Ice-coated conductor galloping characteristics fast prediction method and system based on image driving

Abstract

This invention discloses a rapid prediction method and system for the galloping characteristics of iced conductors based on image-driven computation. The method involves image encoding of acquired iced conductor images, transforming the conductor and icing feature parameters into a two-dimensional image matrix at a uniform scale. A constructed image modeling network extracts high-dimensional structural features from the obtained two-dimensional image matrix, which are then transformed into aerodynamic parameter vectors through a nonlinear mapping function. These aerodynamic parameter vectors are coupled with the conductor's structural dynamic parameters to construct a nonlinear coupled vibration system. Based on time-domain integration and multimodal response analysis, the galloping response characteristics under different icing morphologies and environmental conditions are obtained. Compared to traditional wind tunnel or complex flow field simulations, this method significantly improves the modeling efficiency and practicality of wind-induced conductor response characteristics under varying icing conditions, demonstrating good engineering adaptability and widespread application value.

Description

Technical Field

[0001] This invention belongs to the field of power system safety technology. It relates to a method and system for rapid prediction of the galloping characteristics of icy conductors based on image-driven methods. Background Technology

[0002] With the rapid deployment of ultra-high voltage and extra-high voltage power transmission projects in complex meteorological regions, the operational safety of transmission lines under extreme weather conditions has become increasingly critical. Especially in non-uniform climatic environments with frequent icing, snowfall, and freezing rain, complex and variable ice-like deposits easily form on the conductor surface, significantly altering the conductor's shape, mass distribution, and aerodynamic boundary conditions. This results in a highly nonlinear and time-varying wind-induced response mechanism. Among these, galloping, a typical low-frequency, large-amplitude self-excited vibration mode, can not only cause conductor strand breakage, hardware failure, and tower structural damage, but also potentially lead to localized arcing and even line tripping accidents, seriously threatening the stable operation of the power grid.

[0003] Traditional studies on conductor galloping characteristics mainly rely on wind tunnel tests and numerical simulations. While these methods offer some accuracy in acquiring static aerodynamic characteristics, they still have significant shortcomings in the following aspects: First, wind tunnel tests are costly, parameter tuning is difficult, and they struggle to cover the diversity and evolution of complex natural ice formations. Second, numerical modeling is highly sensitive to boundary conditions and mesh generation, limiting its stability and reliability in predicting multimodal vibration responses. Third, existing methods often rely on explicit modeling of the three aerodynamic force coefficients (lift, drag, and torque), making it difficult to generalize to unknown ice formations or heterogeneous conductor structures when data is scarce or structural information is incomplete.

[0004] In recent years, with the development of computer vision and deep learning, structured image-driven intelligent modeling methods have begun to show significant advantages in complex structure recognition and response prediction. Novel image modeling networks such as the Vision Transformer (ViT) possess strong expressive capabilities in extracting high-dimensional spatial features and understanding non-Euclidean geometric relationships, providing new avenues for modeling unstructured input information. However, purely data-driven models often lack inherent constraints on physical laws, easily leading to problems such as prediction drift and unstable responses, making it difficult to meet the dual requirements of power grid safety monitoring for prediction accuracy and physical consistency. Summary of the Invention

[0005] To address the technical problems of low modeling efficiency, poor prediction accuracy, and insufficient engineering applicability of wind-induced galloping response of transmission lines under complex icing conditions, this invention aims to provide an image-driven method and system for rapid prediction of galloping characteristics of iced transmission lines. This method enables rapid prediction and evolution analysis of the galloping response of transmission lines with different icing conditions under typical wind fields without relying on wind tunnel tests or complex simulations, significantly improving the modeling efficiency and engineering applicability of this type of problem.

[0006] The technical solution to achieve the purpose of this invention is as follows:

[0007] A fast prediction method for image-driven galloping characteristics of icy conductors includes the following steps:

[0008] Image encoding is performed on the acquired images of the icy conductor, and the conductor and icing feature parameters are transformed into a two-dimensional image matrix at a uniform scale;

[0009] The high-dimensional structural features of the obtained two-dimensional image matrix are extracted by constructing an image modeling network, and the extracted high-dimensional structural features are transformed into aerodynamic parameter vectors through a nonlinear mapping function.

[0010] By coupling aerodynamic parameter vectors with the dynamic parameters of the conductor structure, a nonlinear coupled vibration system is constructed. Based on time-domain integration and multimodal response analysis, the galloping response characteristics under different icing morphologies and environmental conditions are obtained.

[0011] In a preferred technical solution, image encoding is performed on the acquired icy conductor image, transforming the conductor and icing feature parameters into a two-dimensional image matrix at a uniform scale, including:

[0012] Image encoding is performed using a contour-texture dual-channel grayscale mapping method. The contour channel maps the contour of the ice-covered cross-section and the slope field of the edge perturbation into grayscale information; the texture channel extracts the local gradient distribution and texture features of the cross-sectional area.

