Aged cable current-carrying capability evaluation method and system based on micro displacement identification

Abstract

The invention discloses an aged cable current-carrying capability evaluation method and system based on micro displacement identification, and the method comprises the steps: obtaining a cable operation video through non-contact shooting, calculating an optical flow field of multiple regions of interest through a Horn-Schunck dense optical flow algorithm and a spatial integral averaging technology, and carrying out the evaluation of the current-carrying capability of an aged cable. A breathing characteristic amplitude locked at power frequency multiplication (100Hz) is extracted through FFT (Fast Fourier Transform) frequency domain analysis and a truncation average method; then, constructing a standard fingerprint database based on the Arrhenius theory, and establishing a nonlinear mapping relationship among the vibration characteristics, the aging state and the current-carrying capacity correction coefficient by using a support vector regression model optimized by a genetic algorithm; and finally, calculating the actual available current-carrying capacity of the aged cable in combination with the environmental parameters. According to the invention, environmental noise interference can be effectively eliminated, non-contact, high-precision and intelligent evaluation of the current-carrying capability of the aged cable is realized without power failure and contact with the cable body, and safe and stable operation of a power system is guaranteed.

Description

Technical Field

[0001] This invention belongs to the field of power cable monitoring technology, specifically relating to a method and system for assessing the current-carrying capacity of aging cables based on micro-displacement identification. Background Technology

[0002] As a key carrier of modern electrical energy transmission, the insulation condition of power cables directly affects the safe and stable operation of the power system. With increasing service life, cable insulation inevitably ages, leading to a significant decrease in its heat resistance and current-carrying capacity. Therefore, accurately assessing the dynamic current-carrying capacity of aging cables is of great importance for preventing overload faults and ensuring the reliability of the power grid.

[0003] Currently, cable current-carrying capacity assessment mainly relies on temperature monitoring, with common methods including thermocouples and fiber Bragg gratings (FBGs). While thermocouples are inexpensive and easy to implement, they are susceptible to electromagnetic interference and have limitations in single-point measurement, making it difficult to guarantee measurement accuracy in complex environments. FBGs achieve high-precision temperature measurement by sensing Bragg wavelength drift, but this method typically requires inserting sensors into the cable, which is not only costly but may also damage the cable's structural integrity, posing potential safety hazards.

[0004] Under the influence of alternating electromagnetic fields, power-carrying cables exhibit micron-level radial mechanical vibrations. This periodic deformation characteristic induced by electromagnetic force is known as the cable's "breathing effect." The vibration amplitude is strongly correlated with the mechanical modulus (viscoelastic properties) of the insulation material, and the mechanical modulus evolves systematically with increasing insulation aging. Based on this, this paper proposes a non-invasive assessment method that accurately captures and quantifies the "breathing" vibration characteristics of cables at different aging stages, establishing a mapping relationship between vibration amplitude and insulation state, thereby achieving a non-destructive and accurate assessment of the current-carrying capacity of aging cables. Summary of the Invention

[0005] To address the above problems, this invention provides a method for evaluating the current-carrying capacity of aging cables based on micro-displacement identification.

[0006] A method for assessing the current-carrying capacity of aging cables based on micro-displacement identification includes the following steps: S1. Obtain a sequence of video images of the power cable to be evaluated in its operating state; S2. Magnification of minute displacements in the video image sequence of the cable under test: The video image sequence is processed using a phase-based Euler video magnification algorithm to amplify the minute displacements of the power cable under test, resulting in an enhanced video sequence; S3. Optical flow calculation and feature extraction: Several regions of interest are defined in the enhanced video sequence, and the temporal vibration features of each region are calculated and extracted using the Horn-Schunck optical flow algorithm; the regions of interest are at least 5 non-overlapping regions of the same specification distributed along the axial direction of the power cable to be evaluated. S4. Multi-source feature fusion and noise suppression based on frequency domain analysis: The time-domain vibration characteristics of each region are subjected to discrete fast Fourier transform and converted into frequency domain spectrum. The frequency domain spectrum data is cleaned based on the truncated average method to obtain feature values. The remaining feature values ​​are calculated by arithmetic average to obtain the breathing characteristic amplitude that characterizes the overall aging state of the power cable to be evaluated. S5. Construct a “vibration-aging-current carrying capacity” mapping model: Based on a pre-established sample database, use the support vector regression (SVR) algorithm to establish a nonlinear mapping relationship between the breathing characteristic amplitude and the cable aging state and current carrying capacity correction coefficient, and optimize to obtain the final SVR decision function model. S6. Calculation of Current Carrying Capacity Correction Factor: Input the breathing characteristic amplitude obtained in step 4 into the final SVR decision function model, and output the actual correction factor for the current carrying capacity of the power cable to be evaluated. ; S7. Final assessment of current carrying capacity: Based on the cable's design rated current carrying capacity and the calculated correction factor, calculate the actual usable current carrying capacity of the aging cable and output the assessment results.

[0007] In a further improvement, in step S1, a high-speed industrial camera is fixed on a tripod and taken at a distance of 2 meters from the cable surface to capture video of the cable to be evaluated, thereby obtaining a video image sequence of the cable to be evaluated.

[0008] Further improvements are made in step S2, where the Euler video upscaling algorithm follows these steps: (a) Complex and Controllable Pyramid Decomposition Each frame of the video image sequence is subjected to complex and controllable pyramid decomposition to obtain the local amplitude spectrum and local phase spectrum of the video image, ensuring that local small phase processing is equivalent to local motion processing: For the input video frame It is decomposed into high-pass residuals, complex bandpass subbands of multiple scales and directions, and low-pass residuals; the decomposition mathematical model is expressed as: ; in, The high-pass residual contains extremely fine texture edges or high-frequency noise information from the sensor that exceed the highest decomposition scale in the image. The low-pass residual contains the fundamental frequency illumination distribution and large-scale contour information of the image; In order to scale and direction The bandpass complex subband signal; x represents the horizontal spatial coordinate of the image pixel; y represents the vertical spatial coordinate of the image pixel; t represents time; For each complex subband signal This can be further expressed in the form of local amplitude and local phase: ; in, The local amplitude spectrum characterizes the image texture intensity; It is a local phase spectrum, and its changes on the time axis directly correspond to the minute local motions of the image; (b) Bandpass filtering and phase amplification For the local phase spectrum obtained by decomposition A bandpass filter is applied in the time dimension to filter out high-frequency noise and DC drift; let the bandpass filter be... The magnification factor is The phase change that contains vibration characteristics and the magnified new phase spectrum The calculation is as follows: ; In the formula: This represents temporal convolution operations; Used to linearly enhance the amplitude of weak vibrations, making them visually recognizable; (c) Video reconstruction and compositing Using the original local amplitude spectrum and the new phase spectrum after amplification Construct the amplified complex subband signal: ; To maintain the image clarity and background stability of the reconstructed video, phase correction is performed only on the bandpass subband signal, while the high-pass residual is not corrected. and low-pass residual Keep the original values ​​unchanged; the final enlarged video sequence The signal is obtained by performing a complex and controllable inverse pyramid transform on the unprocessed residual and the amplified subband signal, and then superimposing and reconstructing the result. ; The above reconstruction process generates the final enlarged video sequence. , As an enhanced video sequence.

