning method for icing disasters on transmission lines, as provided in one embodiment of this specification, is shown.

[0071] like Figure 1 As shown, this data-driven method for early warning of icing disasters on transmission lines may include at least the following steps:

[0072] S1, based on historical statistical data, constructs an ice growth model that reflects the mapping relationship between meteorological data and the increase in ice thickness.

[0073] It is understood that this embodiment does not involve complex physical calculations or deep learning, but instead constructs a data-driven lookup table-based icing growth model. That is, the icing growth model does not rely on precise physical parameters and theoretical equations, but summarizes the mapping relationship between input (meteorological data) and output (increase in icing thickness) through a large amount of historical data of the transmission line itself. This model is more in line with the actual situation of the transmission line section, avoids theoretical model deviations, is not sensitive to sensor instantaneous noise and model mismatch, and exhibits stronger stability and reliability in real complex environments.

[0074] Specifically, in this embodiment, an ice growth model reflecting the mapping relationship between meteorological data and ice thickness increments is constructed based on historical statistical data, including:

[0075] S11, acquire the reference image of the transmission line, and extract the reference width of the transmission line from the reference image of the transmission line.

[0076] Among them, the reference image of the transmission line can be the image of the transmission line captured by the dual-light monitoring equipment (visible light camera and infrared thermal imager) deployed by the system when the transmission line is initially erected on the base tower. The reference width of the transmission line extracted from it is the initial visual diameter (i.e. visual width) of the transmission line.

[0077] Furthermore, the transmission line reference image can also be a transmission line image captured by dual-light monitoring equipment during each ice-free period, with the reference width periodically updated to adapt to the actual situation of wire aging.

[0078] S12, acquire historical dual-light monitoring images of the transmission line and historical meteorological data of the corresponding area;

[0079] In this embodiment, the historical meteorological data for the corresponding region does not refer to data from a single meteorological station covering hundreds of kilometers, but rather to meteorological data distributed along the transmission line, comprising a series of monitoring nodes and representing the "micro-regions" where these nodes are located. Monitoring nodes are key points selectively deployed along the transmission line (e.g., on several towers at intervals), forming a monitoring network distributed along the line. Each monitoring node is responsible for monitoring the status of its own span (i.e., a section of transmission line between two towers) and the transmission lines in adjacent spans. It can be understood that, in terms of monitoring range: historical meteorological data area > monitoring nodes > section of transmission line between two towers.

[0080] S13. Based on historical dual-light monitoring images, the historical visual width of several transmission line segments corresponding to each monitoring node is calculated.

[0081] The historical dual-light monitoring images were captured at fixed time intervals by the system-deployed dual-light monitoring equipment (visible light camera and infrared thermal imager) when the transmission lines were initially erected on the base towers, covering the transmission line segments within their field of view. For each historical dual-light monitoring image, image processing algorithms (such as edge detection and segmentation algorithms) were used to automatically identify and measure the visual width of the transmission line in the image. This visual width is the total width of the actual physical diameter of the transmission line and the projection of the surface ice layer onto the camera's imaging plane.

[0082] S14 directly calculates the difference between the historical visual width and the reference width to obtain the historical icing thickness of the transmission line.

[0083] Specifically, in this embodiment, the historical visual width is directly subtracted from the baseline width measured at the same monitoring location and on the same transmission line segment in an ice-free state (baseline period), decoupling it from physical parameters. The physical meaning of this difference is that the total increase in the visual diameter of the transmission line caused by the growth of ice around the conductor is assumed to be the equivalent ice thickness, thereby avoiding the measurement of complex characteristics such as ice density and adhesion.

[0084] S15, calculate the difference between the historical icing thickness of adjacent time periods to obtain the historical icing thickness increment of the transmission line.

[0085] Specifically, in this embodiment, for the same monitoring location, the difference in historical icing thickness between adjacent time periods is calculated according to a time series (fixed time period, such as every hour), and the resulting historical icing thickness increment can have positive or negative values.

[0086] Understandably, step S15 of this embodiment shifts the focus from "state quantity" (total thickness) to "process quantity" (growth rate). Considering that the risk of icing disasters lies not only in the final thickness but also in the growth rate, and that incremental data directly reflects the dynamic accumulation process of icing under specific meteorological conditions, this embodiment uses it as the most direct dependent variable for constructing the prediction model.

[0087] S16. Based on historical ice thickness increments and corresponding regional historical meteorological data, a lookup-based ice growth model is constructed to reflect the mapping relationship between meteorological data and ice thickness increments.

[0088] Understandably, the core concept of this embodiment lies in creating an icing monitoring and prediction system that does not rely on complex theoretical physics formulas and precise conductor parameters through the creative combination of direct image measurement and lookup table empirical models. This transforms complex mechanical and acoustic measurement problems into more intuitive and stable image recognition and measurement problems, thereby completely solving the parameter acquisition problem. Furthermore, a lookup table model based on historical statistical data is used to characterize the complex physicochemical process of icing growth, replacing theory-driven calculations with data-driven approaches. By leveraging the statistical regularity between meteorological conditions and icing thickness increments, the accuracy, lightweight nature, and generalization ability of the icing growth model are effectively improved. Since the lookup table operation is a low-computation memory access, high-frequency, low-latency icing growth assessment and prediction can run in real time on resource-constrained edge computing devices. Simultaneously, by incorporating the meteorological-icing relationship of each specific micro-region, the system can fully consider the local climate effects caused by topography (wind gaps, valleys), altitude, and the impact of line aging (periodic updates to the reference width), thus achieving strong environmental adaptability, high robustness, high practicality, and forward-looking prediction capabilities in harsh engineering environments.