[0013] Information obtained from the contour and texture channels is fused and mapped into a unified two-dimensional image matrix I. ice = f enc (P s G i W e ),in, f enc ( ) represents the fusion mapping function, P s G represents the structural parameters of the conductor. i W represents the geometric parameters of the icing. e This represents the wind field variable.

[0014] The preferred technical solution further includes: normalizing and morphologically reconstructing the image before image encoding to eliminate background noise interference and enhance edge contours and structural texture information; and introducing image enhancement algorithms, including adaptive histogram equalization, edge sharpening and texture enhancement, into the image encoding process so that the generated two-dimensional image matrix can reflect the geometric and texture differences under different icing morphologies.

[0015] In the preferred technical solution, the extraction of high-dimensional structural features includes:

[0016] An image modeling network composed of a multi-head self-attention mechanism and local structure embedding units is constructed. A hierarchical visual converter structure is used as the backbone to divide the two-dimensional image matrix into multiple fixed-size image blocks and embed position codes to capture the correlation between local and global structures in high-dimensional space.

[0017] The multi-head self-attention mechanism dynamically allocates attention weights among different image patches, enabling the focus to be on key features such as the edges of conductors, abrupt changes in ice thickness, and asymmetrical distribution regions.

[0018] The local structure embedding unit extracts local texture gradient and edge perturbation information through convolution or image local encoder, and fuses it with the global features obtained through the visual converter structure to obtain high-dimensional structural features.

[0019] In the preferred technical solution, after being converted into an aerodynamic parameter vector, the following is also included:

[0020] A physical consistency loss function is introduced during the network training phase of the nonlinear mapping function. L phys Combined with mean square error loss L MSE Joint optimization, total loss function for:

[0021]

[0022] in, Used to control the smoothness of the model output. l,b To dynamically adjust the weighting coefficients;

[0023] A multi-objective regression head is configured at the output of the nonlinear mapping function to perform collaborative learning on the response variables of lift coefficient, drag coefficient, and aerodynamic torque coefficient;

[0024] By employing a dynamic weight reconstruction strategy, the weights of each aerodynamic parameter in the loss function are adaptively adjusted to balance the prediction accuracy of different response variables.

[0025] In the preferred technical solution, the physical consistency loss function L phys for:

[0026]

[0027] in, This represents the j-th component of the aerodynamic parameter vector A. This represents the parameters of the neural network model, where n represents the number of response dimensions. F j () represents the response coupling operator derived from the aerodynamic equation.

[0028] In the preferred technical solution, constructing a nonlinear coupled vibration system includes:

[0029] The predicted aerodynamic parameter vector A is mapped to a time-domain dynamic wind load. Combined with the dynamic parameters of the conductor structure, a nonlinear coupled vibration system is established, which is expressed as:

[0030]

[0031] in, u(t) This represents the displacement vector of the conductor's galloping response. for u(t) The first derivative, for u(t) The second derivative of , M is the mass matrix, K is the stiffness matrix, and C is the damping matrix.

[0032] In the preferred technical solution, based on time-domain integration and multimodal response analysis, the galloping response characteristics under different icing morphologies and environmental conditions are obtained, including:

[0033] The modal reduction method is used to perform characteristic mode decomposition on the conductor structure, retaining the low-frequency modes involved in the galloping, and truncating or approximating the high-order modes;

[0034] Considering the participation and frequency energy distribution of different modes under typical wind-induced excitation modes, the mass, stiffness, and damping mode coupling matrix is ​​used. To achieve dynamic analysis of multimodal responses, Here is the modal matrix;

[0035] For nonlinear coupled vibration systems, explicit or implicit integration algorithms suitable for nonlinear systems are used to solve u(t); during the integration process, aerodynamic-structural coupling damping correction, adaptive time step adjustment, and energy conservation constraints are introduced.

[0036] The key response indicators for conductor galloping after solving include: displacement amplitude and displacement time history, response frequency spectrum and modal energy distribution, dynamic instability boundary and critical wind speed, and the evolution path of modal energy density over time.

[0037] This invention also discloses an image-driven system for rapid prediction of the galloping characteristics of icy conductors, comprising:

[0038] The icy conductor image encoding module encodes the acquired icy conductor images, converting the conductor and icing feature parameters into a two-dimensional image matrix at a uniform scale.

[0039] The aerodynamic parameter vector generation module extracts high-dimensional structural features from the obtained two-dimensional image matrix through the constructed image modeling network, and transforms the extracted high-dimensional structural features into aerodynamic parameter vectors through a nonlinear mapping function.