[0009] Further improvements, the specific steps of step S3 are as follows: S3.1) Select at least five non-overlapping areas along the axial direction of the power cable to be evaluated as observation windows, wherein the observation windows cover the surface texture area of ​​the cable's outer sheath; these are respectively marked as... ,..., Together, they characterize the overall "breathing" state of the power cable being evaluated. S3.2) Global Optimization Solution: For each observation window, the Horn-Schunck optical flow algorithm is used to construct an energy functional that includes constant brightness constraints and smoothness constraints. E : ; In the formula: These are the optical flow velocity components of the pixel in the horizontal and vertical directions, respectively; Image brightness exist Partial derivatives in the spatial direction and in the time direction t; first term The first term is a constant brightness constraint term, used to ensure that the brightness remains unchanged before and after pixel movement; the second term... This is a smoothness constraint term used to constrain the consistency of motion in neighboring pixels; These are the smoothing weight coefficients, used to adjust the strength of the smoothing constraints; The symbol for double integral indicates that the energy function is integrated and summed over the entire image domain; Let be the differential area element in spatial coordinates, representing the horizontal coordinate of the image plane as the integral variable. and vertical coordinates ; To achieve a numerical solution for the energy functional E, the Gauss-Seidel iterative algorithm is employed: First, based on the Euler-Lagrange equations, the energy functional minimization problem is transformed into solving the following system of linear equations: In the discretized mesh, the Laplacian operator is used to approximate the pixels. The deviation between the velocity component and its neighborhood average is represented by a smoothing term, and the specific iterative update formula is as follows: ; In the formula: k represents the current iteration number, and The first The pixel points obtained in the next iteration Optical flow velocity in the horizontal and vertical directions; For smoothing weighting coefficients; , For pixels The average velocity within the neighborhood is calculated using a 4-neighborhood weighted average, defined as: ; Repeat the above calculation process until the error between two consecutive iterations is less than a preset threshold. , Get each pixel within the window Optimal optical flow velocity vector x, y, t, u, and v represent the horizontal coordinate of a pixel, the vertical coordinate of a pixel, time, the horizontal optical flow velocity component, and the vertical optical flow velocity component, respectively. S3.3) Spatial Domain Feature Aggregation: Introducing a spatial averaging operator, the optical flow velocity vector magnitude of all effective pixels within each observation window is integrally averaged to obtain the instantaneous average motion intensity of that region in the current frame, as shown in the following formula: ; in, For the k-th observation window region, Let k be the area of ​​the observation window regions, i.e., the total number of pixels; The instantaneous motion magnitude of a pixel; For pixels exist The horizontal optical flow velocity component at any given time. For pixels exist The vertical optical flow velocity component at any given moment; Let be the vibration amplitude sequence of the k-th region over time; through the above calculations, the vibration amplitude sequences of each observation window distributed along the axial direction of the cable surface are finally obtained, denoted as . ,..., ..., As the temporal vibration characteristics of each region.

[0010] Further improvements, the specific steps of step S4 are as follows: S4.1) DC removal and windowing: First, subtract the mean of all time-domain vibration characteristics from each time-domain vibration characteristic to eliminate DC bias. Then, process the DC-removed signal... Adding a Hanning window, the signal after adding a Hanning window The calculation is as follows: ; In the formula, N represents the total number of sampling points within the observation window, and n represents the sequence number of the current sampling point, with a value range of... ; S4.2) Discrete Fourier Transform: For the preprocessed signal Perform a Discrete Fourier Transform to map from the time domain to the frequency domain; the transform formula is: ; In the formula, Let be the complex spectral value at the k-th frequency point, where j is the imaginary unit; S4.3) Amplitude Extraction: The electrodynamic frequency experienced by the cable under a 50Hz power frequency current is twice the power supply frequency; therefore, the target characteristic frequency is... Locked to Based on the sampling theorem and the relationship between frequency resolution, calculate the spectral index corresponding to 100Hz. : ; In the formula: Extract the spectral magnitude at this index, based on the sampling frame rate of the video capture. As the single-frequency vibration intensity under electrodynamic drive in this region, the vibration amplitude at 100Hz for different observation windows is denoted as follows: , As a frequency domain eigenvalue; Indicates taking the integer part; S4.4) Outlier Removal: The frequency domain feature values ​​are cleaned using the "truncated averaging method": First, the extracted feature amplitude set is... Sort the values ​​in ascending order to obtain an ordered sequence: Then, the maximum and minimum values ​​in the ordered sequence are removed to eliminate the interference of extreme outliers on the overall evaluation; 3) Feature Fusion and Final Index Generation: The remaining intermediate feature amplitudes are arithmetically averaged to obtain the breathing characteristic amplitude, which characterizes the overall aging state of the power cable to be evaluated. The calculation formula is as follows: ; In the formula: The j-th feature magnitude is retained after sorting. The unit for is micrometers. .

