A multi-spectrum vision-based ice thickness-mechanical inversion method for power transmission lines

Abstract

The application discloses a kind of based on multi-spectrum vision transmission line icing thickness-mechanical inversion method, this method only utilizes two-way perception data of visual sensor and multispectral sensor, obtains icing conductor outer diameter and calculates thickness initial value by multispectral vision sensor, identifies icing type and outputs real ice density by multispectral sensor, and then constructs fusion inversion objective function, to minimize the residual between visual geometry calculation and spectrum-physical constraint, dynamically calibrates icing thickness.The application solves the technical problems such as rough density assumption, visual interference from environment, multisensor data asynchronization and the like in traditional icing monitoring, and has the significant advantages of high precision, strong robustness and hardware simplification.

Description

Technical Field

[0001] This invention belongs to the field of power equipment condition monitoring and disaster prevention and mitigation technology, specifically relating to an ice thickness-mechanical inversion calculation method that combines visual thickness detection and multispectral ice density detection for use in transmission line icing monitoring scenarios. Background Technology

[0002] Icing on transmission lines is one of the major natural disasters threatening the safe operation of power grids. Accurate monitoring of ice thickness is fundamental for ice accretion early warning, ice melting decisions, and disaster assessment. Existing ice accretion monitoring technologies mainly include the following categories: Visible light / LiDAR-based visual measurement technology: Obtain the outer diameter of the conductor after it is covered with ice through image processing and 3D reconstruction, and then estimate the ice thickness.

[0003] Mechanical estimation technique based on tension / tilt sensors: By measuring changes in conductor tension, and combining this with a mechanical model, the weight and equivalent thickness of the ice layer can be inferred.

[0004] Microwave / ultrasound-based icing detection technology: Infers the icing state by utilizing changes in the propagation characteristics of electromagnetic waves or sound waves in the icing layer.

[0005] The existing technology has the following prominent drawbacks: Limitations of single-vision measurement: The imaging quality of vision sensors is severely degraded in harsh environments such as fog, rain, snow, and lens damage, resulting in systematic deviations in the geometrically reconstructed icing outer diameter and thickness estimation errors of more than 30%.

[0006] The density assumptions in mechanical inversion are coarse: traditional mechanical methods typically assume that the density of ice accretion is a fixed empirical value (e.g., 0.9 g / cm³). 3 Unable to distinguish rime (~0.9 g / cm³) 3 ), rime ice (~0.25 g / cm) 3 Different icing types, such as mixed frost and rime, result in huge differences in the thickness inverted by the same mechanical changes under different icing types.

[0007] Data asynchrony issues in multi-sensor systems: The solution of using vision + mechanics + meteorology multi-sensor systems suffers from problems such as differences in installation location, asynchronous sampling time, and accumulated calibration drift, which amplifies the data fusion error.

[0008] Lack of adaptive environmental anti-interference mechanism: Existing algorithms usually adopt a data fusion strategy with fixed weights, which cannot dynamically adjust the trust weights of each sensing channel according to the degree of environmental interference. Summary of the Invention

[0009] The purpose of this invention is to overcome the shortcomings of the prior art and provide a method and system for achieving high-precision ice thickness-mechanical inversion using only multispectral visual and multispectral sensors. This method constructs a fusion inversion objective function that includes environmentally adaptive weights, achieving closed-loop verification of visual geometric calculations and spectral-physical density constraints. This significantly improves the accuracy and environmental robustness of ice thickness calculation while greatly simplifying the system hardware architecture.

[0010] System hardware composition This system contains only two sensing layer sensors and one embedded computing terminal, featuring a streamlined architecture and easy deployment. This system design does not rely on any additional sensing devices such as tension sensors, tilt sensors, anemometers, or rain gauges. Both sensors are integrated into the same monitoring terminal, sharing the same clock source and coordinate system on the same embedded computing terminal. This fundamentally eliminates data fusion errors caused by differences in installation location, asynchronous sampling timing, and accumulated calibration drift in multi-sensor systems.