[0040] The icing conductor galloping characteristic prediction module couples aerodynamic parameter vectors with conductor structure dynamic parameters to construct a nonlinear coupled vibration system. Based on time-domain integration and multimodal response analysis, it obtains the galloping response characteristics under different icing morphologies and environmental conditions.

[0041] The present invention also discloses a computer storage medium storing a computer program, wherein when the computer executes the computer program, it implements the image-driven method for rapid prediction of the galloping characteristics of icy conductors as described above.

[0042] The present invention also discloses an electronic device, including a memory and a processor, wherein the memory stores a computer program, and the processor runs the computer program stored in the memory. When the computer program is executed, it implements the image-driven rapid prediction method for the galloping characteristics of icy conductors described above.

[0043] Compared with the prior art, the significant advantages of this invention are:

[0044] 1. The image-driven approach avoids extensive wind tunnel testing and numerical modeling. Starting from the image representation of icy conductors, a contour-texture dual-channel grayscale mapping strategy is adopted. While preserving the original geometric boundary contour features, it enhances the ability to express local detail changes (such as asymmetric protrusions, bifurcation regions, etc.), thereby improving the adaptability of the subsequent perception model to the sensitivity of image structure. This greatly improves the efficiency and operability of the model response prediction, and is especially suitable for disaster prevention and early warning scenarios that require rapid response in practical engineering.

[0045] 2. By integrating deep visual models with nonlinear dynamic analysis, a refined modeling of the relationship between complex icing geometry and structural response is achieved, significantly reducing prediction errors compared to traditional empirical regression models.

[0046] 3. The model supports generalized modeling of various icing morphologies, including irregular, asymmetric, and non-uniform icing structures, and has good scalability and engineering adaptability.

[0047] 4. For the first time, the visual ViT model was introduced into the transmission line galloping prediction task, realizing an integrated, end-to-end modeling process from image to response, and pushing the application boundaries of intelligent sensing technology in power grid structure safety assessment.

[0048] 5. The method of this invention is applicable to actual transmission lines under different climate regions, tower types and conductor layouts, and has good prospects for promotion and application. It can provide power companies with strong technical support for structural status assessment, wind disaster risk prediction and optimization of anti-galling control measures. Attached Figure Description

[0049] Figure 1 A flowchart of an image-driven method for rapid prediction of the galloping characteristics of icy conductors;

[0050] Figure 2 A flowchart illustrating the implementation of an image-driven system for rapid prediction of the galloping characteristics of icy conductors.

[0051] Figure 3 A simplified schematic diagram of an ice-covered cross-section of a conductor;

[0052] Figure 4 This is the pixel grayscale encoding image corresponding to the ice-covered cross section of a conductor. Detailed Implementation

[0053] The principle of this invention is as follows: starting from the image representation of conductor icing, the geometric shape of icing is used as the core input source for structural response prediction. Through multi-stage deep integration of image grayscale modeling, visual feature extraction and parameter mapping, and dynamic response prediction, an end-to-end prediction framework is constructed, which significantly improves the system's ability to analyze complex icing morphologies and the mapping accuracy between them and aerodynamic response.

[0054] Example:

[0055] like Figure 1 As shown, an image-driven method for rapid prediction of the galloping characteristics of icy conductors includes the following steps:

[0056] Image encoding is performed on the acquired images of the icy conductor, and the conductor and icing feature parameters are transformed into a two-dimensional image matrix at a uniform scale;

[0057] The high-dimensional structural features of the obtained two-dimensional image matrix are extracted by constructing an image modeling network, and the extracted high-dimensional structural features are transformed into aerodynamic parameter vectors through a nonlinear mapping function.

[0058] By coupling aerodynamic parameter vectors with the dynamic parameters of the conductor structure, a nonlinear coupled vibration system is constructed. Based on time-domain integration and multimodal response analysis, the galloping response characteristics under different icing morphologies and environmental conditions are obtained.

[0059] In a preferred embodiment, image encoding is performed on the acquired image of the icy conductor, transforming the conductor and icing feature parameters into a two-dimensional image matrix at a uniform scale, including:

[0060] Image encoding is performed using a contour-texture dual-channel grayscale mapping method. The contour channel maps the contour of the ice-covered cross-section and the slope field of the edge perturbation into grayscale information; the texture channel extracts the local gradient distribution and texture features of the cross-sectional area.

[0061] The dual-channel information (information obtained from the contour channel and texture channel) is fused and mapped into a unified two-dimensional image matrix I. ice = f enc (P s G i W e ),in, f enc ( ) represents the fusion mapping function, P s G represents the structural parameters of the conductor. i W represents the geometric parameters of the icing. e This represents the wind field variable.

[0062] In a preferred embodiment, before image encoding, the image is further normalized and morphologically reconstructed to eliminate background noise interference and enhance edge contours and structural texture information; during image encoding, image enhancement algorithms are introduced, including adaptive histogram equalization, edge sharpening and texture enhancement, so that the generated two-dimensional image matrix can reflect the geometric and texture differences under different icing morphologies.