[0089] Specifically, in this embodiment, based on historical ice thickness increments and corresponding historical meteorological data for the region, a lookup-based ice growth model reflecting the mapping relationship between meteorological data and ice thickness increments is constructed, including:

[0090] S161 extracts historical meteorological statistical features and topographic features of each monitoring node from the historical meteorological data of the region, and integrates them to form a comprehensive feature vector of each monitoring node.

[0091] In this embodiment, historical meteorological statistical features include average temperature, average humidity, and average wind speed for each time period. Topographic features may include the elevation of the monitoring node, its slope aspect (windward / leeward), and distance from large bodies of water or valleys (which can be obtained from a GIS system), but this embodiment does not limit these features.

[0092] S162 uses a clustering algorithm to analyze the comprehensive feature vectors of all monitoring nodes, resulting in several cluster groups.

[0093] For example, clustering algorithms such as K-means and hierarchical clustering are used to analyze the comprehensive feature vectors of all monitoring nodes, and monitoring nodes with similar comprehensive feature vectors are automatically grouped into several "cluster groups". For example, all monitoring nodes with "high altitude, windward slope, low temperature and high humidity" may be clustered into the same group.

[0094] S163. Based on the historical ice thickness increment and historical meteorological statistical characteristics of all monitoring nodes in each cluster, a shared lookup table-based ice growth model is constructed for each cluster.

[0095] For each cluster, historical data samples (i.e., historical ice thickness increments and historical meteorological statistical characteristics) from all monitoring nodes within the cluster are merged to form a larger historical statistical database representing the typical environment of that cluster. Using this historical statistical database, a shared lookup-based ice growth model is constructed for the cluster, which summarizes the common patterns of the cluster.

[0096] It is understandable that maintaining a completely independent lookup-based icing growth model for hundreds or thousands of monitoring nodes would present significant complexity and cost challenges in engineering deployment and maintenance. To address this, this embodiment proposes a hierarchical modeling solution based on meteorological statistics and topographic features clustering to strike a balance between personalized adaptation of icing growth models and centralized management. This significantly reduces the number of independent icing growth models that need to be built and maintained while maintaining strong adaptability to meteorological and topographic factors in various regions, thus achieving centralized management. This embodiment does not build a unique model from scratch for each monitoring node; instead, it clusters monitoring nodes with similar meteorological statistics and topographic features, and builds a shared icing growth model for each cluster group.

[0097] Furthermore, in this embodiment, based on the historical icing thickness increments and historical meteorological statistical characteristics of all monitoring nodes in each cluster group, a shared lookup-based icing growth model is constructed for each cluster group, including:

[0098] The average wind speed is used to group the data into intervals. Within each wind speed interval group, the average temperature and average humidity are used as the joint query keys, and the historical ice thickness increment obtained under the corresponding meteorological statistical characteristics is used as the query result. A multidimensional query table reflecting the mapping relationship between wind speed-temperature and humidity data and ice thickness increment is established, which is denoted as the ice growth model.

[0099] It is understood that this embodiment is for an ice growth model constructed for a single cluster group. The construction method for ice growth models of other cluster groups can refer to this embodiment, and will not be repeated here. Specifically, the implementation method of this embodiment is as follows:

[0100] 1. Data Accumulation: During long-term operation, the system uses the historical ice thickness increment ΔH calculated in the aforementioned steps as the key output label, and simultaneously records the meteorological conditions corresponding to the generation of this historical ice thickness increment: the average temperature, average humidity, and average wind speed within the same unit time period. This accumulates a massive amount of historical data samples.

[0101] 2. Data Organization and Table Construction: To improve the granularity and query efficiency of the icing growth model, considering the significant impact of wind speed on icing morphology and growth rate, all historical data samples were first grouped according to the intervals of average wind speed. Then, within each wind speed interval group, a multi-dimensional query table was created, using average temperature and average humidity (often rounded to simplify queries) as the joint query key, and the typical icing thickness increment obtained under those meteorological conditions as the query result. This table covers the common low-temperature and high-humidity intervals in the region.

[0102] For example, the ice growth model covers the average temperature ∈ (-30℃, 0℃), average humidity All values ​​of the ice thickness increment ΔH within the interval (0%RH, 95%RH) are shown in Table 1 below:

[0103] Table 1. Example Table of Ice Growth Models .

[0104] 3. Continuous Learning: The icing growth model is not static. With each winter's operation, the monitoring nodes can continuously learn new techniques. , , The △H} data pairs update their lookup tables, enabling the ice growth model to track slow environmental changes (such as surrounding vegetation growth) or minor equipment drift.