[0011] Further improvements, the specific steps of step S5 are as follows: S5.1) Construction of a standardized fingerprint database: Construct a sample database containing multi-dimensional aging features: S5.1.1) Sample Preparation: Select a short sample cable of the same type as the cable to be tested, place it in a high-temperature aging chamber for artificial accelerated thermal aging experiment, and set the accelerated aging parameters according to the Arrhenius equation: ; In the formula For lifespan, This is the aging temperature. For activation energy, k Using Boltzmann's constant as an example, gradient aging samples with equivalent operating years of 0, 10, 20, and 30 years were prepared by controlling the heating time. Input feature x: In a shielded room, apply 50% of the rated current to each of the above-mentioned aging samples, and measure the breathing characteristic amplitude of each sample cable using the methods in steps S1-S4. ; As input features; Output label y: Perform thermal elongation tests on each aged sample to measure the elongation at break retention rate of the insulation layer under high-temperature load. Calculate the maximum permissible current carrying capacity of each sample under the current insulation condition according to IEC 60287 standard. and the rated current carrying capacity of the cable And define the current carrying capacity correction factor. , serving as the "truth value" label for the model; S5.1.2) Normalization mapping of feature space: Perform Min-Max linear normalization on the input feature x and output label y to obtain low-dimensional nonlinear data: ; Map all input and output data of training samples to the dimensionless interval [0, 1]; where X represents the set of input features x and Y represents the set of output labels y; This represents the i-th input feature after normalization; This represents the i-th output label after normalization; min indicates taking the minimum value, and max indicates taking the maximum value. S5.2) Constructing an SVR model with slack variables: Using the Gaussian radial basis kernel function, the normalized low-dimensional nonlinear data is mapped to a high-dimensional feature space. In order to achieve the optimal balance between "model complexity" and "fitting error", the following convex quadratic programming problem is constructed: Objective function: ; in, Characterizes the smoothness of the model. Characterize empirical risk; The weight vector determines the orientation of the hyperplane in the feature space. This represents the bias term, which determines the position of the hyperplane's intercept relative to the origin. Representing the One sample point exceeds the upper bound of the insensitive loss band; Representing the One sample point exceeds the lower bound of the insensitive loss band; Constraints: ; In the formula, As a penalty factor, The insensitive loss band width, where T denotes matrix transpose. This represents the high-dimensional feature vector after mapping. Indicates the first The true label values ​​of each training sample; Introducing Lagrange multipliers By utilizing the KKT conditions, the primal problem can be transformed into a dual problem, allowing for the introduction of a kernel function and a reduction in computational complexity. Dual objective function: ; Dual constraints: ; In the formula, For the first A Lagrange multiplier, No. Lagrange multiplication of training samples , are the i-th and j-th input feature vectors after normalization, respectively, and m is the total number of training samples. For the normalized i-th true label value, the Gaussian radial basis function is: , For kernel parameters, As a penalty factor; Based on the optimal solution of the dual variables obtained by solving The final SVR regression decision function is: ; In the formula, This is the predicted value of the cable current carrying capacity correction factor after normalization; This is the normalized feature vector of the sample to be tested; These are the normalized support vectors of the training set; S5.3) Global optimization of SVR hyperparameters based on genetic algorithm: A real-number encoded genetic algorithm is used to input two key hyperparameters. and As a chromosome gene, the global optimal solution is searched; the specific implementation steps are as follows: S5.3.1) Population initialization and gene encoding: The size of the random generation is The initial population, each chromosome consists of two gene loci, each representing a hyperparameter. and The search space is mapped to a continuous real number field using a real number logarithmic encoding method to ensure search accuracy. S5.3.2) Fitness function construction: For each individual in the population Substitute the data into the SVR model and perform 5-fold cross-validation using the training set. Calculate the root mean square error under cross-validation and define the fitness function. The reciprocal of RMSE: ; In the formula, For real labels, To validate the predicted values, To prevent tiny constants with a denominator of zero; fitness The larger the value, the higher the model prediction accuracy under that set of parameters; S5.3.3) Genetic evolution operations: Based on the principle of survival of the fittest, the following operator operations are performed on the population to iteratively generate the next generation: Selection operator: The roulette wheel selection method is used, and individuals with higher fitness have a greater probability of being selected to pass on their genes to the next generation; Crossover calculation: Setting crossover probabilities Arithmetic crossover is used to linearly combine selected parent chromosomes to generate new offspring genes, thereby enhancing the local search capability of the population. Mutation operator: Sets the mutation probability Non-uniform Gaussian mutation is used to superimpose random perturbations that conform to a normal distribution on gene values ​​in order to maintain population diversity and prevent premature convergence of the algorithm. S5.3.4) Termination condition and optimal solution output: The maximum number of evolution generations is set to The algorithm terminates when the number of iterations reaches the upper limit or the optimal fitness of the population no longer significantly improves after 10 consecutive generations; it outputs the parameter combination corresponding to the individual with the highest fitness at this point. We use these parameters as the globally optimal hyperparameters; and then train the final SVR decision function model using the optimal parameter combination and the normalized training data.

[0012] Further improvements, the specific steps of step S6 are as follows: S6.1) Isodistribution normalization of input data: Normalize the respiratory characteristic amplitude output in real time from step 4. As the input to be predicted ,right Perform Min-Max normalization to map it to the same [0, 1] feature space as the training set: ; In the formula, and These are the minimum and maximum values ​​of all amplitude samples in the fingerprint database, respectively. S6.2) Model Inference and Output Denormalization: Denormalize the normalized output... Input the final SVR decision function model to calculate the predicted value. ;Utilize the carrying capacity correction factor labels in the fingerprint database Y The statistical measure is used to perform an inverse transformation on the prediction results, restoring them to the actual correction coefficients with physical meaning. : ; In the formula, and These are the minimum and maximum values ​​of all amplitude samples in the historical fingerprint database, respectively.

[0013] Further improvements are made, and the specific steps of step S7 are as follows: S7.1) Based on the manufacturer's nameplate parameters and environmental monitoring data of the power cable to be evaluated, calculate the actual usable current-carrying capacity of the power cable under its current aging state. : ; In the formula, The rated current carrying capacity of the cable. This is the ambient temperature correction factor. If the load falls below the current operating load, an overload warning will be triggered. S7.2) Decision Output and Early Warning: Will Compared with the current actual operating load current of the power cable to be evaluated Compare: like Output "Status is good"; like The system outputs a "warning status" and recommends limiting load growth. like This immediately triggers an "overload alarm".

[0014] A system for evaluating the current-carrying capacity of aging cables based on micro-displacement identification includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, it implements the aforementioned method for evaluating the current-carrying capacity of aging cables based on micro-displacement identification.

[0015] Advantages of this invention: 1. This invention overcomes the limitations of traditional thermocouple-based contact measurement, which has low accuracy, and fiber optic grating-based measurement, which requires invasive installation and is costly. It adopts a non-contact measurement mode based on computer vision, which eliminates the need for power outages and cable structure damage by inspection personnel, greatly reducing operational risks and ensuring the safety of power grid operation.

[0016] 2. In terms of anti-interference and measurement accuracy, this invention achieves precise capture of pixel-level minute displacements by integrating the phase-based Euler video amplification algorithm and the Horn-Schunck global optical flow algorithm. It also innovatively uses FFT frequency domain analysis to lock the cable's unique 100Hz electrodynamic frequency. Combined with multi-ROI truncation averaging technology, it effectively filters out environmental noise interference such as wind, vehicle vibration, and changes in lighting, ensuring the purity of the extracted "breathing feature amplitude".

[0017] 3. In view of the highly nonlinear and saturated characteristics of cable insulation materials during the aging process, this invention constructs a support vector regression (SVR) evaluation model based on genetic algorithm (GA) optimization. Combined with the principle of minimizing structural risk, it has stronger generalization ability and prediction accuracy compared with traditional linear fitting, and can accurately establish the nonlinear mapping relationship between vibration characteristics and current carrying capacity correction coefficient.

