Method for warning safety distance of ultra-high voltage line by combining Beidou positioning with AI image

Abstract

The application belongs to the technical field of line safety, and discloses a Beidou positioning fusion AI image ultra-high voltage line safety distance early warning method, which comprises the following steps: collecting Beidou positioning data, multi-view image data and micro-meteorological data of a target conductor, and performing preprocessing and time synchronization to obtain a Beidou point set, a synchronous image set and an environmental vector; constructing a state field generator, analyzing the conductor motion state, and extracting a physical state component; constructing a state vector, and estimating the optimal state vector at the current time through a state space; according to the optimal state vector at the current time and the future environmental vector, combining the state field generator to analyze and obtain the real-time three-dimensional trajectory and the future predicted trajectory of the conductor full span; further analyzing and obtaining the dynamic minimum spatial clearance between the conductor and the preset crossing object; and generating graded safety warning information based on the dynamic minimum spatial clearance; the application effectively avoids the phase-to-phase short circuit and the cross-regional power grid splitting risk caused by conductor galloping.

Description

Technical Field

[0001] This application relates to the field of line safety technology, and more specifically, to a method for early warning of safe distances for ultra-high voltage lines based on BeiDou positioning fusion with AI images. Background Technology

[0002] Under the influence of wind loads, icing, and other environmental factors, ultra-high voltage (UHV) transmission lines exhibit complex nonlinear galloping of conductors. Their three-dimensional motion trajectories possess characteristics such as spatial continuity, temporal variation, and modal coupling. Existing monitoring systems typically employ two technical solutions: one is based on BeiDou single-point positioning, deploying discrete monitoring terminals along the conductor to acquire displacement sequences of a limited number of points; the other is based on multi-view two-dimensional video analysis, attempting to reconstruct the spatial morphology of the conductor through image matching and three-dimensional reconstruction.

[0003] However, the aforementioned BeiDou scheme can only output the coordinates of a few discrete points and cannot obtain continuous spatial morphology. This means that the calculation of the crossing distance must rely on interpolation or empirical formulas, and the interpolation error can reach several meters when the amplitude of conductor galloping is large. Although the multi-view video scheme can perceive continuous space, its three-dimensional reconstruction calculation is highly complex and it is difficult to achieve real-time processing under conditions of changing lighting, severe weather, and high dynamic motion. It can usually only generate sparse point clouds offline. This means that the calculation of the dynamic spatial clearance between the conductor and the crossing object relies heavily on incomplete spatial information. The accumulation of errors leads to the distortion of clearance judgment. When the amplitude of galloping is large, the actual clearance may have exceeded the red line of safety regulations, and the system may not be able to give a timely warning. This can cause insulation flashover and conductor wear, or even lead to phase-to-phase short circuits or cascading trips, ultimately resulting in cross-regional power grid disconnection accidents.

[0004] In view of this, this application proposes a method for early warning of safe distances for ultra-high voltage power lines based on BeiDou positioning and AI imagery to solve the above problems. Summary of the Invention

[0005] To address the shortcomings of existing technologies, this application aims to provide a method for early warning of safe distances for ultra-high voltage (UHV) transmission lines based on BeiDou positioning and AI imagery. By constructing a state field generator composed of a physical foundation module and a neural residual module, it deeply integrates BeiDou high-precision positioning, multi-view vision, and micro-meteorological data under a unified spatiotemporal reference. Using fixed physical foundation modules to provide mechanical constraints and dynamic neural conditional state adaptation nonlinear residuals, combined with state-space estimation methods, it online inverses the optimal state vector of the conductor. This improves the accuracy of real-time reconstruction and short-term prediction of the continuous three-dimensional galloping trajectory of UHV conductors across the entire span under wind load and icing conditions, and reduces the error in calculating the crossing distance. Simultaneously, it generates tiered safety warnings, solving the problem of misjudgment of crossing distances caused by missing three-dimensional trajectories, and effectively avoiding the risks of phase-to-phase short circuits and cross-regional power grid disconnection caused by extreme galloping.

[0006] This application provides the following technical solution: a method for early warning of safe distances for ultra-high voltage power lines based on BeiDou positioning fusion with AI images, including:

[0007] Collect BeiDou positioning data, multi-view image data, and micro-meteorological data of the conductors of the target UHV line, and perform preprocessing and time synchronization to obtain BeiDou point set, synchronized image set, and environmental vector;

[0008] A state field generator is constructed to analyze and obtain the motion state of the conductor based on the BeiDou point set, the synchronous image set, and the environmental vector, and to extract the physical state components.

[0009] A state vector is constructed based on the physical state components and the conditional state components. The optimal state vector at the current moment is obtained by combining the state space estimation of the synchronous image set and the state field generator.

[0010] Based on the current optimal state vector and the future environment vector, the real-time three-dimensional trajectory of the conductor across the entire span and the predicted future trajectory of the conductor are obtained by combining the state field generator analysis.

[0011] Based on the real-time three-dimensional trajectory and the future predicted trajectory, the dynamic minimum spatial clearance between the conductor and the preset crossing object is obtained; based on the dynamic minimum spatial clearance, hierarchical safety early warning information is generated.

[0012] The technical effects and advantages of the ultra-high voltage line safety distance early warning method based on BeiDou positioning and AI image fusion in this application are as follows:

[0013] 1. By preprocessing and synchronizing BeiDou positioning data, multi-view image data, and micro-meteorological data, BeiDou point sets, synchronized image sets, and environmental vectors are obtained. This enables the alignment of the three heterogeneous data sources under a unified spatiotemporal reference, solving the problem of subsequent judgment distortion caused by data misalignment and time deviation in existing technologies, thereby improving the input consistency of multi-source information fusion.

[0014] 2. By constructing a state field generator consisting of a physical foundation module and a neural residual module, the motion state of the conductor is analyzed and physical state components are extracted based on the BeiDou point set, synchronous image set and environmental vector. This combines the mechanical prior knowledge of the physical model with the data-driven capability of the neural network, solving the problem of insufficient modeling of nonlinear galloping of conductors in existing technologies, thus making the analysis results of conductor motion state closer to the actual dynamic characteristics.

[0015] 3. By constructing a state vector from physical state components and conditional state components, and combining it with a synchronized image set and a state field generator, the state space estimation method is used to obtain the optimal state vector at the current moment. This enables probabilistic fusion of the predicted morphology with the BeiDou point set and visual 3D point cloud observations, reducing dependence on a single sensor or interpolation result, thereby reducing the spatial error of trajectory reconstruction and making subsequent trajectory outputs have higher spatial continuity.