[0105] It is understandable that the data in the lookup-based icing growth model constructed in this embodiment comes entirely from the historical operation of transmission lines within the same cluster group, naturally encompassing the combined influence of local micro-meteorological conditions, topography, and the characteristics of the transmission lines themselves, thus exhibiting optimal environmental adaptability. More importantly, this icing growth model is not only an assessment tool explaining current icing growth, but also... , , With ΔH as the input and ΔH as the output, the system is designed to receive future weather forecasts and perform quantitative extrapolation predictions. This enables it to seamlessly connect with future weather forecast sequences for iterative prediction. This collaboration transforms the system from a real-time status assessment capability to a future trend quantitative prediction capability, providing key decision support for preventative operation and maintenance.

[0106] S2. Construct a sag deformation model that reflects the mapping relationship between the increase in icing thickness and the sag deformation of the transmission line arc.

[0107] Existing methods (such as vibration analysis) indirectly infer icing through frequency changes, but sag changes are influenced by multiple factors such as wind and temperature, making it difficult to directly and accurately separate sag changes caused purely by icing from vibration signals, resulting in a high false alarm rate. Weight sensors, while direct, are easily damaged. Existing mechanical model calculation methods, based on the principles of materials mechanics and structural mechanics, first invert the weight of the icing, then combine it with wind speed to construct an ice-wind coupled force model, and finally calculate the actual tension and deformation based on the mechanical properties of the conductor. However, this existing technology inevitably leads to the following technical problems:

[0108] 1. Calculation is complex and real-time performance is challenging: The mechanical equations or simulation calculations involving multiple coupled variables are large, making it difficult to achieve real-time or high-frequency evaluation on edge computing devices, which affects the alarm response speed.

[0109] 2. Reliance on precise parameters and insufficient robustness: The accuracy of the ice-wind coupled stress model heavily depends on the accuracy of the input parameters, such as the elastic modulus of the conductor, the coefficient of linear expansion, the actual span length, and the tower base stiffness. These parameters may change unpredictably with line aging and environmental changes, leading to deviations in the model output.

[0110] 3. Limited practicality: For maintenance personnel, understanding and calibrating complex physical models has a high barrier to entry. Furthermore, each line, and even different sections of the same line, require a set of parameters for calibration, resulting in a massive workload for implementation and maintenance.

[0111] Therefore, this embodiment adopts a data-driven, practical approach, abandoning explicit modeling of the internal mechanisms of the physical process. Instead, it directly learns and utilizes the external input-output correlation patterns exhibited by the process on a specific object (this transmission line). In sag assessment, it does not concern itself with "how icing causes sag changes through mechanical principles," but only with "how much the mechanical state (sag) of the transmission line typically changes when icing thickness increases by a certain amount on this transmission line." Its output naturally includes all the real and complex comprehensive characteristics of this specific line segment (such as conductor creep, span differences, tower foundation settlement, and other factors that are difficult to quantify precisely in theoretical models), in order to achieve a fast, stable, and engineering-practical risk assessment of the transmission line.

[0112] Specifically, in this embodiment, a sag deformation model reflecting the mapping relationship between the increase in icing thickness and the sag deformation of the transmission line arc is constructed, including:

[0113] S21, acquire the reference image of the transmission line, and extract the reference horizontal position of the transmission line from the reference image.

[0114] Specifically, the reference horizontal position of the transmission line in the image coordinate system (i.e., the vertical position distribution curve of the conductor relative to the line connecting the two suspension points in the image) is extracted from the reference image of the transmission line using image recognition technology (such as conductor extraction and curve fitting), thereby defining the reference state of the initial sag.

[0115] S22, acquire historical dual-light monitoring images of the transmission line;

[0116] S23, based on historical dual-light monitoring images, calculates the historical horizontal position of several transmission line segments corresponding to each monitoring node;

[0117] S24 directly calculates the difference between the historical horizontal position and the benchmark horizontal position to obtain the historical sag amplitude of the transmission line.

[0118] Understandably, this embodiment uses a visual difference method to invert the absolute sag amplitude. The historical horizontal position curve is compared with the baseline horizontal position curve at corresponding pixels, and the difference is directly calculated. This difference represents the amount of sag of the transmission line relative to the baseline horizontal position, i.e., the sag amplitude. Physically, due to the increased icing load, the conductor is stretched and sags, which is directly reflected as a vertical downward shift of pixels in the horizontal position of the conductor in the image, equivalent to the actual sag amplitude.

[0119] S25, calculate the difference between the historical sag amplitudes of adjacent time periods to obtain the historical sag deformation of the transmission line.

[0120] Understandably, this embodiment designs a visual sag monitoring data pipeline that is of the same origin and symmetrical as the icing monitoring system. For the same monitoring location, the difference in sag amplitude between adjacent time periods is calculated according to the time series to obtain the historical sag deformation variable, which also has positive and negative values.

[0121] S26. Based on historical sag variables and historical icing thickness increments for corresponding time periods, a lookup table-based sag deformation model is constructed to reflect the mapping relationship between icing thickness increments and transmission line sag variables.

[0122] Furthermore, in this embodiment, based on historical sag deformation variables and historical icing thickness increments for corresponding time periods, a lookup-based sag deformation model reflecting the mapping relationship between icing thickness increments and transmission line sag deformation variables is constructed, including:

[0123] S261, based on historical sag variables and historical icing thickness increments for corresponding time periods, time-synchronized (icing thickness increment, sag variables) data pairs are obtained.