[0018] 4. Based on the Arrhenius reaction rate equation and IEC standards, this invention constructs a scientific standard fingerprint database, ensuring that the evaluation results have solid theoretical support and engineering application credibility, and providing an efficient, intelligent and reliable technical means for the operation and maintenance decision-making of aging cables. Attached Figure Description

[0019] Figure 1 This is a schematic diagram of the process for evaluating the current-carrying capacity of aging cables based on micro-displacement identification.

[0020] Figure 2 Schematic diagram of hardware layout for implementation scenario.

[0021] Figure 3 This is a schematic diagram of the phase-based Euler video upscaling algorithm.

[0022] Figure 4 A schematic diagram of the region of interest (ROI) and optical flow vectors.

[0023] Figure 5 A schematic diagram illustrating the principles of SVR model construction and prediction. Detailed Implementation

[0024] The technical solution of the present invention will be specifically described below through specific embodiments and in conjunction with the accompanying drawings.

[0025] Example 1 Figure 1 This is a schematic diagram of the process for evaluating the current-carrying capacity of aging cables based on micro-displacement identification.

[0026] like Figure 1 As shown, the flowchart of the method for assessing the current-carrying capacity of aging cables based on micro-displacement identification includes: Step S1: Obtain a video of the aging cable to be evaluated.

[0027] like Figure 2 As shown, a high-speed industrial camera was fixed on a tripod, 2 meters away from the cable surface, ensuring that the cable was completely in the video frame. Considering that the frequency of the electrodynamic force generated by the AC power of the power cable is twice the power supply frequency, the camera sampling frame rate was set to 400fps, the resolution was set to 1920 × 1080, the shooting time was 5 seconds, and a video sequence containing 2000 frames of images was obtained.

[0028] Step S2: Use the phase-based Euler video amplification algorithm to amplify the micro-motion of the video acquired in step 1, thus amplifying the breathing effect of the aging cable.

[0029] The flow of the phase-based Euler video upscaling algorithm is as follows: Figure 3 As shown, the specific implementation steps are as follows: (a) Complex and Controllable Pyramid Decomposition Each frame of the video image sequence is subjected to complex and controllable pyramid decomposition to obtain the local amplitude spectrum and local phase spectrum of the video image, ensuring that local small phase processing is equivalent to local motion processing: For the input video frame It is decomposed into high-pass residuals, complex bandpass subbands of multiple scales and directions, and low-pass residuals; the decomposition mathematical model is expressed as: ; in, The high-pass residual contains extremely fine texture edges or high-frequency noise information from the sensor that exceed the highest decomposition scale in the image. The low-pass residual contains the fundamental frequency illumination distribution and large-scale contour information of the image; In order to scale and direction The bandpass complex subband signal; x represents the horizontal spatial coordinate of the image pixel; y represents the vertical spatial coordinate of the image pixel; t represents time; For each complex subband signal This can be further expressed in the form of local amplitude and local phase: ; in, The local amplitude spectrum characterizes the image texture intensity; It is a local phase spectrum, and its changes on the time axis directly correspond to the minute local motions of the image; (b) Bandpass filtering and phase amplification For the local phase spectrum obtained by decomposition A bandpass filter is applied in the time dimension to filter out high-frequency noise and DC drift; let the bandpass filter be... The magnification factor is The phase change that contains vibration characteristics and the magnified new phase spectrum The calculation is as follows: ; In the formula: This represents temporal convolution operations; Used to linearly enhance the amplitude of weak vibrations, making them visually recognizable; (c) Video reconstruction and compositing Using the original local amplitude spectrum and the new phase spectrum after amplification Construct the amplified complex subband signal: ; To maintain the image clarity and background stability of the reconstructed video, phase correction is performed only on the bandpass subband signal, while the high-pass residual is not corrected. and low-pass residual Keep the original values ​​unchanged; the final enlarged video sequence The signal is obtained by performing a complex and controllable inverse pyramid transform on the unprocessed residual and the amplified subband signal, and then superimposing and reconstructing the result. ; The above reconstruction process generates the final enlarged video sequence. , As an enhanced video sequence.

[0030] Step S3: Optical flow field calculation and region segmentation. Several regions of interest (ROIs) are defined in the enhanced video sequence. The Horn-Schunck optical flow algorithm is used to calculate the optical flow field in each region, and the average vibration amplitude sequence of each region is extracted.

[0031] Based on the enhanced video obtained in step 2 after Euler amplification, the minute vibration field on the cable surface is calculated. The specific steps are as follows: 1) Region of Interest (ROI) division: such as Figure 4 As shown, in the enhanced video image, at least five non-overlapping areas are selected along the cable axis as observation windows, and these observation windows cover the surface texture area of ​​the cable outer sheath; these are respectively labeled as... These areas collectively characterize the overall "breathing" state of the cable.

[0032] 2) Global Optimization Solution: For each Region of Interest (ROI), the Horn-Schunck optical flow algorithm is used to construct an energy functional E that includes constant brightness constraints and smoothness constraints. This method assumes that the image brightness remains constant over a very short time and that the velocity field changes smoothly in space. The constructed energy functional formula is as follows: ; In the formula: These are the optical flow velocity components of the pixel in the horizontal and vertical directions, respectively; Image brightness exist Partial derivatives in the spatial direction and in the time direction t; first term The first term is a constant brightness constraint term, used to ensure that the brightness remains unchanged before and after pixel movement; the second term... This is a smoothness constraint term used to constrain the consistency of motion in neighboring pixels; The smoothing weight coefficient is used to adjust the strength of the smoothing constraint.

[0033] To achieve the numerical solution of the energy functional E, the Gauss-Seidel iterative algorithm is specifically adopted. First, based on the Euler-Lagrange equations, the energy functional minimization problem is transformed into solving the following system of linear equations.

[0034] In the discretized mesh, the Laplacian operator is used to approximate the pixels. The deviation between the velocity component and its neighborhood average is represented by a smoothing term. The specific iterative update formula is as follows: ; In the formula: k represents the current iteration number, and The first The pixel points obtained in each iteration Optical flow velocity in the horizontal and vertical directions; These are the spatiotemporal gradient values ​​at that pixel; These are the smoothing weighting coefficients. , The average velocity within the neighborhood of this pixel is calculated using a 4-neighborhood weighted average, which is defined as follows:

[0035] Repeat the above calculation process until the error between two consecutive iterations is less than a preset threshold. , Get each pixel within the window Optimal optical flow velocity vector .