[0063] In a preferred embodiment, extracting high-dimensional structural features includes:

[0064] An image modeling network composed of a multi-head self-attention mechanism and local structure embedding units is constructed. A hierarchical visual converter structure is used as the backbone to divide the two-dimensional image matrix into multiple fixed-size image blocks and embed position codes to capture the correlation between local and global structures in high-dimensional space.

[0065] The multi-head self-attention mechanism dynamically allocates attention weights among different image patches, enabling the focus to be on key features such as the edges of conductors, abrupt changes in ice thickness, and asymmetrical distribution regions.

[0066] The local structure embedding unit extracts local texture gradient and edge perturbation information through convolution or image local encoder, and fuses it with global features to obtain high-dimensional structural features.

[0067] The global features here refer to the high-level structural features extracted across the entire input image by the visual converter structure, which are used to characterize the overall geometric shape, circumferential distribution pattern, and global correlation between different regions of the icy conductor cross-section.

[0068] In a preferred embodiment, after being converted into an aerodynamic parameter vector, the method further includes:

[0069] A physical consistency loss function is introduced during the network training phase using a nonlinear mapping function. L phys Combined with mean square error loss L MSE Joint optimization, total loss function for:

[0070]

[0071] in, Used to control the smoothness of the model output. l,b To dynamically adjust the weighting coefficients;

[0072] A multi-objective regression head is configured at the output of the nonlinear mapping function to perform collaborative learning on the response variables of lift coefficient, drag coefficient, and aerodynamic torque coefficient;

[0073] By employing a dynamic weight reconstruction strategy, the weights of each aerodynamic parameter in the loss function are adaptively adjusted to balance the prediction accuracy of different response variables.

[0074] In a preferred embodiment, the physical consistency loss function L phys for:

[0075]

[0076] in, This represents the j-th component of the aerodynamic parameter vector A. This represents the parameters of the neural network model, where n represents the number of response dimensions. F j () represents the response coupling operator derived from the aerodynamic equation.

[0077] In a preferred embodiment, constructing a nonlinear coupled vibration system includes:

[0078] The predicted aerodynamic parameter vector A is mapped to a time-domain dynamic wind load. Combined with the dynamic parameters of the conductor structure, a nonlinear coupled vibration system is established, which is expressed as:

[0079]

[0080] in, u(t) This represents the displacement vector of the conductor's galloping response. for u(t) The first derivative, for u(t) The second derivative of , M is the mass matrix, K is the stiffness matrix, and C is the damping matrix.

[0081] In a preferred embodiment, based on time-domain integration and multimodal response analysis, the galloping response characteristics under different icing morphologies and environmental conditions are obtained, including:

[0082] The modal reduction method is used to perform characteristic mode decomposition on the conductor structure, retaining the low-frequency modes involved in the galloping, and truncating or approximating the high-order modes;

[0083] Considering the participation and frequency energy distribution of different modes under typical wind-induced excitation modes, the mass, stiffness, and damping mode coupling matrix is ​​used. To achieve dynamic analysis of multimodal responses, Here is the modal matrix;

[0084] For nonlinear coupled vibration systems, explicit or implicit integration algorithms suitable for nonlinear systems are used to solve u(t); during the integration process, aerodynamic-structural coupling damping correction, adaptive time step adjustment, and energy conservation constraints are introduced.

[0085] The key response indicators for conductor galloping after solving include: displacement amplitude and displacement time history, response frequency spectrum and modal energy distribution, dynamic instability boundary and critical wind speed, and the evolution path of modal energy density over time.

[0086] In another embodiment, an image-driven system for rapid prediction of the galloping characteristics of icy conductors includes:

[0087] The icy conductor image encoding module encodes the acquired icy conductor images, converting the conductor and icing feature parameters into a two-dimensional image matrix at a uniform scale.

[0088] The aerodynamic parameter vector generation module extracts high-dimensional structural features from the obtained two-dimensional image matrix through the constructed image modeling network, and transforms the extracted high-dimensional structural features into aerodynamic parameter vectors through a nonlinear mapping function.

[0089] The icing conductor galloping characteristic prediction module couples aerodynamic parameter vectors with conductor structure dynamic parameters to construct a nonlinear coupled vibration system. Based on time-domain integration and multimodal response analysis, it obtains the galloping response characteristics under different icing morphologies and environmental conditions.

[0090] The following example illustrates the workflow of an image-driven system for rapid prediction of the galloping characteristics of icy conductors, including the following steps:

[0091] The method proposed in this invention takes image recognition and deep learning as its core, breaks through the dependence of traditional numerical modeling on explicit geometric description, and adopts an image-driven three-stage prediction architecture, which includes: an icing conductor image encoding module, an aerodynamic parameter vector generation module, and an icing conductor galloping characteristic prediction module. The three work together to achieve efficient conversion from the input of the conductor icing cross-section image to the output galloping response index.