[0124] Understandably, strict time synchronization means that the icing thickness increment and sag variable are measured under the same external environmental conditions (temperature, wind speed) and within the same time span. This allows the sag variable to be attributed to the icing thickness increment to the greatest extent possible when analyzing this data pair, because other background factors change relatively little or are consistent within the same short time window, thus helping to "cleanly" extract the relationship between the two core variables (icing thickness increment and sag variable).

[0125] S262, group the (ice thickness increment, sag variable) data pairs into intervals based on the ice thickness increment, calculate the statistical value (such as the average value) of the sag variable in each data pair within each ice thickness increment interval group, and obtain the typical sag variable value.

[0126] Understandably, the core objective of this embodiment is to identify the risk level (e.g., low, medium, high, emergency) in a timely and reliable manner, rather than to precisely calculate whether the sag is 1.53 meters or 1.55 meters. Therefore, the mapping from the increase in ice thickness to the typical sag value obtained through big data statistics, although potentially not optimal for a single sample, can still reliably determine whether the current increase in ice thickness typically triggers a "dangerous" level of sag growth, which is sufficient for early warning.

[0127] S263. Using the ice thickness increment range as the query key and the typical sag deformation value as the query result, a two-dimensional query table reflecting the mapping relationship between the ice thickness increment and the sag deformation of the transmission line is established, denoted as the sag deformation model.

[0128] It is understandable that the purpose of the sag deformation model in this embodiment is to establish a direct and simplified mapping between the incremental icing load and the mechanical deformation response of the conductor, rather than repeating or simulating the complex process of "how meteorological conditions affect icing growth," which has already been addressed by the upstream model. That is, sag deformation is essentially a mechanical response problem; the incremental icing thickness ΔH directly determines the additional load, while the sag deformation ΔL is the deformation result caused by the load. While meteorological data such as temperature, humidity, and wind speed affect the formation and growth of icing, once icing forms, the resulting mechanical deformation mainly depends on the weight and distribution of the icing, which is already reflected in ΔH. The influence of topographic data such as span length and conductor parameters is implicitly contained in the mapping relationship formed by historical data. Because each cluster group uses its own historical data to build its model, this data naturally includes all the combined influences of the topography and line parameters of the node. Therefore, it is not necessary to explicitly introduce these complex parameters, which is precisely the advantage of the data-driven sag deformation model in this embodiment.

[0129] S3 acquires real-time dual-light monitoring images of transmission lines and weather forecast data for a preset time period.

[0130] Among them, the system deploys "dual-light monitoring equipment" (visible light camera and infrared thermal imager) to continuously monitor the transmission lines. The visible light camera captures the surface of the transmission lines, which is easily obstructed by heavy fog and blizzards. Therefore, the infrared thermal imaging's ability to penetrate fog and snow enables accurate identification of the visual characteristics of the transmission lines under harsh conditions such as high and low temperatures, aging, salt spray, fog, and freezing rain.

[0131] For example, this embodiment obtains weather forecast data for the next 72 hours (3 days). The next 72 hours are regarded as 72 consecutive unit time periods. The weather forecast data includes the forecast average temperature, forecast average humidity and forecast average wind speed for each unit time period (e.g., hourly).

[0132] S4 inputs meteorological forecast data into the icing growth model, and based on the real-time acquired dual-light monitoring images and the output of the icing growth model, calculates the icing growth trend within a preset time period to obtain the icing prediction value.

[0133] Understandably, the core functions of existing technologies are typically real-time monitoring and current risk assessment. That is, existing systems calculate the icing growth rate based on current data, but this rate only describes the current situation and doesn't address how to quantitatively predict icing trends over a future period. This results in a disaster prevention strategy that is essentially "responsive," only initiating de-icing after the icing reaches a risk threshold, failing to achieve "preventative" resource allocation and intervention. This embodiment, however, explicitly proposes inputting future weather forecast data (temperature, humidity, and wind speed sequences) into its established lookup-based icing growth model for iterative forward calculations, quantitatively predicting future icing growth trends. This elevates the system's capabilities from "current situation awareness" to "proactive prediction," providing crucial decision-making support for preventative de-icing and optimized de-icing resource allocation.

[0134] Specifically, in this embodiment, meteorological forecast data is input into the icing growth model, and based on real-time acquired dual-light monitoring images and the output of the icing growth model, the icing growth trend within a preset time period is calculated to obtain icing prediction values, including:

[0135] S41, extract the meteorological statistical characteristics of the monitoring node for each unit time period within a preset future time period from the meteorological forecast data;

[0136] S42 inputs meteorological statistical features into the icing growth model of the cluster group corresponding to the monitoring node, and outputs the icing thickness increment of the monitoring node in each unit time period within a preset time period in the future.

[0137] The increase in ice thickness is a "growth" rather than a "total amount".