[0036] 3) Spatial Domain Feature Aggregation: A spatial averaging operator is introduced to calculate the integral average of the optical flow velocity vector magnitude of all effective pixels within each region (ROI), yielding the instantaneous average motion intensity of that region in the current frame. The formula is as follows:

[0037] Where k represents the k-th region of interest (k=1, 2, ..., 5). For the k-th observation window region, This represents the area of ​​the region (i.e., the total number of pixels). The instantaneous motion magnitude of a pixel; Let be the vibration amplitude sequence of the k-th region over time. Through the above calculations, the vibration amplitude sequences of five regions distributed axially along the cable surface are finally obtained, denoted as […]. , Used for subsequent feature fusion and current carrying capacity assessment.

[0038] Step S4: Multi-source feature fusion and noise suppression based on frequency domain analysis. The time-domain vibration characteristics of each region are subjected to discrete fast Fourier transform (FFT) to convert them into frequency domain spectra. The data is cleaned using the truncated averaging method, and the remaining eigenvalues ​​are calculated by arithmetic averaging to obtain the breathing characteristic amplitude characterizing the overall aging state of the cable.

[0039] 1) Signal preprocessing and time-frequency domain transformation: The time-domain vibration amplitude sequences of the five regions of interest (ROIs) output in step 3. , Digital signal processing is performed separately. The specific process includes the following three sub-steps: 1. DC removal and windowing: Since the original vibration signal contains a non-zero static offset (DC component), the mean of each sequence is first subtracted to eliminate the DC bias. Then, the DC-removed signal... Add a Hanning window for processing. Signals after adding the window. The calculation is as follows:

[0040] In the formula, N is the total number of sampling points.

[0041] 2. Discrete Fourier Transform (DFT): For preprocessed signals... Perform a Discrete Fourier Transform (DFT) to map the frequency domain from the time domain. The transform formula is:

[0042] In the formula, Let be the complex spectral value at the k-th frequency point, where j is the imaginary unit.

[0043] 3. Amplitude Extraction: The electrodynamic frequency experienced by the cable under the action of power frequency current (50Hz) is twice the power supply frequency. Therefore, the target characteristic frequency is locked at... Based on the sampling theorem and the relationship between frequency resolution, calculate the spectral index corresponding to 100Hz. :

[0044] In the formula: This represents the sampling frame rate of the video capture. Extract the spectral magnitude at this index. As the single-frequency vibration intensity of this region under electrodynamic drive, the vibration amplitude of different ROI regions at 100Hz is denoted as follows: .

[0045] 2) Outlier Removal: Considering that visual measurements may be affected by sudden changes in local lighting or surface contamination in individual areas, leading to abnormal deviations in some measurements, a "truncated averaging method" is used to clean the five frequency domain feature values ​​to improve the robustness of the evaluation results. First, the extracted feature amplitude set... Sort the values ​​in ascending order to obtain an ordered sequence:

[0046] Then, remove the maximum value from the sequence. and minimum value This is to eliminate the interference of extreme outliers on the overall assessment.

[0047] 3) Feature Fusion and Final Index Generation: The arithmetic mean of the remaining three intermediate feature values ​​is calculated to obtain the breathing characteristic amplitude characterizing the overall aging state of the cable. The calculation formula is as follows:

[0048] In the formula: The second to fourth characteristic amplitudes retained after sorting. The final characteristic index after fusion processing, in micrometers (µm). ).

[0049] Step S5: Construct a "Vibration-Aging-Current Carrying Capacity" evaluation model. This model employs the Support Vector Regression (SVR) algorithm, based on a fingerprint database containing cable samples with different aging cycles, and utilizes the Radial Basis Function (RBF) to deeply mine the highly nonlinear relationship between vibration amplitude and insulation aging degree (and the corresponding current carrying capacity reduction ratio). The results obtained in Step 4... Input the model, output the normalized predicted carrying capacity correction coefficient. The SVR model construction and prediction principle is as follows: Figure 5 As shown. The specific process is as follows: 1) Construction of a standardized fingerprint database: In order to ensure the generalization ability of the model, a sample database containing multi-dimensional aging features is first constructed.

[0050] 1. Sample Preparation: Short cable samples of the same type as the cable to be tested were selected and placed in a high-temperature aging chamber for artificial accelerated thermal aging experiments. Accelerated aging parameters were set according to the Arrhenius equation:

[0051] In the formula For lifespan, The aging temperature is set to [value]. The activation energy is 1.15 eV. By controlling the heating time, gradient aging samples with equivalent operating years of 0, 10, 20, and 30 years were prepared.

[0052] Input feature (x): In a shielded room, 50% of the rated current is applied to each of the above-mentioned aging samples, and the breathing characteristic amplitude of each sample cable is measured using steps 1-4 of the present invention. .

[0053] Output label (y): Perform thermal elongation tests (IEC 60811-2-1 standard) on each aged sample to measure the elongation at break retention of the insulation layer under high-temperature load. Calculate the maximum permissible current carrying capacity of each sample under the current insulation condition according to IEC 60287 standard. and the rated current carrying capacity of the cable And define the current carrying capacity correction factor. , as the "truth value" label of the model.

[0054] 2. Normalization Mapping of Feature Space: Due to the scale difference between the input vibration amplitude and the output correction coefficient of the cable vibration, Min-Max linear normalization is performed on the input feature x and the output label y to accelerate gradient descent convergence and improve the accuracy of kernel function calculation.

[0055] Map all input and output data of training samples to the dimensionless interval [0, 1].

[0056] 2) Constructing an SVR model incorporating slack variables: The normalized low-dimensional nonlinear data is mapped to a high-dimensional feature space using the Gaussian radial basis function (RBF kernel). To achieve an optimal balance between model complexity and fitting error, the following convex quadratic programming problem is constructed: Objective function:

[0057] in, Characterize the smoothness of the model (structural risk). Characterize empirical risk (training error).

[0058] Constraints:

[0059] In the formula, As a penalty factor, For insensitive loss band width, , These are slack variables.

[0060] Introducing Lagrange multipliers By utilizing the KKT conditions, the primal problem can be transformed into a dual problem, allowing for the introduction of a kernel function and a reduction in computational complexity. Dual objective function:

[0061] Dual constraints:

[0063] In the formula, For Lagrange multipliers, Here, m represents the normalized input feature vector, and m is the total number of training samples. For the normalized true label values, the Gaussian radial basis function is: , For kernel parameters, This is a penalty factor.

[0064] Based on the optimal solution of the dual variables obtained by solving The final SVR regression decision function is:

[0065] In the formula, This is the predicted value of the cable current carrying capacity correction factor after normalization; This is the normalized feature vector of the sample to be tested; These are the normalized support vectors of the training set.