[0092] I. Image Encoding Module for Iced Conductors

[0093] This implementation first introduces image recognition as a carrier for representing complex icing morphology. Through grayscale encoding and texture extraction techniques, the cross-sectional information of icing conductors with shape differences is input into the model system in the form of a pixel matrix.

[0094] Specifically, high-resolution cameras, CT slices, or scanning devices are used to sample cross-sections of typical icy conductors to acquire raw image data. To ensure the integrity and distinguishability of information representation, the image processing module first performs image normalization and morphological reconstruction to eliminate background noise interference, while enhancing edge contours and structural texture information.

[0095] The ice-covered sample construction covers typical ice types (shell ice, curtain ice, mixed ice) and ice thickness ranges [d] min ,d max [Tractor structure types (single conductor, double-split, quadruple-split, etc.) and typical wind speed disturbance model V(t)=V0+δ] sin(ωt) (where V(t) is the instantaneous wind speed at time t; V0 is the average wind speed; δ is the wind speed disturbance amplitude; ω is the wind speed disturbance angular frequency; t is the time variable), construct a multi-scale structure-environment joint database.

[0096] During the encoding process, a contour-texture dual-channel grayscale mapping strategy was proposed: the contour channel is used to capture the geometric shape features of the icing cross-section. Edge detection operators (such as the Canny operator and the Sobel gradient operator) are used to extract the boundary contours of the conductors and the icing, and the local perturbation slope field is further calculated to reflect the irregular protrusions and asymmetries of the icing morphology. The texture channel is used to express local details and material microstructure features. By constructing the image gradient field and the grayscale co-occurrence matrix, subtle texture changes on the icing surface (such as differences in ice thickness and non-uniform distribution areas) are captured to enhance the model's sensitivity to perturbations of the icing microstructure.

[0097] The dual-channel information (information obtained from the contour channel and texture channel) is then mapped into a unified two-dimensional grayscale matrix representation I through a fusion strategy. ice = f enc (P s Gi W e ), where P s Indicates conductor structural parameters (such as diameter, number of splits), G i W represents the geometric parameters of icing (such as thickness, type, and asymmetry). e This represents wind field variables (such as wind speed and wind direction). This fusion mapping function... f enc ( To ensure consistency in spatial resolution and grayscale of the input image, the two channels are fused to construct a composite image input. This image encoding incorporates conductor cross-sectional contours, local perturbation features, and texture gradient information, thereby improving the sensitivity and stability of the image representation to changes in the microstructure of icing.

[0098] To further enhance the robustness of image representation, this implementation also incorporates image enhancement algorithms such as adaptive histogram equalization, edge sharpening, and texture enhancement during the grayscale encoding process. This ensures that the generated grayscale image accurately reflects the geometric and texture differences under different icing morphologies, providing high-fidelity input for the subsequent aerodynamic parameter vector generation module.

[0099] Through the above processing, the geometric and textural features of complex icy conductors are transformed into standardized and computable grayscale image representations, breaking through the dependence of traditional methods on explicit geometric modeling and providing unified, stable and physically sensitive input data for deep learning models.

[0100] II. Aerodynamic Parameter Vector Generation Module

[0101] After completing the grayscale encoding of the icing conductor image, the second stage of this implementation is the "aerodynamic parameter vector generation module." The core objective of this module is to generate a standardized image representation I. ice The vector is transformed into a high-dimensional vector A that can characterize the aerodynamic response of the conductor, realizing a nonlinear mapping from visual information to aerodynamic parameters, and providing input for subsequent galloping response prediction.

[0102] The specific implementation steps are as follows:

[0103] 1. Image Feature Extraction

[0104] This module employs a deep image modeling network based on Vision Transformer (ViT) as the backbone for feature extraction. ViT captures the correlation between local and global structures in a high-dimensional space by dividing the input grayscale image into several fixed-size image patches and embedding positional codes.

[0105] To enhance sensitivity to the microstructure of ice accretion, the network is designed with a multi-head attention mechanism, which can dynamically allocate attention weights among different regions of the image, focusing on key features such as the edges of the conductor, abrupt changes in ice thickness, and asymmetric distribution areas.

[0106] Simultaneously, a local structure embedding unit is introduced, which extracts local texture gradient and edge perturbation information through convolution or image local encoder, and fuses it with global features to improve the ability to capture subtle geometric changes.