[0016] 4. By acquiring future micro-meteorological data through networking, a future environmental vector is obtained. Based on the current optimal state vector and the future environmental vector, combined with a state field generator, the real-time three-dimensional trajectory and future predicted trajectory of the conductor span are obtained. This can restore the abstract low-dimensional state to a continuous, high-dimensional full-field trajectory, and at the same time realize the advance prediction of the future short-term trajectory. This solves the problem that existing technologies cannot predict risks due to a lack of predictive capabilities, thus providing a complete spatial data foundation for dynamic clearance calculation.

[0017] 5. By analyzing the dynamic minimum spatial clearance between the conductor and the preset crossing object based on the real-time three-dimensional trajectory and the future predicted trajectory, and generating graded safety early warning information based on the clearance, it is possible to quantitatively compare the three-dimensional spatial distance with the statutory safety threshold, avoid the misjudgment of clearance caused by the lack of trajectory in the existing technology, and thus output early warning according to the level when the clearance is close to the danger value, reducing the risk of phase-to-phase short circuit and cross-regional power grid disconnection caused by extreme galloping.

[0018] In summary, this application not only improves the local defects of the existing technology in the aspects of conductor full span trajectory reconstruction, multi-source data fusion and dynamic clearance calculation, but also makes data acquisition, state estimation, trajectory prediction and early warning output form a closed processing chain, which improves the accuracy, real-time and reliability of UHV line safety distance early warning, and is more suitable for active safety protection of long span transmission lines under complex meteorological conditions. Attached Figure Description

[0019] Figure 1 This is a schematic diagram of the method for early warning of safe distance of ultra-high voltage lines based on Beidou positioning and AI image fusion in this application;

[0020] Figure 2 This is a schematic diagram of the execution flow of the physical foundation module of this application;

[0021] Figure 3 This is a schematic diagram of the method for obtaining the optimal state vector at the current moment in this application;

[0022] Figure 4 This is a schematic diagram of the method for obtaining the dynamic minimum spatial clearance between the conductor and the preset intersecting object according to this application. Detailed Implementation

[0023] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0024] Example 1:

[0025] Please see Figure 1 As shown, this embodiment provides a method for early warning of safe distances for ultra-high voltage power lines based on BeiDou positioning fusion with AI images, including:

[0026] The system collects BeiDou positioning data, multi-view image data, and micro-meteorological data of the conductors of the target UHV line, performs preprocessing and time synchronization, and obtains BeiDou point sets, synchronized image sets, and environmental vectors.

[0027] In one specific implementation, the BeiDou positioning data consists of a three-dimensional point coordinate sequence obtained by installing M BeiDou high-precision monitoring terminals at equal intervals on the conductor of the target UHV line; the multi-view image data consists of a multi-view video stream covering the entire span of the conductor obtained by installing R industrial cameras with synchronous triggering function on adjacent towers; and the micro-meteorological data consists of wind speed, wind direction, temperature, and humidity obtained by installing micro-meteorological sensors on the top of the towers. The three-dimensional point coordinate sequence and micro-meteorological data are cleaned; distortion correction is performed on the video stream; and all data are synchronized in time using a BeiDou precision clock signal, and the spatial coordinates calculated by BeiDou are unified to a linear timescale. In the local coordinate system of the line with the center of the road as the origin, the seven-parameter Bursa model transformation is a well-known technique and will not be elaborated further. The BeiDou point set and the synchronized image set are obtained, and then the synchronized micro-meteorological data are packaged into an environmental vector. The environmental vector is obtained by splicing wind speed, wind direction, temperature and humidity. The unified spatiotemporal reference is the premise of multi-source fusion. Only when all data are ensured to point to the same section of the conductor at the same time and in the same coordinate system can the subsequent joint observation have physical meaning. This step eliminates the systematic error caused by sensor heterogeneity, lays the spatiotemporal consistency foundation for subsequent high-precision state estimation, and directly supports the realization of multi-source information complementary gain.

[0028] A state field generator is constructed. Based on the BeiDou point set, the synchronous image set, and the environmental vector, the motion state of the conductor is analyzed and the physical state components are obtained.

[0029] In one specific implementation, the state field generator consists of a physical foundation module and a neural residual module, wherein the physical foundation module is used to calculate the ground state motion response of the conductor.

[0030] Reference Figure 2 The execution of the physical foundation module includes:

[0031] Based on real-time environmental vector analysis of conductor unit length self-weight load, unit length icing load, and unit length wind load, a comprehensive unit length load is obtained through vector synthesis. Specifically, the conductor bare wire diameter D, conductor cross-sectional area, and conductor unit length mass are obtained from line design data, conductor model parameter tables, or on-site verification measurement data. The conductor unit length self-weight load is obtained by calculating the product of conductor unit length mass and gravitational acceleration. The estimated equivalent icing thickness b at the previous moment is obtained based on the icing thickness state component output from the previous filtering cycle, the change in conductor unit length mass, or the change in conductor sag. When b is obtained from the conditional state component output from the previous filtering cycle, a one-dimensional icing thickness state component is set in the conditional state component, and the output value of the icing thickness state component is used as b. When b is obtained from the change in conductor unit length mass, the change in conductor unit length mass is recorded as Δm, and based on... b is calculated; when b is obtained from the change in conductor sag, under the condition that the current wind speed, temperature, span and tension boundary conditions remain unchanged, b is searched in one dimension with the goal of minimizing the difference between the theoretical catenary sag and the observed sag, and the b that minimizes the sag difference is obtained.

[0032] Within the initial filtering period, the estimated equivalent icing thickness *b* from the previous moment is determined based on manual inspection records, icing monitoring sensor outputs, line maintenance system records, or historical icing records for the same section, and is used as the initial parameter for state-space estimation; *D+2b* is taken as the total diameter of the conductor after icing, and the icing cross-sectional area per unit length is... The product of ice density and cross-sectional area of ​​ice accretion per unit length is calculated to obtain the mass of ice accretion per unit length. The product of the mass of ice accretion per unit length and gravitational acceleration is then calculated to obtain the ice load per unit length. Ice density is determined based on the type of ice accretion, field sampling results, or records from the line maintenance system. When field sampling results are unavailable, the ice density is determined using empirical values ​​corresponding to the current temperature, humidity, and type of ice accretion. The wind load per unit length is determined based on wind speed, wind direction, the total diameter of the conductor after icing, and the wind pressure coefficient. In one possible implementation, the wind load per unit length is calculated as follows: based on wind speed... The angle between the wind direction and the axis of the conductor Calculate the reference wind pressure When the reference wind pressure is expressed in engineering units kN / m 2 When expressed, the reference wind pressure When the reference wind pressure is expressed in SI units N / m 2 When expressed, the reference wind pressure ,in, The air density is used as the basis; the wind pressure coefficient is obtained by multiplying the wind pressure non-uniformity coefficient and the shape coefficient; the wind pressure non-uniformity coefficient is determined based on the spatial distribution difference of wind speed within the target span, the line height, and the terrain type; the shape coefficient is determined based on the conductor cross-sectional shape, the icing cross-section type, and the wind direction angle. The wind pressure non-uniformity coefficient and the shape coefficient are obtained by matching them from the transmission line design specifications or line design data; the wind load per unit length is obtained by multiplying the reference wind pressure, the total diameter of the conductor after icing, and the wind pressure coefficient; in this embodiment, N and m are used as the calculation units when performing load vector synthesis; when the reference wind pressure q is expressed in engineering units kN / m 2 When expressing the load, first multiply q by 1000 to convert it to N / m², and then calculate the wind load per unit length. The self-weight load per unit length of the conductor and the icing load per unit length act in the vertical direction, and the wind load per unit length acts in the horizontal wind pressure direction perpendicular to the conductor axis. The self-weight load per unit length of the conductor, the icing load per unit length, and the wind load per unit length are vectorized according to direction to obtain the comprehensive load per unit length. The sum of the self-weight load per unit length of the conductor and the icing load per unit length is taken as the vertical load per unit length for the calculation of the vertical catenary.