[0066] 4) Global optimization of SVR hyperparameters based on genetic algorithm (GA): Although the dual form simplifies the computation, the predictive performance of the model still highly depends on the penalty factor C determining the dual variable. (upper bound constraints) and kernel parameters (This determines the mapping structure of the feature space). Therefore, a real-valued genetic algorithm (GA) is used, employing these two key hyperparameters as chromosome genes to search for the global optimum. The specific implementation steps are as follows: 1. Population initialization and gene encoding: The size of the random generation is The initial population. Each individual (chromosome) consists of two gene loci, each representing a hyperparameter. and Considering the large range of parameter values, a real logarithmic encoding method is adopted to map the search space to a continuous real number field to ensure search accuracy.

[0067] 2. Construction of the Fitness Function: Fitness is the sole criterion for evaluating the quality of an individual. For every individual in a population... Substitute it into the SVR model and use the training set for 5-Fold Cross Validation.

[0068] Calculate the root mean square error (RMSE) under cross-validation and define the fitness function. The reciprocal of RMSE:

[0069] In the formula, For real labels, To validate the predicted values, To prevent small constants with a denominator of zero (take...) Fitness The larger the value, the higher the model prediction accuracy under that set of parameters.

[0070] 3. Genetic evolutionary operations: Based on the principle of survival of the fittest, the following operator operations are performed on the population to iteratively generate the next generation: Selection operator: The Roulette Wheel Selection method is used, where individuals with higher fitness have a greater probability of being selected and passed on to the next generation.

[0071] Crossover operator: Sets the crossover probability. Arithmetic crossover is used to linearly combine selected parent chromosomes to generate new offspring genes, thereby enhancing the population's local search capability.

[0072] Mutation operator: Sets the mutation probability. Non-uniform Gaussian mutation is employed, which superimposes a normally distributed random perturbation onto the gene values ​​to maintain population diversity and prevent premature convergence of the algorithm.

[0073] 4. Termination conditions and optimal solution output: The maximum number of evolution generations is set to The algorithm terminates when the maximum number of iterations is reached or the optimal fitness of the population no longer significantly improves after 10 consecutive generations. The output is the parameter combination corresponding to the individual with the highest fitness at this point. This parameter is used as the globally optimal hyperparameter. Using the optimal parameter combination and normalized training data, the final SVR decision function model is trained.

[0074] Step S6: Calculation of current carrying capacity correction factor 1) Isodistribution normalization of input data: Normalize the respiratory characteristic amplitude output in real time from step 4. As the input to be predicted .right Perform Min-Max normalization to map it to the same [0, 1] feature space as the training set:

[0075] In the formula, and These are the minimum and maximum values ​​of all amplitude samples in the fingerprint database, respectively.

[0076] 2) Model Inference and Output Denormalization: Normalized output is denormalized... Input the trained SVR model to calculate the predicted value. By utilizing the statistics of the current carrying capacity correction coefficient label Y in the fingerprint database, the prediction results are inversely transformed to restore the actual correction coefficient with physical meaning. :

[0077] In the formula, and These are the minimum and maximum values ​​of all amplitude samples in the historical fingerprint database (training set), respectively.

[0078] Step S7: Final assessment of the actual current-carrying capacity of the aging cable.

[0079] 1) Calculate the actual usable current-carrying capacity under the current aging state by combining the cable's factory nameplate parameters and environmental monitoring data. :

[0080] In the formula, The rated current carrying capacity of the cable. This is a correction factor for ambient temperature. If... If the load is lower than the current operating load, an overload warning will be triggered.

[0081] 2) Decision output and early warning.

[0082] The system will With respect to the actual operating load current of the cable Compare: like Output "Status is good"; like The system outputs a "warning status" and recommends limiting load growth. like This immediately triggers an "overload alarm".

[0083] The above is only one specific implementation method of the present invention, but the design concept of the present invention is not limited thereto. Any non-substantial modifications made to the present invention using this concept shall be considered as infringing the protection scope of the present invention.

Claims

1. A method for evaluating the current-carrying capacity of aging cables based on micro-displacement identification, characterized in that, Includes the following steps: S1. Obtain a sequence of video images of the power cable to be evaluated in its operating state; S2. Magnification of minute displacements in the video image sequence of the cable under test: The video image sequence is processed using a phase-based Euler video magnification algorithm to amplify the minute displacements of the power cable under test, resulting in an enhanced video sequence; S3. Optical flow calculation and feature extraction: Several regions of interest are defined in the enhanced video sequence, and the temporal vibration features of each region are calculated and extracted using the Horn-Schunck optical flow algorithm; the regions of interest are at least 5 non-overlapping regions of the same specification distributed along the axial direction of the power cable to be evaluated. S4. Multi-source feature fusion and noise suppression based on frequency domain analysis: The time-domain vibration characteristics of each region are subjected to discrete fast Fourier transform and converted into frequency domain spectrum. The frequency domain spectrum data is cleaned based on the truncated average method to obtain feature values. The remaining feature values ​​are calculated by arithmetic average to obtain the breathing characteristic amplitude that characterizes the overall aging state of the power cable to be evaluated. S5. Constructing a "Vibration-Aging-Current Carrying Capacity" Mapping Model: Based on a pre-established sample database, the Support Vector Regression (SVR) algorithm is used to establish a nonlinear mapping relationship between the vibration characteristic amplitude and the cable aging state and current carrying capacity correction coefficient, and the final SVR decision function model is obtained through optimization. S6. Calculation of Current Carrying Capacity Correction Factor: Input the breathing characteristic amplitude obtained in step 4 into the final SVR decision function model, and output the actual correction factor for the current carrying capacity of the power cable to be evaluated. ; S7. Final assessment of current carrying capacity: Based on the cable's design rated current carrying capacity and the calculated correction factor, calculate the actual usable current carrying capacity of the aging cable and output the assessment results.

2. The method for evaluating the current-carrying capacity of aging cables based on micro-displacement identification as described in claim 1, characterized in that, In step S1, a high-speed industrial camera is fixed on a tripod and taken at a distance of 2 meters from the cable surface to capture video of the cable to be evaluated, thereby obtaining a video image sequence of the cable to be evaluated.