[0011] Parameter and symbol definitions The inherent parameters of the conductors are entered once before system deployment: Real-time sensor input: Visual geometric computation The icing area is represented as an annular cylinder. With the visual sensor calibrated to the corresponding focal length, the total outer diameter D_v of the conductor and ice can be directly measured, and the ice thickness d_vis can be calculated. d_vis = (D_v - D) / 2 Calculate the ice volume V_v: V_v = (π / 4) × (D_v 2 - D 2 ) This thickness is entirely derived from visual geometry calculations, and its accuracy is directly affected by the imaging quality of the visual sensor under adverse weather conditions. When the lens is fogged, obscured by snow, or there is insufficient lighting, D_v will produce a systematic error that is too large or too small, and V_v will also deviate from the true value. This is the fundamental reason why subsequent physical constraint inversion calibration is required.

[0012] Multispectral combined mechanical ice weight calculation The true ice density ρ_type:R provided by a multispectral sensor λ0 The multispectral reflectance at wavelength λ0 is obtained from measured calibration. ρ_type=ρ_guess⋅(1-R λ0) Based on the visual volume V_v mentioned above, calculate the physical ice weight W_ice: W_ice = ρ_type × g × V_v This leads to the total load calculated using multispectral joint mechanics: W_v = W_c + W_ice Traditional methods here directly use a fixed empirical density ρ_guess (e.g., 0.9 g / cm³). 3 Replacing ρ_type with ) results in the icing being low-density rime (~0.25 g / cm³) regardless of whether the actual icing is low-density rime. 3 (or high-density rime (~0.9 g / cm³)) 3 Using the same density to estimate weight in previous methods resulted in significant errors. This solution replaces ρ_guess with ρ_type, making W_v approximate the true value in the density dimension. The only uncertainty comes from the geometric error of the visual volume D_v, which is what the objective function will address next.

[0013] Construction of the fusion inversion objective function: core innovative method Define a scalar objective function J(d) with the desired thickness d as the independent variable, and obtain the final calculated thickness by minimizing this function: J(d) = (d_vis - d) 2 + λ × |ρ_guess × V(d) - ρ_type × V(d)| 2 Where V(d) = (π / 4) × [(2d + D) 2 - D 2 ], which is the ice volume corresponding to a thickness of d.

[0014] (I) Independent Variable and Geometric Relationship The unknown quantity d represents the final calibrated icing thickness. Its relationship with the outer diameter of the conductor is as follows: D(d) = 2d + D That is, when the thickness is d, the corresponding outer diameter of the wire is 2d + D. This relationship replaces the sensor-measured D_v in the function, becoming an adjustable variable.

[0015] (ii) First term -- Visual geometric constraint term: (d_vis - d) 2 (iii) Second term – Physical density consistency constraint term: λ × |ΔM| 2 Expand the inner expression: ΔM = (ρ_guess - ρ_type) × (π / 4) × [(2d + D) 2 - D 2 ] (iv) Environmental weight coefficient λ -- Dynamic adjustment driven by pure data Since the system does not contain meteorological sensors, the value of λ is dynamically determined by analyzing the inherent consistency of the data from the two sensors. λ = f(Confidence_visual, Confidence_spectral) λ is entirely determined by the quality index output by the sensor itself, requiring no input from any external meteorological sensor, thus maintaining the purity of the dual-sensor closed-loop system. Specifically, the visual image sharpness index can be obtained through traditional image feature extraction methods such as edge gradient intensity, texture entropy, and focus evaluation functions; the spectral matching confidence score can be obtained through the probability distribution entropy value of spectral feature matching.