[0033] Based on the unit length vertical load, span, elevation difference between the two suspension points, and conductor arc length constraint or reference horizontal tension in the comprehensive unit length load, and combined with the catenary equation, the static equilibrium position curve of the conductor is calculated, and the equivalent horizontal tension under the current working condition is simultaneously calculated; specifically, with the lower suspension point A as the origin, the span direction as the x-axis, and the vertical direction as the z-axis, the coordinates of the other suspension point B are... The vertical load per unit length is denoted as... Let the equivalent horizontal tension be denoted as H, and let... The static equilibrium position curve is represented as follows: ;in, The range of values ​​is The above catenary equation is used to describe the static equilibrium position curve of the conductor under a current unit length vertical load, the equivalent horizontal tension H, and the horizontal offset. The elevation constraint of suspension point B is determined jointly by the elevation constraint and the arc length constraint of the conductor; the elevation constraint of suspension point B is: The conductor arc length constraint is that the conductor curve length within the span is equal to the conductor arc length S. The conductor arc length S is obtained from the line design data, the benchmark static calibration results, or the conductor arc length under windless and ice-free conditions. When the line design data has given the reference horizontal tension corresponding to the current working condition, the reference horizontal tension is used as the initial or known value of H, and the static equilibrium position curve is calculated by the catenary equation.

[0034] To solve for the equivalent horizontal tension H and the horizontal offset A system of nonlinear equations is established, consisting of the elevation constraint at suspension point B and the arc length constraint of the conductor; the elevation constraint at suspension point B is expressed as... The arc length constraint of the conductor is expressed as: Where S is the conductor arc length corresponding to the current span; the equivalent horizontal tension H and the horizontal offset are combined. As the variable to be solved, As the first residual, As the second residual, a two-dimensional residual function is constructed; the two-dimensional residual function is solved by the Newton-Raphson iteration method, the secant iteration method, or the least squares iteration method, so that the first residual and the second residual simultaneously meet the convergence requirements; the initial values ​​of the iteration are determined according to the line design horizontal tension, the windless and ice-free reference tension, the design sag, or the equivalent horizontal tension output from the previous filtering cycle.

[0035] The iteration termination conditions include: the ratio of the difference in equivalent horizontal tension obtained from two adjacent iterations to the current estimated equivalent horizontal tension is not greater than the relative tension convergence threshold; the difference in horizontal offset obtained from two adjacent iterations is not greater than the offset convergence threshold; the difference between the theoretical height and the actual height of suspension point B is not greater than the height residual threshold; and the difference between the calculated conductor curve length and the conductor arc length S is not greater than the arc length residual threshold. The relative tension convergence threshold is obtained by conversion based on the tension sensor error or the allowable tension error in the line design; the offset convergence threshold is determined based on the conductor spatial offset step length; the height residual threshold is determined based on the larger of the BeiDou positioning elevation error and the camera 3D reconstruction elevation error; and the arc length residual threshold is determined based on the line design arc length error or the conductor arc length calibration error.

[0036] Once the iteration converges, the final equivalent horizontal tension H and horizontal offset will be obtained. Substituting into the catenary equation, the static equilibrium position curve of the conductor is obtained; the static equilibrium position curve of the conductor is used to represent the static vertical coordinate z(x) of the conductor at any horizontal position x within the span; in order to connect with the subsequent modal superposition and visual point cloud matching steps, the span interval [0, L] is divided into N discrete conductor positions according to the preset spatial step size, and the static vertical coordinates corresponding to each discrete conductor position are calculated to form a discretized static equilibrium position sequence; the preset spatial step size is determined according to the spacing between Beidou monitoring terminals, the spatial resolution of multi-view image reconstruction and the minimum half wavelength corresponding to the highest order of the target modality, and the preset spatial step size is not greater than the minimum value of the above three.

[0037] The unit mass of the ice-covered conductor is calculated based on the conductor's mass per unit length and the mass of ice accretion per unit length. Specifically, the conductor's mass per unit length is added to the ice accretion per unit length to obtain the total unit mass of the ice-covered conductor. The total unit mass of the ice-covered conductor is used for subsequent natural frequency vector calculations.

[0038] Calculate the natural frequency vector of the conductor based on its total mass per unit area and horizontal tension; specifically, for a conductor fixed at both ends, calculate the natural frequency vector based on its order. Pi Gear spacing Equivalent horizontal tension and the total mass per unit of ice-containing conductor The i-th order angular frequency is calculated; in one possible implementation, the formula for calculating the i-th order angular frequency is: The ratio of the i-th angular frequency to 2π is used to obtain the i-th transverse natural frequency. Typically, the first K natural frequencies are calculated, where K is determined based on the spectral analysis of the galloping energy distribution. The results are arranged by order to form a natural frequency vector, which is used to characterize the basic dynamic characteristics of the conductor under specific icing and tension conditions.

[0039] The modal superposition method is employed to analyze the linear vibration response of the conductor based on the natural frequency vector. The conductor vibration is decomposed into the superposition of the first K natural modes, yielding the linear dynamic displacement field, the foundation dynamic velocity field, and the reference amplitude corresponding to each mode. Specifically, for a conductor fixed at both ends, the i-th mode shape function is taken as sin(iπx / L), where x is the spatial coordinate along the conductor span direction, and L is the span. At any horizontal position x, the product of the first K mode shape functions and the corresponding generalized coordinates is... The linear dynamic displacement field is obtained by summing the products of the first K mode shape functions and their corresponding generalized velocity coordinates. The basic dynamic velocity field is obtained by summing the products of the first K mode shape functions and their corresponding generalized velocity coordinates. The generalized coordinates are the displacement components in the state vector corresponding to the i-th mode, and the generalized velocity coordinates are the velocity components in the state vector corresponding to the i-th mode, or the derivatives of the generalized coordinates with respect to time. The reference amplitudes corresponding to each mode are calculated based on the subsequent quasi-steady aerodynamic theory and are used to limit the initial vibration magnitude of each mode under the assumption of linear, steady-state harmonic excitation.