[0107] 2. Construction of nonlinear mappings

[0108] Extracted high-dimensional structural features F vit Through the nonlinear mapping function Φ θ Converted into aerodynamic parameter vector A:

[0109] in i The network parameters A can include multidimensional indicators (i.e. response variables) that characterize the aerodynamic behavior of the conductor, such as lift coefficient, drag coefficient, and aerodynamic torque coefficient, but do not need to rely on the traditional calculation method of the three-part force coefficients.

[0110] This mapping uses a deep neural network to learn complex nonlinear relationships, enabling the model to automatically capture the coupling patterns between icing geometry, wind field conditions, and aerodynamic response.

[0111] 3. Physical Consistency Constraints

[0112] To ensure that the predicted aerodynamic parameters satisfy fundamental physical laws, this implementation introduces a physical consistency loss function during the network training phase. L phys Combined with mean square error loss L MSE Joint optimization, total loss function for:

[0113]

[0114] in: L MSE Characterizes the error between the predicted value and the actual value; L phys Physical consistency constraints are applied to the predicted vector A using aerodynamic conservation equations or constraints based on conductor dynamics. Used to control the smoothness of the model output; l,b To dynamically adjust the weight coefficients, different targets can be adapted according to the training stage.

[0115] Physical consistency loss function L phys for:

[0116]

[0117] in This represents the j-th component of the aerodynamic response vector A; Represents the parameters of the neural network model; j represents the index of the j-th aerodynamic response component; n represents the response dimension, i.e., the number of components in the aerodynamic response vector A; F j This indicates that the response coupling operator, derived from the aerodynamic equation, constrains the predicted output and guides the model to learn structural response characteristics with physical consistency.

[0118] This constraint ensures that the model can still output aerodynamic parameters that conform to physical laws when faced with unknown icing morphology or heterogeneous conductor structures, thereby improving prediction reliability and generalization ability.

[0119] 4. Multi-objective regression and collaborative learning

[0120] The output is configured with a multi-task regression head to collaboratively learn multi-dimensional response variables such as lift, drag, and torque.

[0121] The network can adaptively adjust the weights of each aerodynamic parameter in the loss function through a dynamic weight reconstruction strategy, thereby balancing the prediction accuracy of different response variables and improving the overall prediction stability.

[0122] 5. High-fidelity input guarantee

[0123] To ensure the accuracy of aerodynamic parameter prediction, this implementation method can combine a multi-scale structure-environment joint database, including typical icing types, thickness ranges, conductor splitting forms, and wind speed disturbance models, to achieve high-fidelity training under complex environments and diverse icing conditions.

[0124] Through the above processing, the grayscale image input I... ice It is effectively transformed into a high-dimensional aerodynamic parameter vector A, which fully expresses the aerodynamic characteristics of the icing conductor under different wind fields, providing accurate and physically consistent input for the third-stage galloping response prediction module, and realizing the closed loop of image-driven aerodynamic characterization and dynamic response prediction.

[0125] III. Prediction Module for Icy Conductor Galloping Characteristics

[0126] After the aerodynamic parameter vector is generated, the third stage of this implementation is the "response characteristic prediction module". Its core objective is to couple the aerodynamic parameter vector A with the conductor structure dynamic parameters M, K, and C to construct a nonlinear multi-degree-of-freedom vibration system, thereby enabling rapid prediction and dynamic evolution analysis of the wind-induced galloping response of the icing conductor.

[0127] The specific implementation steps are as follows:

[0128] 1. Construction of Nonlinear Coupled Vibration System

[0129] The predicted aerodynamic parameter vector A is mapped to a time-domain dynamic wind load. And combined with the conductor structure parameters M (mass matrix), K (stiffness matrix), and C (damping matrix), a nonlinear coupled vibration equation is established:

[0130]

[0131] Where u(t) represents the displacement vector of the conductor galloping response.

[0132] Considering the nonlinear characteristics of iced conductors, such as large-amplitude self-excited vibration, aerodynamic-structural coupling stiffness variation, and modal coupling effects, nonlinear terms can be introduced into the system equations during numerical solutions. Corrections are made to stiffness and damping:

[0133]

[0134] It is the aerodynamic damping matrix, which is the equivalent damping effect caused by fluid-structure interaction. It is the aerodynamic stiffness matrix, which is the equivalent stiffness effect caused by aerodynamic forces.

[0135] 2. Modal order reduction and multimodal response analysis

[0136] To improve the solution efficiency, this implementation method adopts the modal reduction method. By performing characteristic mode decomposition on the conductor structure, the low-frequency modes that mainly participate in the dancing are retained, while the high-order modes are truncated or approximated to reduce the computational cost.

[0137] Considering the participation degree and frequency energy distribution of different modes under typical wind-induced excitation modes, the modal coupling matrix is ​​used to... To achieve dynamic analysis of multimodal responses, ensuring the ability to capture the evolution of dominant vibration modes and nonlinear modal interactions.

