Cable insulation aging state evaluation method based on operation data analysis

Abstract

The invention relates to a cable insulation aging state evaluation method based on operation data analysis. The method comprises the following steps: S1, acquiring cable operation data in real time through a heterogeneous sensor array; s2, adopting a layered preprocessing strategy on the edge side; s3, constructing a hybrid deep learning model; s4, outputting by adopting an improved evidence theory fusion model; s5, outputting a fourth-level aging state label and a confidence evidence theory fusion model; electrical parameters and non-electrical parameters of the cable are collected in real time through the heterogeneous sensor array, and multi-dimensional data such as partial discharge signals, three-phase current harmonic distortion rate, distributed optical fiber temperature, mechanical vibration acceleration and environment dew point temperature are covered. The multi-source data fusion mode can comprehensively reflect the operation state of cable insulation, the limitation of single parameter evaluation is avoided, and the accuracy of the evaluation result is greatly improved.

Description

Technical Field

[0001] This invention belongs to the field of cable monitoring technology, specifically a method for assessing the aging status of cable insulation based on operational data analysis. Background Technology

[0002] In power systems, cables serve as a crucial carrier for power transmission, and their insulation condition directly impacts the safety and stability of power supply. As cables age, insulation aging becomes increasingly prominent, potentially leading to cable failures, power outages, and even severe economic losses and safety hazards.

[0003] Traditional methods for assessing the aging condition of cable insulation often rely on a single testing method or a limited set of parameters, such as judging the insulation condition solely by measuring insulation resistance. This method has many limitations and cannot comprehensively and accurately reflect the actual aging condition of cable insulation. Furthermore, a single parameter is easily affected by external interference, leading to inaccurate assessment results. On the other hand, it cannot fully consider various influencing factors during cable operation, such as temperature changes, mechanical vibration, and environmental conditions, making it difficult to achieve dynamic assessment and real-time monitoring of the aging condition of cable insulation.

[0004] Therefore, the present invention provides a method for evaluating the aging status of cable insulation based on operational data analysis. Summary of the Invention

[0005] In order to overcome the shortcomings of the prior art, at least one technical problem raised in the background art is solved.

[0006] The technical solution adopted by this invention to solve its technical problem is: a method for evaluating the aging state of cable insulation based on operational data analysis, comprising: S1. Real-time acquisition of cable operation data, including electrical and non-electrical parameters, through a heterogeneous sensor array: S2. Employ a layered preprocessing strategy at the edge: First layer: Process partial discharge signals based on an improved wavelet threshold denoising algorithm, retaining frequency bands with an energy ratio of ≥8%; The second layer: eliminates the latency of multi-source data acquisition through the spatiotemporal alignment module, with a synchronization error of ≤1ms; S3. Construct a hybrid deep learning model, including: Parallel operation of a three-dimensional convolutional neural network and a gated recurrent unit, wherein the 3D-CNN processes the time-frequency-spatial three-dimensional features of partial discharge signals, and the GRU analyzes the nonlinear temporal correlation of temperature and vibration data; The model enhances the training data through an adversarial generative network to generate virtual samples under extreme conditions. S4, Adopting an improved method The output of the evidence theory fusion model includes: The credibility weight of each piece of evidence is dynamically calculated, and the weight update frequency is positively correlated with the cable load change rate. Bayesian network recalibration is triggered when the conflict factor K > 0.5; S5. Output four-level aging status labels and confidence levels: The four aging status labels include healthy, warning, critical, and faulty; The confidence level is calculated using Bootstrap sampling, requiring a 95% confidence interval width ≤ 0.1.

[0007] A further improvement to the technical solution of the present invention is that the method also includes an abnormal operating condition backtracking function: The cable insulation health index HI mutation point in the stored historical data is... Full-dimensional data from the first and last 10 minutes; The causal reasoning model is used to locate the causes of mutations. The causes are classified as follows: lightning overvoltage, mechanical damage, and insulation hydrolysis.

[0008] A further improvement to the technical solution of this invention lies in that: the electrical parameters acquired in step S1 include the time-frequency characteristics of the partial discharge signal in the frequency band of 50kHz-1MHz, the three-phase current harmonic distortion rate, and the dielectric loss tangent. The partial discharge signal acquisition employs a differential sensor array, and the spatial arrangement satisfies… Sampling theorem: The distance between adjacent sensors ≤ the shortest wavelength of the signal / 2; Non-electrical parameters include distributed fiber temperature with a spatial resolution ≤1m, mechanical vibration acceleration in the 0.1-10kHz frequency band, and ambient dew point temperature.