[0040] According to the quasi-steady aerodynamic theory, the amplitude of the lateral lift acting on a unit length of conductor is determined based on air density, relative wind speed, the total diameter of the conductor after icing, and the lift coefficient; specifically, based on air density... The relative wind speed at the nth discrete traverse position Total diameter of the conductor after icing Angle of attack at the position of the nth discrete traverse Harmony and angle of attack Corresponding lift coefficient The lateral lift amplitude acting on a unit length of conductor at the nth discrete conductor position is calculated; in one possible implementation, the lateral lift amplitude acting on a unit length of conductor at the nth discrete conductor position... The calculation formula is: Among them, relative wind speed The wind speed and direction were calculated from the micrometeorological data, along with the velocity of the conductor at the nth discrete conductor position. When the influence of the conductor's velocity on the incoming flow velocity is ignored, the projected velocity of the wind speed in the plane perpendicular to the conductor's axis is taken as... Angle of attack The angle of attack is determined based on wind direction, conductor axial direction, conductor vibration velocity direction, and reference direction of the icing section. In simplified calculations, the angle between the projection of the wind direction onto the plane perpendicular to the conductor axial direction and the reference direction of the icing section is used. .

[0041] Lift coefficient Based on the type of icing cross-section and angle of attack The lift coefficient is obtained by matching from a pre-established lookup table; the lift coefficient lookup table is established from wind tunnel test data, publicly available engineering test curves, or historical calibration data of similar lines; when the angle of attack is missing... When the sampling points are completely corresponding, linear interpolation is performed on the lift coefficients corresponding to adjacent angle-of-attack sampling points to obtain... When the i-th mode is the current dominant mode, based on the lateral lift amplitude acting on a unit length of conductor at the n-th discrete conductor position, the... Angular frequency and the first The instantaneous lateral lift at the nth discrete conductor position is obtained by calculating the initial phase parameters of the first mode. In one embodiment, the lateral lift at the nth discrete conductor position acts on a unit length of conductor. The calculation formula is: ,in, For the first angular frequency, For the first Initial phase parameters of the first mode; initial phase parameters Determined based on the generalized displacement and generalized velocity coordinates at the start of the current filtering cycle; when initial phase observations are lacking, It is set to 0 and then updated and corrected through observations during subsequent state-space estimation.

[0042] The lateral lift acting on a unit length of conductor is integrated along the span direction using mode shape weighted integrals to obtain the generalized aerodynamic amplitude of the corresponding mode. Specifically, for the i-th mode, the lateral lift at each discrete conductor position is multiplied by the function value of the i-th mode shape function at the corresponding position, and then integrated or discretely summed along the span direction to obtain the generalized aerodynamic amplitude corresponding to the i-th mode. When the lateral lift is approximately uniform along the span direction and the two ends of the conductor are fixed, the mode shape function in the form of sin(iπx / L) can be used for analytical integration. When the wind field, icing thickness, or end elevation is not uniform along the span direction, numerical integration is performed based on the actual value of the lateral lift at each discrete conductor position, and the simplified result of the uniformly distributed lateral aerodynamic force is no longer used.

[0043] The i-th mode of the conductor can be simplified as a single-degree-of-freedom forced vibration system, and the reference amplitude corresponding to the i-th mode can be calculated based on the generalized aerodynamic amplitude, modal mass, angular frequency, and damping ratio. Specifically, the i-th modal mass is obtained by integrating the total mass per unit mass of the ice-covered conductor with the square of the i-th modal shape function along the span direction. In one possible implementation, the i-th modal shape function is used... Calculate the i-th modal mass. The calculation formula is: According to the i-th modal mass With the i-th order angular frequency Calculate the i-th modal stiffness i-th modal stiffness The calculation formula is: ;

[0044] Based on the amplitude of the lateral lift force acting on a unit length of conductor at the position of the nth discrete conductor. and the i-th order mode shape function Calculate the amplitude of the i-th order generalized aerodynamic force The amplitude of the i-th order generalized aerodynamic force The calculation formula is: ;

[0045] Let the i-th order damping ratio be denoted as . ,:

[0046] When using discrete traverse position calculation, based on the nth discrete traverse position... Spatial step size between adjacent discrete conductor positions and the lateral lift amplitude at the nth discrete conductor position Calculate the amplitude of the i-th order generalized aerodynamic force The amplitude of the i-th order generalized aerodynamic force The calculation formula is: ;

[0047] When the excitation angular frequency When the resonance criterion for the i-th modal is not met, the reference amplitude corresponding to the i-th modal is... Determined based on the frequency response relationship of forced vibration under single-degree-of-freedom conditions, specifically according to the stiffness of the i-th modal. , i-th modal mass Excitation angular frequency The i-th order damping ratio , i-th order angular frequency Calculate the reference amplitude corresponding to the i-th order mode using the generalized aerodynamic amplitude of the i-th order. The reference amplitude corresponding to the i-th mode The calculation formula is: When the excitation angular frequency When the resonance criterion of the i-th modal is met, the reference amplitude corresponding to the i-th mode is calculated using the resonance approximation. Specifically, based on the i-th modal stiffness The i-th order damping ratio Calculate the reference amplitude corresponding to the i-th order mode using the generalized aerodynamic amplitude of the i-th order. The reference amplitude corresponding to the i-th mode The calculation formula is: Wherein, the i-th order damping ratio The i-th order modal resonance criterion is determined based on historical vibration attenuation tests of similar lines, online vibration attenuation curve fitting results, or line design damping parameters; the excitation angular frequency is used to determine the i-th order modal resonance criterion. With the i-th order angular frequency The ratio of the absolute value of the difference to the i-th order angular frequency is less than or equal to the i-th order resonance determination threshold, which is determined based on the spectral resolution of the i-th order damping ratio and the lateral lift time series; specifically, the angular frequency interval corresponding to the spectral resolution is denoted as . ,in , The analysis duration for the lateral lift time series was determined based on the dominant period of conductor galloping and the sampling period of micrometeorological data; specifically, Take spectral resolution The minimum analysis time required is no greater than half the minimum interval between two adjacent angular frequencies; the damping ratio of the i-th order will be determined by... The larger value in the range is used as the threshold for determining the i-th order resonance; when the excitation angular frequency is not calculated separately. At that time, Take as the i-th order angular frequency The resonance approximate reference amplitude is used as the conservative reference amplitude; reference amplitude The initial vibration magnitude of the i-th mode of the conductor is used to characterize the current wind field, icing, and damping conditions.