[0138] in, The modal matrix is ​​a matrix composed of the mode vectors (mode vectors) of each order of the structure.

[0139] 3. Time-domain integral and stability control

[0140] For nonlinear coupled vibration systems, the improved Newmark-β integral method or other explicit or implicit integration algorithms suitable for nonlinear systems are used to solve for u(t).

[0141] During the integration process, aerodynamic-structural coupling damping correction, adaptive time step adjustment, and energy conservation constraints are introduced to control the stability and accuracy of numerical integration and avoid divergent calculations under large amplitudes or critical wind speeds.

[0142] 4. Dance Response Output

[0143] After the system is solved, the key response indicators of conductor galloping are output, including but not limited to: displacement amplitude and displacement time history u(t); response frequency spectrum and modal energy distribution; dynamic instability boundary and critical wind speed; and the evolution path of modal energy density over time.

[0144] These indicators can be used to intuitively characterize the dynamic response characteristics and safety margin of icing conductors under complex wind fields.

[0145] 5. Parameter Uncertainty and Robustness Analysis

[0146] This implementation method can introduce uncertainties in parameters such as ice thickness, ice shape distribution, and wind speed disturbance, and perform sensitivity analysis on the system response through Monte Carlo simulation or random sampling methods.

[0147] During the prediction process, the distribution of galloping response under different icing morphologies and environmental conditions can be obtained, providing a reliable basis for engineering design and risk assessment.

[0148] Through the above processing, the aerodynamic parameter vector A is effectively transformed into the conductor galloping response u(t), realizing an end-to-end prediction closed loop from image grayscale representation to structural dynamic response. This module can efficiently and accurately capture the nonlinear vibration characteristics of complex iced conductors, significantly improving prediction efficiency and engineering adaptability, and providing a reliable data foundation for risk monitoring and decision support in smart grids.

[0149] IV. System Integration and Output Interfaces

[0150] The three phase modules described above are all encapsulated as independent function units, which can be called at the module level through Python scripts, MATLAB functions, or C++ dynamic link libraries, supporting GPU parallel training and inference. The overall system can be embedded into the power grid icing monitoring platform, combined with the conductor image acquisition terminal and the real-time wind field database, to construct an end-to-end response prediction process.

[0151] Ultimately, the system outputs include, but are not limited to: conductor displacement time history, vibration amplitude spectrum, dominant frequency components, and predicted ultimate displacement under different wind field conditions. It can also automatically generate a galloping response risk assessment report for operation and maintenance personnel to use for decision-making.

[0152] The method described in this invention is not only applicable to conventional power transmission conductors, but can also be extended to various structural forms such as overhead ground wires and composite material conductors, and has wide engineering applicability and scalability.

[0153] In another embodiment, a computer storage medium stores a computer program thereon, which, when executed by a computer, implements the image-driven rapid prediction method for the galloping characteristics of icy conductors described in any of the preceding embodiments.

[0154] In another embodiment, an electronic device includes a memory and a processor, wherein the memory stores a computer program, and the processor runs the computer program stored in the memory, wherein the computer program, when executed, implements the image-driven rapid prediction method for the galloping characteristics of icing conductors as described in any of the preceding embodiments.

[0155] The above embodiments are preferred embodiments of the present invention, but the embodiments of the present invention are not limited to the above embodiments. Any changes, modifications, substitutions, combinations, or simplifications made without departing from the spirit and principle of the present invention shall be considered equivalent substitutions and shall be included within the protection scope of the present invention.