[0016] Objective function solution and thickness output Calculation process: Initialization: d = d_vis (using the initial visual value as the starting point for iteration) Calculate the objective function value J(d) corresponding to the current d. The gradient descent method or Brent's one-dimensional search is used to search along the d-axis for the element that minimizes J(d). Analytical solution verification: Differentiate J(d) and set it to zero. The partial derivative of the objective function with respect to d is: make The analytical solution can be obtained as follows: d = [d_vis + 2λ(ρ_guess - ρ_type) 2 × π × D] / [1 + 4λ(ρ_guess -ρ_type) 2 × π] Constraint clipping: d Clip to a reasonable physical range [0, d_vis × (1+ε)] (ε is the allowed upward floating ratio to prevent unreasonable extrapolation in negative density scenes) Output the final calibration thickness d_final = d Let K = 2λπ(ρ_guess - ρ_type)2 , then d = (d_vis + K×D) / (1 + 2K).

[0017] Advantages of this invention: Superior to traditional methods: Traditional methods use tensile sensors to measure total weight and then infer thickness (using fixed theoretical density values); this scheme uses spectral density measurement + visual measurement of initial thickness, and mechanical inversion to verify the final thickness value. The perception path is completely different, avoiding the wind vibration noise and low-pass filter delay problems of weighing sensors.

[0018] High precision: The introduction of multispectral measured density ρ_type solves the systematic error caused by the assumption of fixed ice density in traditional mechanical models, and the accuracy of ice type identification can reach more than 95%.

[0019] Strong robustness: By dynamically adjusting the ratio of visual and physical constraints through the environmental weight coefficient λ, the system can still output reliable ice thickness data under extreme weather conditions such as sandstorms, dense fog, and blizzards.

[0020] Self-calibration capability: The objective function realizes bidirectional verification of visual geometric information and physical mechanical information, forming a complete technical closed loop from visual initial value → physical quality verification → thickness calibration.

[0021] Simplified hardware: Only two sensors are needed to complete high-precision icing monitoring, which reduces hardware costs by more than 40% and installation and maintenance costs by more than 60% compared with traditional multi-sensor solutions.

[0022] Purely data-driven: It can achieve environmental adaptive weight adjustment without relying on any external meteorological sensors, making the system highly independent and flexible in deployment. Attached Figure Description

[0023] The present invention will be further described in detail below with reference to the accompanying drawings and embodiments: Figure 1 The method steps of this invention; Figure 2 : Schematic diagram of ice thickness parameters. Detailed Implementation

[0024] The present invention will be further explained below with reference to specific implementation schemes, but it is not limited to the present invention. The structures, proportions, sizes, etc. shown in the accompanying drawings are only used to complement the content disclosed in the specification, so as to enable those skilled in the art to understand and read, and are not intended to limit the conditions under which the present invention can be implemented. Therefore, they have no substantial technical significance. Any modification of the structure, change of the proportion relationship or adjustment of the size, without affecting the effect and purpose that the present invention can produce, should still fall within the scope of the technical content disclosed in the present invention.

[0025] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to specific embodiments and accompanying drawings. The embodiments described herein are for illustrative purposes only and do not constitute any limitation on the scope of protection of this invention.

[0026] Example 1: Icing monitoring under normal weather conditions Step S1: Install a monitoring device integrating a vision camera and a multispectral analyzer on the transmission tower to ensure that the sensor's field of view covers the ice-covered area of ​​the conductor.

[0027] Step S2: Power on the system and initialize it, then import the wire parameters (e.g., wire diameter D = 18.8 mm, weight W_c = 1.52 kg / m).

[0028] Step S3: The vision sensor captures an image of the wire, extracts the contour, and calculates the visual outer diameter D_v = 25.6 mm and the initial thickness d_vis = (25.6 - 18.8) / 2 = 3.4 mm.

[0029] Step S4: The multispectral sensor emits light in the characteristic wavelength band and receives the reflected signal, calculating the true density ρ_type = 0.28 g / cm³. 3 .

[0030] Step S5: The embedded computing terminal analyzes the image sharpness index (normal edge gradient intensity) and spectral matching confidence (low entropy value, high matching degree), determines it to be a normal working condition, and sets λ = 0.2.

[0031] Step S6: Substitute the values ​​into the objective function and solve. The calculated value is K = 2 × 0.2 × π × (0.9 - 0.28). 2 ≈0.484, d = (3.4 + 0.484 × 18.8) / (1 + 2 × 0.484) ≈ 3.15 mm.