[0048] The calculated first K order reference amplitudes Ai are arranged according to their order to form a reference amplitude vector, which is used together with the base displacement field, base dynamic velocity field, displacement gradient field, equivalent horizontal tension and natural frequency vector to form a physical characteristic vector.

[0049] The basic displacement field is obtained by superimposing the static equilibrium position curve and the linear dynamic displacement field.

[0050] Spatial numerical differentiation of the basic displacement field yields the displacement gradient field.

[0051] The physical characteristic vector is obtained by splicing together the basic displacement field, basic dynamic velocity field, displacement gradient field, equivalent horizontal tension, natural frequency vector, and reference amplitudes of each order.

[0052] The physical feature vector is used as input to the neural residual module to obtain the corrected conductor displacement and velocity. The neural residual module can be constructed using a multi-layer fully connected neural network, such as including an input layer, three hidden layers, and an output layer. The number of nodes in the input layer is equal to the dimension of the physical feature vector, and the number of nodes in the output layer is the total dimension of the three-dimensional displacement and velocity of the discrete sampling points of the conductor. Each hidden layer is followed by a batch normalization layer and a ReLU activation function. The residual connection is set to directly add the input physical feature vector after linear transformation to the output layer to learn the nonlinear residual mapping between the physical model prediction and the true value.

[0053] Training methods for neural residual modules include:

[0054] Q sets of residual training data are pre-collected. The residual training data includes physical feature vectors and corresponding measured conductor displacement and velocity, or high-precision simulated conductor displacement and velocity. The measured conductor displacement and velocity are obtained by joint calibration of Beidou monitoring terminal, multi-view 3D reconstruction results, or vibration sensor. The high-precision simulated conductor displacement and velocity are calculated by finite element dynamic model under the same wind speed, wind direction, temperature, icing thickness, span, and tension conditions. The method for setting Q is as follows: first, a working condition combination sample set is formed according to the value range of wind speed, wind direction, temperature, icing thickness, span, and tension. After each round of sample expansion, the neural residual module is retrained using the newly added samples and the mean square error of the validation set is calculated. When the relative decrease rate of the mean square error of the validation set is less than the sample expansion benefit threshold after two consecutive sample expansions, the addition of training samples is stopped, and the current number of samples is determined as Q. The sample expansion benefit threshold is determined according to the lower limit of the measurement error of the validation set label.

[0055] The physical feature vector is used as the input to the neural residual module, which outputs the corresponding corrected conductor displacement and velocity. The goal is to minimize the error between the corrected conductor displacement and velocity and the actual conductor displacement and velocity or the high-precision simulated conductor displacement and velocity. Backpropagation and gradient descent optimization algorithms are used to iteratively optimize the network parameters of the neural residual module until it converges. The convergence criteria are: the relative decrease rate of the validation set mean square error in P consecutive training rounds is less than the error decrease threshold, and the relative change in network parameters is less than the parameter change threshold. The error decrease threshold is determined based on the lower limit of the measurement error of the validation set labels, and the parameter change threshold is determined based on the minimum effective update amplitude after normalization of the network parameters. The number of consecutive rounds P is determined based on the short-term fluctuation width of the validation error curve. The network parameters corresponding to the convergence of the neural residual module are obtained, and the neural residual module constructed using these parameters is used as the trained neural residual module.

[0056] Physical state components are extracted from the corrected conductor displacement and velocity. Specifically, the corrected conductor displacement field and the corrected conductor velocity field are spatially discretized along the conductor span direction to obtain the three-dimensional displacement vector and three-dimensional velocity vector of each discrete point. The spatially discretized sampling points are selected at equal intervals along the conductor span direction, and the number of sampling points is not less than 2K+1, or the spatial sampling interval is not greater than L / (2K), to ensure that spatial aliasing does not occur during modal coordinate extraction. Using the pre-stored first K order modal mode functions, modal projection is performed on the corrected conductor displacement field to obtain the generalized displacement coordinates of each order. Modal projection is performed on the corrected conductor velocity field to obtain the generalized velocity coordinates of each order. The i-th order generalized displacement coordinate is: the product of the projection of the corrected conductor displacement field in the target vibration direction and the i-th order modal mode function is integrated along the span direction, and then divided by the integral of the square of the i-th order modal mode function along the span direction. In one possible implementation, the i-th order generalized displacement coordinate is... The calculation formula is: ,in, The projection of the corrected conductor displacement field onto the target vibration direction; the i-th order generalized velocity coordinate is: the product of the projection of the corrected conductor velocity field onto the target vibration direction and the i-th order mode shape function is integrated along the span direction, and then divided by the integral of the square of the i-th order mode shape function along the span direction. In one possible implementation, the i-th order generalized velocity coordinate is... The calculation formula is: ,in, The projection of the corrected conductor velocity field onto the target vibration direction is used; the generalized displacement coordinates and generalized velocity coordinates of each order are used together as physical state components.

[0057] get and The methods include:

[0058] The position is obtained from the static equilibrium position curve of the conductor. Tangential unit vector at point The wind direction unit vector in the micro-meteorological data The unit vector perpendicular to the tangential direction of the conductor Projecting the vector onto the plane and normalizing it, we obtain the unit vector of the target vibration direction. ,Right now: ;when When the value is less than the direction determination threshold, it indicates that the wind direction is approximately consistent with the tangential direction of the conductor. In this case, the vertical unit vector ez is positioned perpendicular to the tangential unit vector of the conductor. Projecting and normalizing the vector onto the plane yields the unit vector of the target vibration direction. ,Right now: The direction determination threshold is determined based on the angular resolution of the wind direction sensor, and the sine value corresponding to the angular resolution of the wind direction sensor is used as the direction determination threshold.

[0059] The corrected conductor displacement field is placed at position ,time The dot product of the three-dimensional displacement vector at the point of origin and the unit vector of the target vibration direction is used to obtain the projection of the corrected conductor displacement field onto the target vibration direction. The corrected conductor displacement field will be applied at position. ,time The dot product of the three-dimensional velocity vector at the point of origin and the unit vector of the target vibration direction is used to obtain the projection of the corrected conductor velocity field onto the target vibration direction. .

[0060] A state vector is constructed based on the physical state components and the conditional state components. The state vector is then combined with a synchronized image set and a state field generator to perform state space estimation, thereby obtaining the optimal state vector at the current moment.