3. The method for evaluating the current-carrying capacity of aging cables based on micro-displacement identification as described in claim 1, characterized in that, In step S2, the Euler video upscaling algorithm steps are as follows: (a) Complex and Controllable Pyramid Decomposition Each frame of the video image sequence is subjected to complex and controllable pyramid decomposition to obtain the local amplitude spectrum and local phase spectrum of the video image, ensuring that local small phase processing is equivalent to local motion processing: For the input video frame It is decomposed into high-pass residuals, complex bandpass subbands of multiple scales and directions, and low-pass residuals; the decomposition mathematical model is expressed as: ； in, The high-pass residual contains extremely fine texture edges or high-frequency noise information from the sensor that exceed the highest decomposition scale in the image. The low-pass residual contains the fundamental frequency illumination distribution and large-scale contour information of the image; In order to scale and direction The bandpass complex subband signal; x represents the horizontal spatial coordinate of the image pixel; y represents the vertical spatial coordinate of the image pixel; t represents time; For each complex subband signal This can be further expressed in the form of local amplitude and local phase: ； in, The local amplitude spectrum characterizes the image texture intensity; It is a local phase spectrum, and its changes on the time axis directly correspond to the minute local motions of the image; (b) Bandpass filtering and phase amplification For the local phase spectrum obtained by decomposition A bandpass filter is applied in the time dimension to filter out high-frequency noise and DC drift; let the bandpass filter be... The magnification factor is The phase change that contains vibration characteristics and the magnified new phase spectrum The calculation is as follows: ； In the formula: This represents temporal convolution operations; Used to linearly enhance the amplitude of weak vibrations, making them visually recognizable; (c) Video reconstruction and compositing Using the original local amplitude spectrum and the new phase spectrum after amplification Construct the amplified complex subband signal: ； To maintain the image clarity and background stability of the reconstructed video, phase correction is performed only on the bandpass subband signal, while the high-pass residual is not corrected. and low-pass residual Keep the original values ​​unchanged; the final enlarged video sequence The signal is obtained by performing a complex and controllable inverse pyramid transform on the unprocessed residual and the amplified subband signal, and then superimposing and reconstructing the result. ； The above reconstruction process generates the final enlarged video sequence. , As an enhanced video sequence.

4. The method for evaluating the current-carrying capacity of aging cables based on micro-displacement identification as described in claim 1, characterized in that: The specific steps of step S3 are as follows: S3.1) Select at least five non-overlapping areas along the axial direction of the power cable to be evaluated as observation windows, wherein the observation windows cover the surface texture area of ​​the cable's outer sheath; these are respectively marked as... ,..., Together, they characterize the overall "breathing" state of the power cable being evaluated. S3.2) Global Optimization Solution: For each observation window, the Horn-Schunck optical flow algorithm is used to construct an energy functional that includes constant brightness constraints and smoothness constraints. E : ； In the formula: These are the optical flow velocity components of the pixel in the horizontal and vertical directions, respectively; Image brightness exist Partial derivatives in the spatial direction and in the time direction t; first term This is a constant brightness constraint term, used to ensure that the brightness remains unchanged before and after pixel movement; Second item This is a smoothness constraint term used to constrain the consistency of motion in neighboring pixels; These are the smoothing weight coefficients, used to adjust the strength of the smoothing constraints; The symbol for double integral indicates that the energy function is integrated and summed over the entire image domain; Let be the differential area element in spatial coordinates, representing the horizontal coordinate of the image plane as the integral variable. and vertical coordinates ; To achieve a numerical solution for the energy functional E, the Gauss-Seidel iterative algorithm is employed: First, based on the Euler-Lagrange equations, the energy functional minimization problem is transformed into solving the following system of linear equations: In the discretized mesh, the Laplacian operator is used to approximate the pixels. The deviation between the velocity component and its neighborhood average is represented by a smoothing term, and the specific iterative update formula is as follows: ； In the formula: k represents the current iteration number, and The first The pixel points obtained in the next iteration Optical flow velocity in the horizontal and vertical directions; For smoothing weighting coefficients; , For pixels The average velocity within the neighborhood is calculated using a 4-neighborhood weighted average, defined as: ； Repeat the above calculation process until the error between two consecutive iterations is less than a preset threshold. , Get each pixel within the window Optimal optical flow velocity vector x, y, t, u, and v represent the horizontal coordinate of a pixel, the vertical coordinate of a pixel, time, the horizontal optical flow velocity component, and the vertical optical flow velocity component, respectively. S3.3) Spatial Domain Feature Aggregation: Introducing a spatial averaging operator, the optical flow velocity vector magnitude of all effective pixels within each observation window is integrally averaged to obtain the instantaneous average motion intensity of that region in the current frame, as shown in the following formula: ； in, For the k-th observation window region, Let k be the area of ​​the observation window regions, i.e., the total number of pixels; The instantaneous motion magnitude of a pixel; For pixels exist The horizontal optical flow velocity component at any given time. For pixels exist The vertical optical flow velocity component at any given moment; Let be the vibration amplitude sequence of the k-th region over time; through the above calculations, the vibration amplitude sequences of each observation window distributed along the axial direction of the cable surface are finally obtained, denoted as . ,..., ..., As the temporal vibration characteristics of each region.

5. The method for evaluating the current-carrying capacity of aging cables based on micro-displacement identification as described in claim 1, characterized in that: The specific steps of step S4 are as follows: S4.1) DC removal and windowing: First, subtract the mean of all time-domain vibration characteristics from each time-domain vibration characteristic to eliminate DC bias. Then, process the DC-removed signal... Adding a Hanning window, the signal after adding a Hanning window The calculation is as follows: ； In the formula, N represents the total number of sampling points within the observation window, and n represents the sequence number of the current sampling point, with a value range of... ; S4.2) Discrete Fourier Transform: For the preprocessed signal Perform a Discrete Fourier Transform to map from the time domain to the frequency domain; the transform formula is: ； In the formula, Let be the complex spectral value at the k-th frequency point, where j is the imaginary unit; S4.3) Amplitude Extraction: The electrodynamic frequency experienced by the cable under a 50Hz power frequency current is twice the power supply frequency; therefore, the target characteristic frequency is... Locked to ; Based on the sampling theorem and the relationship between frequency resolution, calculate the spectral index corresponding to 100Hz. : ； In the formula: Extract the spectral magnitude at this index, based on the sampling frame rate of the video capture. As the single-frequency vibration intensity under electrodynamic drive in this region, the vibration amplitude at 100Hz for different observation windows is denoted as follows: , As a frequency domain eigenvalue; Indicates taking the integer part; S4.4) Outlier Removal: The frequency domain feature values ​​are cleaned using the "truncated average method": First, the extracted feature amplitude set is... Sort the values ​​in ascending order to obtain an ordered sequence: Then, the maximum and minimum values ​​in the ordered sequence are removed to eliminate the interference of extreme outliers on the overall evaluation; 3) Feature Fusion and Final Index Generation: The remaining intermediate feature amplitudes are arithmetically averaged to obtain the breathing characteristic amplitude, which characterizes the overall aging state of the power cable to be evaluated. The calculation formula is as follows: ； In the formula: The j-th feature magnitude is retained after sorting. The unit for is micrometers. .