[0032] Example 2: Icing Monitoring under Extreme Weather Steps S1~S4: Same as in Example 1, but the environment is a blizzard, and the visual sensor image is obscured by snowflakes.

[0033] Step S5: The embedded computing terminal analyzes the image sharpness index (significantly decreased edge gradient intensity, increased texture entropy), indicating a decrease in visual reliability; however, the spectral matching confidence remains high (ice type identified as rime, ρ_type = 0.85 g / cm). 3 Let λ = 1.2.

[0034] Step S6: Substitute the values ​​into the objective function to solve. The initial visual value d_vis = 5.2 mm (may be higher due to snow occlusion), and the calculated value is K = 2 × 1.2 × π × (0.9 - 0.85). 2 ≈ 0.057, d = (5.2 + 0.057 ×18.8) / (1 + 2 × 0.057) ≈ 4.83 mm.

[0035] Effect Analysis: Although λ increases significantly, since ρ_type ≈ ρ_guess (rime density is close to empirical value), the K value itself is very small, therefore d It is still close to d_vis. This shows that when the density is consistent, even if the vision is unreliable, the physical constraints will not produce incorrect correction directions, demonstrating the inherent safety of the objective function.

[0036] Matters not covered in this invention are common knowledge.

[0037] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for inverting the icing thickness-mechanical properties of transmission lines based on multispectral vision, characterized in that: By constructing a fusion inversion objective function that includes environmental adaptive weights, closed-loop verification of visual geometric measurement and spectral-physical density constraints is achieved, significantly improving the accuracy of ice thickness calculation and environmental robustness, while greatly simplifying the system hardware architecture. System hardware composition: It comprises two sensing layer sensors and an embedded computing terminal, with a streamlined architecture and convenient deployment, specifically: Visual sensor, initial value calculation of ice thickness: Acquire images of ice-covered conductors, obtain the outer diameter D_v of the conductor's outer envelope after icing through traditional stereo vision three-dimensional geometric calculation, and calculate the initial value of ice thickness d_vis = (D_v - D) / 2; Multispectral sensor for ice density acquisition: It emits light signals of multiple characteristic bands to the ice surface and receives the reflected / transmitted spectra, and outputs the measured real ice density ρ_type; Embedded computing terminal, calculation execution: Built-in objective function solving engine, performs all inversion calculations, and outputs the final calibration thickness; By integrating both sensors into the same monitoring terminal and sharing the same clock source and coordinate system on the same embedded computing terminal, the data fusion error caused by differences in installation location, asynchronous sampling timing, and accumulated calibration drift in multi-sensor systems is fundamentally eliminated. Parameter and symbol definitions: The inherent parameters of the conductors are entered once before system deployment: D, outer diameter of bare conductor, known parameters of conductor's factory specifications; W_c, the self-weight load per unit length of the conductor at the factory, a known parameter on the conductor's nameplate at the factory; g, acceleration due to gravity, constant (taken as local standard value 9.8 m / s²) 2 ); ρ_guess, a fixed empirical value for ice density (preferably 0.9 g / cm³). 3 ), industry default value; Real-time sensor input: D_v, the outer diameter of the icy wire measured by the vision sensor; vision sensor; d_vis, initial value of ice thickness, calculated from visual sensor results; ρ_type, the current true density of ice accretion measured by the multispectral sensor; multispectral sensor; ρ_guess, theoretical density of icing, standard value: 0.9 g / cm³ 3 ; Visual geometric computation: The icing area is represented as an annular cylinder. With the visual sensor calibrated to the corresponding focal length, the total outer diameter D_v of the conductor and ice can be directly measured, and the ice thickness d_vis can be calculated. d_vis = (D_v - D) / 2 Calculate the ice volume V_v: V_v = (π / 4) × (D_v 2 - D 2 ) This thickness is entirely derived from visual geometry calculations, and its accuracy is directly affected by the imaging quality of the visual sensor under adverse weather conditions. When the lens is fogged, obscured by snow, or there is insufficient light, D_v will produce a systematic error of being too large or too small, and V_v will also deviate from the true value. This is the fundamental reason why physical constraint inversion calibration is required in the future. Multispectral combined mechanical ice weight calculation: The true ice density ρ_type:R provided by a multispectral sensor λ0 The multispectral reflectance at wavelength λ0 is obtained from measured calibration. ρ_type=ρ_guess⋅(1-R λ0 ) Based on the visual volume V_v mentioned above, calculate the physical ice weight W_ice: W_ice = ρ_type × g × V_v This leads to the total load calculated using multispectral joint mechanics: W_v = W_c + W_ice Traditional methods here directly use a fixed empirical density ρ_guess (e.g., 0.9 g / cm³). 3 Replacing ρ_type with ) results in the icing being low-density rime (~0.25 g / cm³) regardless of whether the actual icing is low-density rime. 3 (or high-density rime (~0.9 g / cm³)) 3 All previous methods estimated the weight using the same density, resulting in a huge error. This solution uses ρ_type instead of ρ_guess, making W_v close to the true value in the density dimension. The only uncertainty comes from the geometric error of the visual volume D_v, which is what the objective function needs to solve next. The core innovative method for constructing the fusion inversion objective function: Define a scalar objective function J(d) with the desired thickness d as the independent variable, and obtain the final calculated thickness by minimizing this function: J(d) = (d_vis - d) 2 + λ × |ρ_guess × V(d) - ρ_type × V(d)| 2 Where V(d) = (π / 4) × [(2d + D) 2 - D 2 ], which is the ice volume corresponding to a thickness of d.