[0061] In one specific implementation, refer to Figure 3 Methods for obtaining the optimal state vector at the current moment include:

[0062] A state vector is constructed based on the physical state components and the conditional state components. The physical and conditional state components are concatenated to obtain the state vector. The conditional state components are conditional encoding vectors used to adjust the output behavior of the neural residual module, representing the nonlinear residual characteristics not directly modeled by the physical foundation module in the current observations. The conditional state components are used as variables to be estimated during state space estimation, and are jointly updated with the physical state components through unscented Kalman filtering, with initial values ​​set to zero vectors. The dimension of the conditional state components is not less than the number of residual principal components used to represent the unmodeled residuals. Specifically, the residual matrix of the training samples is decomposed into principal components, sorted by eigenvalues ​​from largest to smallest, and the variance contribution rate corresponding to each principal component is accumulated sequentially. The conditional state components are considered valid when the accumulated variance contribution rate is not lower than the residual retention threshold. When the number of principal components is used as the number of residual principal components, the residual retention threshold is determined based on the residual reconstruction error. Specifically, the training sample residuals are reconstructed using the aforementioned principal components. When the mean square error of the reconstructed residuals is not greater than the lower limit of the measurement error corresponding to the BeiDou positioning error, multi-view 3D reconstruction error, and vibration velocity measurement error, the corresponding cumulative variance contribution rate is determined as the residual retention threshold. The state vector of the previous moment and the environment vector of the current moment are used as inputs to the state field generator to obtain the predicted conductor shape at the current moment. Specifically, the physical state components are physically transferred according to the state transition matrix and the first process noise. For the i-th mode, the generalized displacement coordinates and generalized velocity coordinates are used to form a two-dimensional modal state sub-vector. The time interval between adjacent filtering cycles is denoted as... The i-th order angular frequency is denoted as Then the state transition relationship of the i-th modal state subvector is determined by the analytical solution of simple harmonic motion: the generalized displacement coordinates are according to and Weighted update, generalized velocity coordinates according to and Weighted update; the two-dimensional state transition sub-matrices corresponding to the first K modes are arranged into block diagonal matrices according to their order to obtain the state transition matrices corresponding to the physical state components: the transition of the physical state components is based on the analytical solution of the conductor simple harmonic motion model, that is, the generalized coordinates of each order are phase-advanced according to the natural frequency; the conditional state components use the identity matrix as the transition matrix; the first process noise covariance matrix is ​​constructed based on the variance of the historical generalized coordinate prediction residual sample, and the second process noise covariance matrix is ​​constructed based on the variance of the difference sequence of the conditional state components of adjacent filtering periods; the prior state estimate at the current moment is obtained based on the state transition results, and the conductor spatial shape corresponding to the prior state estimate is used as the predicted conductor shape at the current moment.

[0063] Based on the predicted conductor morphology, a visual 3D point cloud of the conductor is extracted from the synchronized image set. Specifically, for the h-th industrial camera, the predicted conductor morphology at the current moment is projected onto the 2D image plane of the h-th industrial camera using the calibrated camera projection matrix. All projection points on the 2D image plane are connected to form the predicted conductor pixel trajectory. Using the predicted pixel trajectory as the center line, a preset width is extended to both sides to generate a pixel-level strip region, which is marked as the ROI region. The preset width is set as follows: for the h-th industrial camera, the camera reprojection error, state prediction position error, and conductor radius projection error are converted into corresponding pixel errors. The camera reprojection error directly uses the pixel reprojection error obtained during camera calibration. The state prediction position error and conductor radius error are first expressed as the 3D error radius in the local coordinate system of the line, and then projected onto the 2D image plane using the camera projection matrix of the h-th industrial camera to obtain the corresponding pixels. Error radius: The half-width of the ROI region is set to an integer pixel not less than the sum of the pixel reprojection error, the state prediction position error, and the pixel error radius. The state prediction position error is obtained statistically based on the residual between the predicted trajectory and the actual observed trajectory in the previous filtering cycle. Traverse feature extraction, matching, and 3D reconstruction are performed within the ROI region. Considering the sparse texture of the traverse surface, edge feature-based detection methods are preferred to output a visual 3D point cloud. The edge number threshold is determined based on the length of the predicted pixel trajectory of the traverse and the edge sampling interval. The edge number threshold is an integer not less than the ratio of the length of the predicted pixel trajectory of the traverse to the edge sampling interval, where the edge sampling interval is determined based on the image resolution and the pixel width of the traverse. The matching confidence threshold is determined based on the confidence distribution of historical correct matching samples. Specifically, the mean confidence of historical correct matching samples is taken minus three standard deviations, or the lower quantile of the confidence distribution of historical correct matching samples is taken as the matching confidence threshold.

[0064] Within the ROI region, feature detection algorithms, such as feature detection algorithms, feature description algorithms, and nearest neighbor matching algorithms, can be used to detect feature points on the conductor surface, such as spacers, vibration dampers, and texture spots. Subsequently, the corresponding feature description algorithm is used to calculate a feature vector for each feature point. Based on the similarity of descriptors, such as calculating Euclidean distance or Hamming distance, feature matching algorithms, such as nearest neighbor matching, are used to match feature points belonging to the same physical point in different camera images, forming a set of matching pairs. For each pair of matching points in the set of matching pairs, the triangulation principle in multi-view geometry is used to calculate its corresponding three-dimensional spatial coordinates by solving linear equations or optimizing reprojection errors, thus obtaining a visual three-dimensional point cloud.

[0065] The visual 3D point cloud is transformed from the camera coordinate system to the local coordinate system of the line, and the transformation parameters are obtained through pre-calibrated camera extrinsic parameters. Then, the BeiDou point set and the transformed visual 3D point cloud are stitched together in the world coordinate system to form a joint observation vector Z. Before stitching, outliers in the visual point cloud with reprojection errors greater than a first preset threshold are removed, and points in the BeiDou point set with estimated positioning errors greater than a second preset threshold are removed. If the BeiDou device outputs a positioning confidence score, points with positioning confidence scores lower than the positioning confidence score threshold are removed. The first preset threshold is set based on the statistical distribution of camera calibration reprojection errors. The second preset threshold is set based on the nominal positioning accuracy calculated by BeiDou RTK to cover the upper limit of errors caused by multipath effects and ionospheric anomalies. The positioning confidence score threshold is determined based on the confidence distribution of the effective positioning samples output by the BeiDou device.

[0066] Using BeiDou point sets and visual 3D point clouds as observations, and based on a state-space estimation method, the observed and predicted values ​​are fused to update and output the optimal state vector for the current moment. Specifically, the BeiDou point sets and visual 3D point clouds are concatenated in the world coordinate system to form a joint observation vector Z, and an observation noise covariance matrix R corresponding to the joint observation vector Z is constructed simultaneously. The observation noise covariance matrix R is a diagonal block matrix, including the BeiDou observation noise covariance and the visual 3D reconstruction noise covariance. The BeiDou observation noise covariance is set based on the positioning error estimate calculated by RTK; the visual 3D reconstruction noise covariance is estimated online based on the triangulation intersection angle and feature point reprojection error, or obtained from historical calibration samples. The joint observation vector Z represents the actual observation value, and the observation noise covariance matrix R represents the observation uncertainty. The joint observation vector Z and the observation noise covariance matrix R together constitute a joint observation data set.