Claims

1. A method for rapid prediction of the galloping characteristics of icy conductors based on image-driven imaging, characterized in that, Includes the following steps: Image encoding is performed on the acquired images of the icy conductor, transforming the conductor and icing feature parameters into a two-dimensional image matrix at a uniform scale; specifically, this includes: Image encoding is performed using a contour-texture dual-channel grayscale mapping method. The contour channel maps the contour of the ice-covered cross-section and the slope field of the edge perturbation into grayscale information; the texture channel extracts the local gradient distribution and texture features of the cross-sectional area. Information obtained from the contour and texture channels is fused and mapped into a unified two-dimensional image matrix I. ice = f enc (P s G i W e ),in, f enc ( ) represents the fusion mapping function, P s G represents the structural parameters of the conductor. i W represents the geometric parameters of the icing. e Represents wind field variables; A constructed image modeling network is used to extract high-dimensional structural features from the obtained two-dimensional image matrix. These extracted high-dimensional structural features are then transformed into aerodynamic parameter vectors using a nonlinear mapping function. The extraction of high-dimensional structural features includes: An image modeling network composed of a multi-head self-attention mechanism and local structure embedding units is constructed. A hierarchical visual converter structure is used as the backbone to divide the two-dimensional image matrix into multiple fixed-size image blocks and embed position codes to capture the correlation between local and global structures in high-dimensional space. The multi-head self-attention mechanism dynamically allocates attention weights among different image patches, enabling the focus to be on key features such as the edges of conductors, abrupt changes in ice thickness, and asymmetrical distribution regions. The local structure embedding unit extracts local texture gradient and edge perturbation information through convolution or image local encoder, and fuses it with the global features obtained through the visual converter structure to obtain high-dimensional structural features. By coupling aerodynamic parameter vectors with the dynamic parameters of the conductor structure, a nonlinear coupled vibration system is constructed. Based on time-domain integration and multimodal response analysis, the galloping response characteristics under different icing morphologies and environmental conditions are obtained; specifically including: The predicted aerodynamic parameter vector A is mapped to a time-domain dynamic wind load. Combined with the dynamic parameters of the conductor structure, a nonlinear coupled vibration system is established, which is expressed as: in, u(t) This represents the displacement vector of the conductor's galloping response. for u(t) The first derivative, for u(t) The second derivative of , M is the mass matrix, K is the stiffness matrix, C is the damping matrix; The modal reduction method is used to perform characteristic mode decomposition on the conductor structure, retaining the low-frequency modes involved in the galloping, and truncating or approximating the high-order modes; Considering the participation and frequency energy distribution of different modes under typical wind-induced excitation modes, the mass, stiffness, and damping mode coupling matrix is ​​used. To achieve dynamic analysis of multimodal responses, Here is the modal matrix; For nonlinear coupled vibration systems, explicit or implicit integration algorithms suitable for nonlinear systems are used to solve u(t); during the integration process, aerodynamic-structural coupling damping correction, adaptive time step adjustment, and energy conservation constraints are introduced. The key response indicators for conductor galloping after solving include: displacement amplitude and displacement time history, response frequency spectrum and modal energy distribution, dynamic instability boundary and critical wind speed, and the evolution path of modal energy density over time.

2. The image-driven method for rapid prediction of the galloping characteristics of icy conductors according to claim 1, characterized in that, Also includes: Before image encoding, the image is normalized and morphologically reconstructed to eliminate background noise interference and enhance edge contours and structural texture information. Image enhancement algorithms, including adaptive histogram equalization, edge sharpening, and texture enhancement, are introduced into the image encoding process so that the generated two-dimensional image matrix can reflect the geometric and texture differences under different icing morphologies.

3. The image-driven method for rapid prediction of the galloping characteristics of icy conductors according to claim 1, characterized in that, After being converted into an aerodynamic parameter vector, it also includes: A physical consistency loss function is introduced during the network training phase of the nonlinear mapping function. L phys Combined with mean square error loss L MSE Joint optimization, total loss function for: in, Used to control the smoothness of the model output. λ,β To dynamically adjust the weighting coefficients; A multi-objective regression head is configured at the output of the nonlinear mapping function to perform collaborative learning on the response variables of lift coefficient, drag coefficient, and aerodynamic torque coefficient; By employing a dynamic weight reconstruction strategy, the weights of each aerodynamic parameter in the loss function are adaptively adjusted to balance the prediction accuracy of different response variables.

4. The image-driven method for rapid prediction of galloping characteristics of icy conductors according to claim 3, characterized in that, Physical consistency loss function L phys for: in, This represents the j-th component of the aerodynamic parameter vector A. This represents the parameters of the neural network model, where n represents the number of response dimensions. F j () represents the response coupling operator derived from the aerodynamic equation.

5. A rapid prediction system for the galloping characteristics of icy conductors based on image-driven methods, used to implement the rapid prediction method for the galloping characteristics of icy conductors based on image-driven methods according to any one of claims 1-4, characterized in that, include: The icy conductor image encoding module encodes the acquired icy conductor images, converting the conductor and icing feature parameters into a two-dimensional image matrix at a uniform scale. The aerodynamic parameter vector generation module extracts high-dimensional structural features from the obtained two-dimensional image matrix through the constructed image modeling network, and transforms the extracted high-dimensional structural features into aerodynamic parameter vectors through a nonlinear mapping function. The icing conductor galloping characteristic prediction module couples aerodynamic parameter vectors with conductor structure dynamic parameters to construct a nonlinear coupled vibration system. Based on time-domain integration and multimodal response analysis, it obtains the galloping response characteristics under different icing morphologies and environmental conditions.

6. A computer storage medium having a computer program stored thereon, characterized in that, When the computer executes the computer program, it implements the image-driven rapid prediction method for the galloping characteristics of icy conductors as described in any one of claims 1-4.

7. An electronic device comprising a memory and a processor, characterized in that, The memory stores a computer program, and the processor runs the computer program stored in the memory. When the computer program is executed, it implements the image-driven rapid prediction method for the galloping characteristics of icy conductors as described in any one of claims 1-4.

Images