2. The method for inverting icing thickness and mechanics of transmission lines based on multispectral vision according to claim 1, characterized in that: (I) Independent Variable and Geometric Relationship The unknown quantity d represents the final calibrated icing thickness; its relationship with the outer diameter of the conductor is as follows: D(d) = 2d + D That is, when the thickness is d, the corresponding outer diameter of the wire is 2d + D; this relationship replaces the D_v measured by the sensor in the function and becomes an adjustable variable; (ii) First term -- Visual geometric constraint term: (d_vis - d) 2 (iii) Second term – Physical density consistency constraint term: λ × |ΔM| 2 Expand the inner expression: ΔM = (ρ_guess - ρ_type) × (π / 4) × [(2d + D) 2 - D 2 ] (iv) Environmental weight coefficient λ -- Dynamic adjustment driven by pure data Since the system does not contain meteorological sensors, the value of λ is dynamically determined by analyzing the inherent consistency between the two sensor data streams: λ = f(Confidence_visual, Confidence_spectral). λ is entirely determined by the quality index output by the sensor itself, without requiring any input from external meteorological sensors, thus maintaining the purity of the dual-sensor closed-loop system. Among them, the visual image sharpness index can be obtained through traditional image feature extraction methods such as edge gradient intensity, texture entropy, and focus evaluation function; the spectral matching confidence can be obtained through the probability distribution entropy value of spectral feature matching. Objective function solution and thickness output: Calculation process: Initialization: d = d_vis (using the initial visual value as the starting point for iteration) Calculate the objective function value J(d) corresponding to the current d. The gradient descent method or Brent's one-dimensional search is used to search along the d-axis for the element that minimizes J(d). Analytical solution verification: Differentiate J(d) and set it to zero. The partial derivative of the objective function with respect to d is: make The analytical solution can be obtained as follows: d = [d_vis + 2λ(ρ_guess - ρ_type) 2 × π × D] / [1 + 4λ(ρ_guess - ρ_type) 2 × p] Constraint clipping: d Clip to a reasonable physical range [0, d_vis × (1+ε)] (ε is the allowed upward floating ratio to prevent unreasonable extrapolation in negative density scenes) Output the final calibration thickness d_final = d Let K = 2π(ρ_guess - ρ_type) 2 , d = (d_vis + K×D) / (1 + 2K)。

Images