[0067] Extract the traverse arc length coordinates corresponding to each observation point in the joint observation vector Z. Use the traverse arc length coordinates and state vector corresponding to each observation point as input to the state field generator to construct the observation equation. For any observation point, calculate the nearest point from the observation point to the predicted traverse shape curve, and use the cumulative arc length of the corresponding nearest point along the traverse axis as the traverse arc length coordinates corresponding to the observation point. When the nearest distance exceeds the observation matching distance threshold, the observation point is marked as an outside point and does not participate in the filtering update. The observation equation represents the predicted position of any observation point, that is, the three-dimensional coordinates calculated by the state field generator under the traverse arc length coordinates of the observation point and the current state vector.

[0068] The observation matching distance threshold is determined based on the standard deviation of BeiDou positioning, the standard deviation of visual 3D reconstruction position, and the standard deviation of state prediction position. Specifically, the square root of the sum of the squares of the three standard deviations is used to obtain the comprehensive position standard deviation, and the comprehensive position standard deviation is multiplied by the matching confidence factor to obtain the observation matching distance threshold. The matching confidence factor is determined based on the outlier elimination confidence level. When the three-standard-deviation criterion is adopted, the matching confidence factor is 3.

[0069] Unscented Kalman filtering is used to perform nonlinear state estimation of the predicted conductor morphology at the current time, generating 2n+1 sampling points, i.e., Sigma points. In each filtering cycle, based on the prior state estimation at the time and the corresponding covariance matrix, 2n+1 Sigma points are generated according to the standard unscented transformation formula, where n is the dimension of the state vector. The Sigma point generation formula and parameter values ​​are well-known techniques in this field and will not be described in detail here.

[0070] Substitute each Sigma point into the observation equation to obtain a set of predicted observations. Calculate the Kalman gain based on the predicted observations, and then use the residual between the actual observations and the predicted observations to update the predicted traverse shape at the current time. At the same time, update the state covariance to obtain the posterior optimal state estimate, which serves as the optimal state vector at the current time.

[0071] The method for obtaining the optimal state vector at the current moment unifies multi-source heterogeneous data into a single optimal state, so that subsequent trajectory reconstruction does not depend on which sensor is selected, but trusts the optimal result after fusion; at the same time, the online inversion of ice thickness and generalized coordinates enables the system to have the intelligent ability to know its load by observing its movement.

[0072] The system obtains the future wind speed, wind direction, temperature, and humidity of the target line's spatial location within the prediction period through meteorological service interfaces, local micro-meteorological prediction models, or line operation and maintenance platforms. It then performs time and spatial interpolation according to the conductor span position to form a future environment vector consistent with the input dimension of the state field generator. Based on the current optimal state vector and the future environment vector, the system analyzes the data using the state field generator to obtain the real-time three-dimensional trajectory of the conductor across the entire span and the predicted future trajectory of the conductor.

[0073] In one specific implementation, the method for obtaining the real-time three-dimensional trajectory and the predicted future trajectory of the traverse includes:

[0074] The optimal state vector is used as input to the state field generator to obtain the real-time three-dimensional trajectory of the conductor across the entire span; the optimal state vector and the future environment vector are used as input to the state field generator to obtain the predicted future trajectory of the conductor.

[0075] From discrete, low-dimensional state vectors, a continuous, high-dimensional full-field trajectory is recovered through forward propagation of the generator. This is the inverse process of the state-space model. This step transforms the abstract state into an intuitive trajectory, providing a direct data foundation for cross-distance calculation, visual monitoring, and operation and maintenance decision-making.

[0076] Based on the real-time three-dimensional trajectory and the future predicted trajectory, the dynamic minimum spatial clearance between the conductor and the preset crossing object is obtained; based on the dynamic minimum spatial clearance, hierarchical safety early warning information is generated.

[0077] Reference Figure 4 Methods for obtaining the dynamic minimum spatial clearance between the conductor and the preset crossing include:

[0078] Calculate the 3D Euclidean distance from all points in the current conductor point cloud to the nearest point on the preset crossing, obtain the current minimum distance, traverse all crossing models, and obtain the current minimum clearance composed of the current minimum distances; the preset crossings include crossings of lines, ground features, buildings, tree barriers, or traffic facilities; the crossing model is generated from the line design drawings, laser point clouds, GIS data, or field survey data, and uniformly converted to the line local coordinate system; for linear crossings, a line segment model is used; for planar crossings, a triangular mesh model is used; for volumetric crossings, a bounding box or 3D solid model is used.

[0079] The predicted minimum clearance from the traverse point cloud to each intersecting object model within the prediction period is calculated by combining the future predicted trajectory. Specifically, at each prediction time, the shortest distance from all points in the predicted traverse point cloud to each intersecting object model is calculated to obtain the predicted minimum clearance corresponding to that prediction time. All prediction times within the prediction period are traversed to obtain the minimum clearance for the prediction period. The current minimum clearance is compared with the minimum clearance for the prediction period, and the minimum value is taken as the dynamic minimum spatial clearance.

[0080] Warning levels are determined based on the dynamic minimum space clearance: when the dynamic minimum predicted clearance is less than the first clearance, it is marked as Level 1, corresponding to an emergency state; when the dynamic minimum predicted clearance is not less than the first clearance and less than the second clearance, it is marked as Level 2, corresponding to an alarm state; when the dynamic minimum predicted clearance is not less than the second clearance and less than the third clearance, it is marked as Level 3, corresponding to a warning state; when the dynamic minimum predicted clearance is not less than the third clearance, a normal state is output, and no warning information is generated. The first clearance is set according to the legally mandated minimum allowable clearance distance set by industry standards; the second clearance is set by superimposing a dynamic warning margin on the legally mandated minimum allowable clearance distance; the third clearance is set by superimposing a dynamic warning margin on the second clearance. The dynamic warning margin is determined jointly by the prediction uncertainty margin and the system response displacement. The prediction uncertainty margin is the product of the standard deviation of the predicted trajectory position and the warning confidence factor, which is set according to the warning false alarm tolerance; when using the three-standard-deviation criterion, the warning confidence factor is 3. The system response displacement is the product of the system response time and the maximum radial velocity of the conductor; the system response time is obtained by adding the warning calculation cycle, communication delay, alarm release delay, and on-site handling or control response time. Adding the prediction uncertainty margin to the system response displacement yields the dynamic warning margin; this step, by comparing the dynamic prediction distance with the statutory rigid threshold and the dynamic adaptive margin, minimizes false alarms while ensuring absolute safety. This step transforms technical data into business decisions, elevating the system from a monitoring tool to a proactive defense system.

[0081] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

[0082] Finally: The above description is only a preferred embodiment of this application and is not intended to limit this application. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the protection scope of this application.