[0138] S43, based on the real-time acquired dual-light monitoring images, calculates the real-time visual width of several transmission line segments corresponding to the monitoring node;

[0139] S44 directly calculates the difference between the real-time visual width and the reference width to obtain the real-time icing thickness of the transmission line;

[0140] S45, the real-time icing thickness of each transmission line segment corresponding to the monitoring node is accumulated with the icing thickness increment of each unit time cycle within the future preset time period to obtain the icing prediction value of each transmission line segment corresponding to the monitoring node within each unit time cycle within the future preset time period.

[0141] It is understandable that the prediction of icing growth trends in this embodiment does not involve complex algorithms. Essentially, it invokes the process controller of a lookup-based icing growth model, dividing the meteorological forecast data for a preset future time period into unit time segments. Then, for each future period, using its meteorological characteristics as input, it queries the icing growth model to obtain the icing thickness increment for that period. Next, starting with the real-time icing thickness measured by real-time dual-light imagery, it accumulates the icing thickness increments for all future periods in chronological order, thereby generating an icing thickness prediction sequence from the current moment to a future moment. This sequence can quantitatively predict the cumulative icing thickness at any point within a preset future time period, achieving dynamic simulation of the future icing accumulation process.

[0142] S5. Input the output of the icing growth model into the sag deformation model, and based on the real-time acquired dual-light monitoring images and the output of the sag deformation model, calculate the trend of sag deformation within a preset time period to obtain the sag prediction value.

[0143] The output of the icing growth model is the future icing thickness increment sequence. This is used as the driving input to query the sag deformation model in sequence, thereby simultaneously deducing the typical sag deformation sequence for the corresponding time in the future (i.e., the amount of sag change that the conductor will usually produce under this icing increment).

[0144] Understandably, this embodiment elevates the early warning judgment from a "scalar comparison" problem to a "vector localization" problem in a two-dimensional state space of "ice thickness-sag". The key role of the sag deformation model is to provide a second state dimension, independent of ice thickness, that is indispensable for constituting this decision space.

[0145] Specifically, in this embodiment, the output of the icing growth model is input into the sag deformation model, and based on the real-time acquired dual-light monitoring images and the output of the sag deformation model, the trend of sag deformation within a preset time period is calculated to obtain the predicted sag value, including:

[0146] S51, input the output of the ice growth model into the sag deformation model, and output the typical sag deformation value of the monitoring node in each unit time period within the future preset time period;

[0147] S52, based on the real-time acquired dual-light monitoring images, calculates the real-time horizontal position of several transmission line segments corresponding to the monitoring nodes;

[0148] S53, directly calculate the difference between the real-time horizontal position and the reference horizontal position to obtain the real-time sag amplitude of the transmission line;

[0149] S54, the real-time sag amplitude of each transmission line segment corresponding to the monitoring node is accumulated with the typical sag deformation value of each unit time period within the future preset time period to obtain the sag prediction value of each transmission line segment corresponding to the monitoring node within each unit time period within the future preset time period.

[0150] Understandably, this embodiment serves as the coordinating hub of the entire forecasting system, where the sag deformation model (a mapping table describing how much the transmission line will sag given an increase in icing thickness) is activated. The sag deformation model directly receives the forecast results from the icing growth model as its input, thus converting meteorological forecast information into mechanical response information. The two models are interconnected, forming a seamless, data-driven forecasting pipeline: future meteorological forecast sequence -> icing growth model -> future icing thickness increment sequence -> sag deformation model -> future sag variable sequence -> future sag prediction value sequence.

[0151] It is understandable that the core calculations of the entire prediction process in this embodiment are two table lookups (ice growth model, sag deformation model) and two accumulations. The computational complexity is extremely low, making it fully applicable to real-time operation on resource-constrained edge monitoring equipment. This enables advanced preventive early warning capabilities to be deployed at low cost and on a large scale to every important line.

[0152] S6. Cross-validate the predicted values ​​of icing and sag, and issue graded early warnings and initiate targeted pre-treatment measures based on the cross-validation results.

[0153] Understandably, this embodiment abandons the "large and comprehensive" accumulation of general data. Instead, it constructs a highly correlated and robust dedicated sensing combination based on the physical essence (low temperature, high humidity, condensation) and apparent characteristics (thickening of shape, sag) of icing formation. It creatively introduces a dual-dimensional sensing and joint decision-making mechanism for icing load (ice thickness) and mechanical response (sag). In this concept, the sag deformation model acts as the quantifier of mechanical response. Its collaborative logic is as follows: the icing growth model predicts / monitors the input (increased icing thickness), while the sag deformation model evaluates the system state response (sag deformation value) in real time. The early warning system then fuses, compares, and cross-validates these two types of information to trigger the early warning level with the highest degree of matching with the actual risk and targeted pre-treatment measures. This forms a synergy of "prediction-guided precise intervention," avoiding the practicality problems of uncertainty in existing complex models and the lack of forward-looking early warning capabilities, greatly improving de-icing efficiency and the safety of transmission line operation and maintenance.

[0154] Specifically, in this embodiment, the predicted icing values ​​and sag values ​​are cross-validated. Based on the cross-validation results, graded early warnings are issued and targeted pre-response measures are initiated, including:

[0155] S61 defines a cross-risk matrix with ice thickness as the horizontal axis and sag amplitude as the vertical axis. Each cell in the cross-risk matrix corresponds to a comprehensive early warning level and a targeted pre-treatment measure.