6. The method for evaluating the current-carrying capacity of aging cables based on micro-displacement identification as described in claim 1, characterized in that: The specific steps of step S5 are as follows: S5.1) Construction of a standardized fingerprint database: Construct a sample database containing multi-dimensional aging features: S5.1.1) Sample Preparation: Select a short sample cable of the same type as the cable to be tested, place it in a high-temperature aging chamber for artificial accelerated thermal aging experiment, and set the accelerated aging parameters according to the Arrhenius equation: ； In the formula For lifespan, This is the aging temperature. For activation energy, k Using Boltzmann's constant as an example, gradient aging samples with equivalent operating years of 0, 10, 20, and 30 years were prepared by controlling the heating time. Input feature x: In a shielded room, apply 50% of the rated current to each of the above-mentioned aging samples, and measure the breathing characteristic amplitude of each sample cable using the methods in steps S1-S4. ; As input features; Output label y: Perform thermal elongation tests on each aged sample to measure the elongation at break retention rate of the insulation layer under high-temperature load. Calculate the maximum permissible current carrying capacity of each sample under the current insulation condition according to IEC 60287 standard. and the rated current carrying capacity of the cable ; And define the current carrying capacity correction factor. , as a "truth value" label for the model; S5.1.2) Normalization mapping of feature space: Perform Min-Max linear normalization on the input feature x and output label y to obtain low-dimensional nonlinear data: ； Map all input and output data of training samples to the dimensionless interval [0, 1]; where X represents the set of input features x and Y represents the set of output labels y; This represents the i-th input feature after normalization; This represents the i-th output label after normalization; min indicates taking the minimum value, and max indicates taking the maximum value. S5.2) Constructing an SVR model with slack variables: Using the Gaussian radial basis kernel function, the normalized low-dimensional nonlinear data is mapped to a high-dimensional feature space. In order to achieve the optimal balance between "model complexity" and "fitting error", the following convex quadratic programming problem is constructed: Objective function: ； in, Characterizes the smoothness of the model. Characterize empirical risk; The weight vector determines the orientation of the hyperplane in the feature space. This represents the bias term, which determines the position of the hyperplane's intercept relative to the origin. Representing the One sample point exceeds the upper bound of the insensitive loss band; Representing the One sample point exceeds the lower bound of the insensitive loss band; Constraints: ； In the formula, As a penalty factor, The insensitive loss band width, where T denotes matrix transpose. This represents the high-dimensional feature vector after mapping. Indicates the first The true label values ​​of each training sample; Introducing Lagrange multipliers By utilizing the KKT conditions, the primal problem can be transformed into a dual problem, allowing for the introduction of a kernel function and a reduction in computational complexity. Dual objective function: ； Dual constraints: ； In the formula, For the first A Lagrange multiplier, No. Lagrange multiplication of training samples , are the i-th and j-th input feature vectors after normalization, respectively, and m is the total number of training samples. For the normalized i-th true label value, the Gaussian radial basis function is: , For kernel parameters, As a penalty factor; Based on the optimal solution of the dual variables obtained by solving The final SVR regression decision function is: ； In the formula, This is the predicted value of the cable current carrying capacity correction factor after normalization; This is the normalized feature vector of the sample to be tested; These are the normalized support vectors of the training set; S5.3) Global optimization of SVR hyperparameters based on genetic algorithm: A real-number encoded genetic algorithm is used to input two key hyperparameters. and As a chromosome gene, the global optimal solution is searched; the specific implementation steps are as follows: S5.3.1) Population initialization and gene encoding: The size of the random generation is The initial population, each chromosome consists of two gene loci, each representing a hyperparameter. and The search space is mapped to a continuous real number field using a real number logarithmic encoding method to ensure search accuracy. S5.3.2) Fitness function construction: For each individual in the population Substitute the data into the SVR model and perform 5-fold cross-validation using the training set. Calculate the root mean square error under cross-validation and define the fitness function. The reciprocal of RMSE: ； In the formula, For real labels, To validate the predicted values, To prevent tiny constants with a denominator of zero; fitness The larger the value, the higher the model prediction accuracy under that set of parameters; S5.3.3) Genetic evolution operations: Based on the principle of survival of the fittest, the following operator operations are performed on the population to iteratively generate the next generation: Selection operator: The roulette wheel selection method is used, and individuals with higher fitness have a greater probability of being selected to pass on their genes to the next generation; Crossover calculation: Setting crossover probabilities Arithmetic crossover is used to linearly combine selected parent chromosomes to generate new offspring genes, thereby enhancing the local search capability of the population. Mutation operator: Sets the mutation probability Non-uniform Gaussian mutation is used to superimpose random perturbations that conform to a normal distribution on gene values ​​in order to maintain population diversity and prevent premature convergence of the algorithm. S5.3.4) Termination condition and optimal solution output: The maximum number of evolution generations is set to The algorithm terminates when the number of iterations reaches the upper limit or the optimal fitness of the population no longer significantly improves after 10 consecutive generations; it outputs the parameter combination corresponding to the individual with the highest fitness at this point. We use these parameters as the globally optimal hyperparameters; and then train the final SVR decision function model using the optimal parameter combination and the normalized training data.

7. The method for evaluating the current-carrying capacity of aging cables based on micro-displacement identification as described in claim 1, characterized in that: The specific steps of step S6 are as follows: S6.1) Isodistribution normalization of input data: Normalize the respiratory characteristic amplitude output in real time from step 4. As the input to be predicted ,right Perform Min-Max normalization to map it to the same [0, 1] feature space as the training set: ； In the formula, and These are the minimum and maximum values ​​of all amplitude samples in the fingerprint database, respectively. S6.2) Model Inference and Output Denormalization: Denormalize the normalized output... Input the final SVR decision function model to calculate the predicted value. ;Utilize the carrying capacity correction factor labels in the fingerprint database Y The statistical measure is used to perform an inverse transformation on the prediction results, restoring them to the actual correction coefficients with physical meaning. : ； In the formula, and These are the minimum and maximum values ​​of all amplitude samples in the historical fingerprint database, respectively.

8. The method for evaluating the current-carrying capacity of aging cables based on micro-displacement identification as described in claim 1, characterized in that: The specific steps of step S7 are as follows: S7.1) Based on the manufacturer's nameplate parameters and environmental monitoring data of the power cable to be evaluated, calculate the actual usable current-carrying capacity of the power cable under its current aging state. : ； In the formula, The rated current carrying capacity of the cable. This is the ambient temperature correction factor. If the load falls below the current operating load, an overload warning will be triggered. S7.2) Decision Output and Early Warning: Will Compared with the current actual operating load current of the power cable to be evaluated Compare: like Output "Status is good"; like The system outputs a "warning status" message, suggesting that load growth be limited. like This immediately triggers an "overload alarm".

9. A system for assessing the current-carrying capacity of aging cables based on micro-displacement identification, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the method for evaluating the current-carrying capacity of aging cables based on micro-displacement identification as described in any one of claims 1 to 8.

Images