Claims

1. A method for early warning of safe distances for ultra-high voltage power lines based on BeiDou positioning and AI imagery, characterized in that: include: Collect BeiDou positioning data, multi-view image data, and micro-meteorological data of the conductors of the target UHV line, and perform preprocessing and time synchronization to obtain BeiDou point set, synchronized image set, and environmental vector; A state field generator is constructed. Based on the BeiDou point set, the synchronous image set and the environmental vector, the motion state of the conductor is analyzed and the physical state components are obtained. A state vector is constructed based on the physical state components and the conditional state components. The state vector is then combined with a synchronized image set and a state field generator to perform state space estimation and obtain the optimal state vector at the current moment. The system acquires future micro-meteorological data via the internet, obtains future environmental vectors, and analyzes the current optimal state vector and future environmental vectors using a state field generator to obtain the real-time three-dimensional trajectory of the conductor across the entire span and the predicted future trajectory of the conductor. Based on the real-time three-dimensional trajectory and the future predicted trajectory, the dynamic minimum spatial clearance between the conductor and the preset crossing object is obtained by analysis. Based on the aforementioned dynamic minimum spatial clearance, hierarchical safety early warning information is generated.

2. The method for early warning of safe distance of ultra-high voltage lines based on BeiDou positioning and AI image fusion as described in claim 1, characterized in that, Methods for obtaining physical state components include: The state field generator consists of a physical foundation module and a neural residual module. The physical foundation module is used to calculate the ground state motion response of the conductor and obtain the physical feature vector. The neural residual module is used to perform nonlinear residual correction on the physical feature vector to obtain the corrected conductor displacement and velocity. Modal projection is performed on the corrected conductor displacement and velocity to obtain generalized displacement coordinates and generalized velocity coordinates of each order. The generalized displacement coordinates and generalized velocity coordinates are used together as physical state components.

3. The method for early warning of safe distance of ultra-high voltage lines based on BeiDou positioning and AI image fusion according to claim 2, characterized in that, Methods for obtaining physical feature vectors include: Calculate the natural frequency vector based on the total unit mass and equivalent horizontal tension of the ice-covered conductor; The linear dynamic displacement field, the foundation dynamic velocity field, and reference amplitudes of each order are obtained by using the modal superposition method and based on the natural frequency vector analysis. The basic displacement field is obtained by superimposing the static equilibrium position curve with the linear dynamic displacement field, and then the displacement gradient field is obtained by differentiation. The physical characteristic vector is obtained by splicing together the basic displacement field, basic dynamic velocity field, displacement gradient field, equivalent horizontal tension, natural frequency vector, and reference amplitudes of each order.

4. The method for early warning of safe distance of ultra-high voltage lines based on BeiDou positioning and AI image fusion as described in claim 3, characterized in that, Methods for obtaining static equilibrium position curves and equivalent horizontal tension include: The self-weight load per unit length of the conductor, the icing load per unit length, and the wind load per unit length are calculated based on real-time environmental vectors, and the comprehensive load per unit length is obtained through vector synthesis. Based on the unit length vertical load, span, and elevation difference between the two suspension points in the comprehensive unit length load, as well as the conductor arc length or reference horizontal tension, the static equilibrium position curve and equivalent horizontal tension are obtained by solving the catenary equation.

5. The method for early warning of safe distance of ultra-high voltage lines based on BeiDou positioning and AI image fusion according to claim 2, characterized in that, Methods for obtaining corrected conductor displacement and velocity include: By using the physical feature vector as input to the neural residual module, the corrected conductor displacement and velocity are obtained.

6. The method for early warning of safe distance of ultra-high voltage lines based on BeiDou positioning and AI image fusion according to claim 2, characterized in that, Methods for extracting physical state components from modified conductor displacement and velocity include: The modified conductor displacement field and the modified conductor velocity field are spatially discretized along the conductor span direction to obtain the three-dimensional displacement vector and three-dimensional velocity vector of each discrete point; using the pre-stored first K order mode shape functions, modal projection is performed on the modified conductor displacement field to obtain the generalized displacement coordinates of each order; modal projection is performed on the modified conductor velocity field to obtain the generalized velocity coordinates of each order, and the generalized displacement coordinates and generalized velocity coordinates of each order are used together as physical state components.

7. The method for early warning of safe distance of ultra-high voltage lines based on BeiDou positioning and AI image fusion as described in claim 1, characterized in that, Methods for obtaining the optimal state vector at the current moment include: A state vector is constructed based on the physical state components and the conditional state components; the state vector of the previous moment and the environment vector of the current moment are used as inputs to the state field generator to obtain the predicted traverse shape at the current moment; based on the predicted traverse shape, the visual three-dimensional point cloud of the traverse is extracted from the synchronous image set; using the BeiDou point set and the visual three-dimensional point cloud as observations, the observations and predictions are fused based on the state space estimation method to update and output the optimal state vector at the current moment.

8. The method for early warning of safe distance of ultra-high voltage lines based on BeiDou positioning and AI image fusion as described in claim 7, characterized in that, Methods for extracting visual 3D point clouds of conductors include: Within the ROI region, perform wire feature extraction, matching, and 3D reconstruction to output a visual 3D point cloud; Methods for obtaining the ROI region include: For the h-th industrial camera, the predicted wire shape is projected onto the two-dimensional image plane of the h-th industrial camera using the calibrated camera projection matrix. All projection points on the two-dimensional image plane are connected to form the wire prediction pixel trajectory. The predicted pixel trajectory is used as the center line to expand to both sides with a preset width to generate a pixel-level strip region, which is marked as the ROI region.

9. The method for early warning of safe distance of ultra-high voltage lines based on BeiDou positioning and AI image fusion according to claim 1, characterized in that, Methods for obtaining the real-time three-dimensional trajectory of a conductor across its entire span and the predicted future trajectory of the conductor include: The optimal state vector is used as input to the state field generator to obtain the real-time three-dimensional trajectory of the conductor across the entire span; the optimal state vector and the future environment vector are used as input to the state field generator to perform state prediction and obtain the future predicted trajectory of the conductor.

10. The method for early warning of safe distance of ultra-high voltage lines based on BeiDou positioning and AI image fusion according to claim 1, characterized in that, Methods for obtaining the dynamic minimum spatial clearance between the conductor and the preset crossing include: Calculate the three-dimensional Euclidean distance from each traverse point in the traverse point cloud to the nearest point on the cross-section model at the current time, obtain the current nearest distance for each cross-section model, and take the minimum value among the current nearest distances for all cross-section models as the current minimum net distance; The predicted minimum clearance between the point cloud of the traverse and each intersecting object model is calculated by combining the future predicted trajectory. The current minimum clearance is compared with the predicted minimum clearance during the prediction period, and the minimum value is taken as the dynamic minimum spatial clearance between the traverse and the preset intersecting object. The warning level is divided according to the dynamic minimum spatial clearance.

Images