[0156] In this embodiment, the comprehensive warning level output by the cross-risk matrix is ​​not the vague "high risk" or abstract comprehensive risk score of the prior art, but a specific description such as "high ice thickness - medium sag risk" or "medium ice thickness - high sag risk".

[0157] Understandably, this embodiment transforms the existing "single-index early warning" into "multi-dimensional state space localization and decision-making," elevating early warning judgment from a scalar comparison problem to a vector localization problem in a two-dimensional state space of ice thickness and sag. The key role of the sag deformation model is to provide a second state dimension, independent of ice thickness, that is indispensable for constituting this decision space.

[0158] S62, the calculated icing prediction value and sag prediction value are used as coordinate points to locate in the cross-risk matrix, and the movement trajectory of the coordinate points in the cross-risk matrix within the future preset time period is obtained, so as to determine the evolution of the future warning level and the targeted pre-response measures to be initiated.

[0159] Understandably, by predicting the movement trajectory of coordinate points in the future cross-risk matrix, we can move from "where the risk is high now" to "when and how the risk will evolve in the future," providing unprecedentedly refined guidance for preventative resource allocation.

[0160] For example, the implementation method for cross-validating and deciding on measures based on icing predictions and sag predictions can be as follows:

[0161] Scenario A (Consistent Verification, Enhanced Early Warning): If both the icing prediction and sag prediction fall into the "high-risk" range (e.g., high ice thickness, large sag), the highest level of warning is triggered (e.g., "high ice thickness - high sag risk"), and the most urgent de-icing pre-treatment measures are initiated (e.g., immediately coordinating a drone swarm equipped with specially designed vibrators to simulate manual striking frequencies and break the thin ice layer). At this point, the icing prediction and sag prediction mutually corroborate each other, eliminating the possibility of false alarms from a single sensor and maximizing the confidence level of the warning.

[0162] Scenario B (Revealing Hidden Risks, Avoiding Missed Reports): If the predicted icing value is in the low-to-medium risk range, but the predicted sag value is abnormally high (the ice is not very thick, but the sag is excessive), this indicates other risks not primarily caused by ice thickness, such as: line slack, fitting slippage, or the presence of undetected concentrated ice formations. The system will therefore trigger a "medium ice thickness - high sag risk" warning and initiate a special inspection (such as deploying a high-definition pan-tilt unit or dispatching a drone for detailed inspection) instead of directly de-icing. This is a function that existing monitoring models struggle to achieve, effectively avoiding serious false alarms or missed reports due to different threat sources.

[0163] Scenario C (Identifying Errors or Delays, Optimizing Response): If the predicted icing value is in the high-risk range, but the predicted sag value has not yet increased significantly, this indicates a physical delay in the mechanical deformation response due to rapid icing growth. The system will trigger a "High Ice Thickness - Medium Sag Risk" warning. Targeted pre-emptive measures will focus on "continuous close monitoring" and "preparing de-icing resources," rather than immediately executing de-icing actions. This avoids warning delays caused by mechanical response lags and prevents over-response before deformation reaches a dangerous level.

[0164] It can be seen that the handling logic of scenarios B and C essentially utilizes the inconsistency of two-dimensional information to perform preliminary fault diagnosis and root cause analysis. A significant deviation between the predicted icing value and the predicted sag value is a strong anomaly signal, driving the system to perform a more in-depth investigation, demonstrating the intelligence in the perception-assessment-decision closed loop. The cross-risk matrix defined in this embodiment is essentially a well-defined and interpretable set of rules. This set of rules transforms complex risk assessment into a rapid lookup and logical judgment of the coordinates of (predicted icing value, predicted sag value), making the decision-making process transparent, traceable, and manually verifiable.

[0165] Understandably, this embodiment, through in-depth interaction and comparison of icing thickness and sag amplitude information, can identify "atypical icing" or "compound" disaster risks. The output comprehensive early warning level directly guides maintenance personnel to take the most targeted measures (whether to prioritize de-icing or to prioritize line inspection), achieving precise alignment between early warning and response and improving emergency response efficiency. This is of great significance for ensuring power grid safety, because many serious accidents are not caused by the thickest icing, but by hidden dangers such as uneven icing and line defects that are difficult to detect under ordinary monitoring.

[0166] Example 2:

[0167] Please see the appendix Figure 2 , Figure 2 This is a schematic diagram of a data-driven transmission line icing disaster early warning system provided as an embodiment of this specification.

[0168] like Figure 2As shown, the data-driven transmission line icing disaster early warning system may include at least:

[0169] Model building module 1 is used to build an ice growth model that reflects the mapping relationship between meteorological data and ice thickness increment based on historical statistical data, and to build an arc sag deformation model that reflects the mapping relationship between ice thickness increment and transmission line arc sag variable.

[0170] Data acquisition module 2 is used to acquire dual-light monitoring images of the transmission line in real time and meteorological forecast data for a preset time period in the future;

[0171] The icing prediction module 3 is used to input meteorological forecast data into the icing growth model, and based on the real-time acquired dual-light monitoring images and the output of the icing growth model, to calculate the icing growth trend within a preset time period and obtain the icing prediction value.

[0172] The sag prediction module 4 is used to input the output of the icing growth model into the sag deformation model, and based on the real-time acquired dual-light monitoring images and the output of the sag deformation model, to estimate the trend of sag deformation within a preset time period and obtain the sag prediction value.

[0173] The disaster early warning module 5 is used to cross-validate the icing prediction value and the sag prediction value, and to issue graded early warnings and initiate targeted pre-response measures based on the cross-validation results.

[0174] It is understood that the technical concept of the data-driven transmission line icing disaster early warning system provided in this embodiment is similar to the technical concept of the aforementioned data-driven transmission line icing disaster early warning method, and will not be repeated here.

[0175] The above description is merely a preferred embodiment disclosed in this application and an explanation of the technical principles employed. Those skilled in the art should understand that the scope of protection involved in this disclosure is not limited to technical solutions formed by specific combinations of the above-described technical features, but should also cover other technical solutions formed by arbitrary combinations of the above-described technical features or their equivalents without departing from the above-disclosed concept. For example, technical solutions formed by substituting the above features with (but not limited to) technical features with similar functions disclosed in this disclosure.

[0176] Furthermore, while the operations are described in a specific order, this should not be construed as requiring these operations to be performed in the specific order shown or in a sequential order. In certain environments, multitasking and parallel processing may be advantageous. Similarly, while several specific implementation details are included in the above discussion, these should not be construed as limiting the scope of this disclosure. Certain features described in the context of individual embodiments may also be implemented in combination in a single embodiment. Conversely, various features described in the context of a single embodiment may also be implemented individually or in any suitable sub-combination in multiple embodiments.

Claims

1. A data-driven early warning method for icing disasters on power transmission lines, characterized in that, Includes the following steps: Based on historical statistical data, an ice growth model is constructed that reflects the mapping relationship between meteorological data and the increase in ice thickness. Construct a sag deformation model that reflects the mapping relationship between the increase in icing thickness and the sag deformation of the transmission line; Real-time acquisition of dual-light monitoring images of transmission lines and meteorological forecast data for a preset time period; Meteorological forecast data is input into the icing growth model, and based on real-time acquired dual-light monitoring images and the output of the icing growth model, the icing growth trend within a preset time period is calculated to obtain the icing prediction value. The output of the ice growth model is input into the sag deformation model, and based on the real-time acquired dual-light monitoring images and the output of the sag deformation model, the trend of sag deformation in the future preset time period is calculated to obtain the sag prediction value. Cross-validate the predicted values ​​of icing and sag, and issue graded early warnings and initiate targeted pre-response measures based on the cross-validation results.

2. The data-driven early warning method for icing disasters on transmission lines as described in claim 1, characterized in that, Based on historical statistical data, an ice growth model reflecting the mapping relationship between meteorological data and ice thickness increments is constructed, including: Obtain a reference image of the transmission line and extract the reference width of the transmission line from the reference image; Acquire historical dual-light monitoring images of transmission lines and historical meteorological data for the corresponding areas; Based on historical dual-light monitoring images, the historical visual width of several transmission line segments corresponding to each monitoring node was calculated. The historical icing thickness of the transmission line is obtained by directly calculating the difference between the historical visual width and the baseline width. The difference between historical icing thicknesses in adjacent time periods is calculated to obtain the historical icing thickness increment of the transmission line; Based on historical ice thickness increments and corresponding regional historical meteorological data, a lookup table-based ice growth model is constructed to reflect the mapping relationship between meteorological data and ice thickness increments.

3. The data-driven early warning method for icing disasters on transmission lines as described in claim 2, characterized in that, Based on historical ice thickness increments and corresponding regional historical meteorological data, a lookup-based ice growth model reflecting the mapping relationship between meteorological data and ice thickness increments is constructed, including: Historical meteorological statistical features and topographic features of each monitoring node are extracted from the historical meteorological data of the region and fused to form a comprehensive feature vector of each monitoring node; Clustering algorithms were used to analyze the comprehensive feature vectors of all monitoring nodes, resulting in several cluster groups. Based on the historical ice thickness increments and historical meteorological statistical characteristics of all monitoring nodes in each cluster, a shared lookup table-based ice growth model is constructed for each cluster.

4. The data-driven early warning method for transmission line icing disasters as described in claim 3, characterized in that, Historical meteorological statistical characteristics include average temperature, average humidity, and average wind speed over each time period. Based on the historical icing thickness increments and historical meteorological statistical characteristics of all monitoring nodes in each cluster, a shared lookup-based icing growth model is constructed for each cluster, including: The average wind speed is used to group the data into intervals. Within each wind speed interval group, the average temperature and average humidity are used as the joint query keys, and the historical ice thickness increment obtained under the corresponding meteorological statistical characteristics is used as the query result. A multidimensional query table reflecting the mapping relationship between wind speed-temperature and humidity data and ice thickness increment is established, which is denoted as the ice growth model.

5. The data-driven early warning method for icing disasters on transmission lines as described in claim 1, characterized in that, A sag deformation model reflecting the mapping relationship between the increase in icing thickness and the sag deformation of the transmission line arc is constructed, including: Acquire a reference image of the transmission line and extract the reference horizontal position of the transmission line from the reference image; Acquire historical dual-light monitoring images of transmission lines; Based on historical dual-light monitoring images, the historical horizontal positions of several transmission line segments corresponding to each monitoring node were calculated. The historical sag amplitude of the transmission line is obtained by directly calculating the difference between the historical horizontal position and the benchmark horizontal position; The historical sag deformation of the transmission line is obtained by calculating the difference between the historical sag amplitudes of adjacent time periods. Based on historical sag variables and corresponding historical icing thickness increments over time periods, a lookup table-based sag deformation model is constructed to reflect the mapping relationship between icing thickness increments and transmission line sag variables.

6. The data-driven early warning method for icing disasters on transmission lines as described in claim 5, characterized in that, Based on historical sag variables and corresponding historical icing thickness increments over time periods, a lookup-based sag deformation model is constructed to reflect the mapping relationship between icing thickness increments and transmission line sag variables, including: Based on historical sag variables and corresponding historical icing thickness increments over the same time period, time-synchronized data pairs (icing thickness increment, sag variables) are obtained. The data pairs (ice thickness increment, sag variable) are grouped into intervals according to the ice thickness increment. Within each ice thickness increment interval group, the statistical value of the sag variable in each data pair is calculated to obtain the typical sag variable value. Using the ice thickness increment range as the query key and the typical sag deformation value as the query result, a two-dimensional query table reflecting the mapping relationship between the ice thickness increment and the sag deformation of the transmission line is established, denoted as the sag deformation model.

7. The data-driven early warning method for icing disasters on transmission lines as described in claim 2, characterized in that, Meteorological forecast data is input into the icing growth model, and based on real-time acquired dual-light monitoring images and the output of the icing growth model, the icing growth trend within a preset time period is calculated to obtain icing prediction values, including: Meteorological statistical characteristics of monitoring nodes for each unit time period within a preset future time period are extracted from meteorological forecast data; By inputting meteorological statistical characteristics into the icing growth model, the output is the increase in icing thickness of the monitoring node in each unit time period within a preset future time period. Based on real-time acquired dual-light monitoring images, the real-time visual width of several transmission line segments corresponding to the monitoring nodes is calculated. The real-time icing thickness of the transmission line is obtained by directly calculating the difference between the real-time visual width and the reference width. The real-time icing thickness of each transmission line segment corresponding to the monitoring node is accumulated with the icing thickness increment for each unit time period within the future preset time period to obtain the predicted icing value for each transmission line segment corresponding to the monitoring node for each unit time period within the future preset time period.

8. The data-driven early warning method for icing disasters on transmission lines as described in claim 5, characterized in that, The output of the icing growth model is input into the sag deformation model, and based on the real-time acquired dual-light monitoring images and the output of the sag deformation model, the trend of sag deformation within a preset time period is calculated to obtain the predicted sag value, including: The output of the icing growth model is input into the sag deformation model, and the output is the typical sag deformation value of the monitoring node in each unit time period within a preset time period in the future. Based on real-time acquired dual-light monitoring images, the real-time horizontal position of several transmission line segments corresponding to the monitoring nodes is calculated. The real-time sag amplitude of the transmission line is obtained by directly calculating the difference between the real-time horizontal position and the reference horizontal position. The real-time sag amplitude of each transmission line segment corresponding to the monitoring node is accumulated with the typical sag deformation value of each unit time period within the future preset time period to obtain the predicted sag value of each transmission line segment corresponding to the monitoring node within each unit time period within the future preset time period.

9. The data-driven early warning method for transmission line icing disasters as described in claim 1, characterized in that, Cross-validation was performed on the predicted icing and sag values. Based on the cross-validation results, tiered early warnings were issued and targeted pre-response measures were initiated, including: With ice thickness as the horizontal axis and sag amplitude as the vertical axis, a cross-risk matrix is ​​defined. Each cell in the cross-risk matrix corresponds to a comprehensive early warning level and a targeted pre-treatment measure. The calculated icing and sag prediction values ​​are used as coordinate points to locate the cross-risk matrix, and the movement trajectory of the coordinate points in the cross-risk matrix within a preset time period is obtained to determine the evolution of the future warning level and the targeted pre-response measures to be initiated.

10. A data-driven early warning system for icing disasters on power transmission lines, characterized in that, include: The model building module is used to build an ice growth model that reflects the mapping relationship between meteorological data and ice thickness increment based on historical statistical data, and to build an arc sag deformation model that reflects the mapping relationship between ice thickness increment and transmission line arc sag variable. The data acquisition module is used to acquire real-time dual-light monitoring images of the transmission line and meteorological forecast data for a preset time period. The icing prediction module is used to input meteorological forecast data into the icing growth model, and based on the real-time acquired dual-light monitoring images and the output of the icing growth model, to estimate the icing growth trend within a preset time period and obtain the icing prediction value. The sag prediction module is used to input the output of the icing growth model into the sag deformation model, and based on the real-time acquired dual-light monitoring images and the output of the sag deformation model, to estimate the trend of sag deformation within a preset time period and obtain the sag prediction value. The disaster early warning module is used to cross-validate the icing prediction value and the sag prediction value, and to issue graded early warnings and initiate targeted pre-response measures based on the cross-validation results.

Images