[0009] A further improvement to the technical solution of this invention lies in the following: the improved wavelet threshold denoising algorithm in step S2 includes: The threshold is determined adaptively, and the threshold calculation formula is as follows: ; in, The standard deviation of noise. For signal length, The calculated wavelet threshold; Nonlinear compression is applied to the high-frequency coefficients using the following formula: ; in, These are the high-frequency wavelet coefficients after nonlinear compression processing. These are the original high-frequency wavelet coefficients. To obtain The larger of the two values ​​is 0, i.e., if If the value is greater than 0, then the value is [value]. ;if If ≤0, then the value is 0; A further improvement to the technical solution of the present invention is that the GAN data augmentation in step S3 includes: The generator takes the latent space encoding of the actual samples as input and outputs the following: ; in, Represents actual samples With generated samples The 2-norm distance between two samples is used to measure the degree of difference between them. This is a sample of actual cable operation data collected. These are virtual samples generated by a generator to augment training data and simulate data conditions under extreme circumstances. The standard deviation of the actual sample reflects the dispersion of the actual sample data. The discriminator output simultaneously uses Distance optimization, gradient penalty coefficient λ=10 A further improvement to the technical solution of this invention lies in that: the 3D-CNN in step S3 adopts a multi-scale feature extraction structure, including: First convolutional layer: kernel size 7×7×7, stride 2, extracts macroscopic defect features; Second convolutional layer: kernel size 3×3×3, stride 1, capturing microscopic discharge pulse clusters.

[0010] A further improvement to the technical solution of this invention is that the confidence weight calculation in step S4 includes: The weights of electrical parameters are expressed by the following formula: ,in, This indicates the weight of electrical parameters in the cable insulation aging assessment. This represents the current operating time of the cable, in years. The weights of non-electrical parameters are expressed by the following formula: ,in, This indicates the weight of non-electrical parameters in the cable insulation aging assessment.

[0011] A further improvement to the technical solution of this invention lies in the fact that, in the evidence theory fusion of step S4, the basic probability allocation function satisfies: ; in, It is the basic probability assignment function, which represents the allocation of probabilities to the body of evidence. The degree of support for the assessment results is used to measure the extent to which different pieces of evidence support the assessment results in the fusion of evidence theories for cable insulation aging assessment.

[0012] For the first In cable insulation aging assessment, this type of evidence includes different types of evidence such as electrical parameter evidence, temperature evidence, and vibration data evidence.

[0013] For the first Cross-entropy loss is a measure of the difference between the predicted outcome and the actual outcome of evidence. The smaller the loss, the more accurate the prediction. The temperature coefficient acts as a moderating element in the formula, affecting the distribution of probability allocation. =0.1, It is about all the evidence. The summation is used to normalize the probability distributions, so that the sum of the probability distributions of all evidence is 1.

[0014] A further improvement of the technical solution of the present invention is that: in step S5, a health tag is output when the cable insulation health index HI≥0.8, a warning tag is output when 0.6≤HI<0.8, a critical tag is output when 0.4≤HI<0.6, and a fault tag is output when HI<0.4.

[0015] A further improvement to the technical solution of the present invention lies in the following: after the aging status label is generated in step S5, further processing is performed: When the system is assessed as being in a critical state three times in a row, the life prediction module of the digital twin system is automatically triggered. When the mean square error between the predicted result and the measured cable insulation health index HI exceeds 0.05, online model fine-tuning is triggered.

[0016] The beneficial effects of this invention are as follows: 1. Real-time acquisition of electrical and non-electrical parameters of the cable via a heterogeneous sensor array, covering multi-dimensional data such as partial discharge signals, three-phase current harmonic distortion rate, dielectric loss tangent, distributed fiber temperature, mechanical vibration acceleration, and ambient dew point temperature. This multi-source data fusion method comprehensively reflects the operating status of cable insulation, avoiding the limitations of single-parameter evaluation and greatly improving the accuracy of evaluation results. 3D-CNN can process the time-frequency-space three-dimensional features of partial discharge signals, and GRU can analyze the nonlinear temporal correlation of temperature and vibration data. Simultaneously, adversarial generative networks are used to enhance training data and generate virtual samples under extreme conditions, improving the model's generalization ability and adaptability. Furthermore, an improved evidence theory is used to fuse the model output, dynamically calculating the credibility weights of each piece of evidence. The weight update frequency is positively correlated with the cable load change rate, allowing for reasonable allocation of weights for each parameter based on the actual operating conditions of the cable. When the conflict factor K > 0.5, Bayesian network recalibration is triggered, ensuring the reliability and stability of the evaluation results. 2. Output four levels of aging status labels and confidence levels. Labels include healthy, warning, critical, and fault. Confidence levels are calculated using Bootstrap sampling, requiring a 95% confidence interval width ≤ 0.1 to make the assessment results intuitive and easy to understand, facilitating timely action by maintenance personnel. Simultaneously, when three consecutive assessments indicate a critical state, the life prediction module of the digital twin system is automatically triggered. When the mean square error between the predicted result and the measured cable insulation health index HI exceeds 0.05, online model fine-tuning is triggered, enabling continuous monitoring of cable insulation status and dynamic model optimization. Furthermore, it stores comprehensive data for 10 minutes before and after the HI mutation point in historical data and uses a causal reasoning model to pinpoint the mutation cause, aiding in the analysis of cable insulation aging and providing support for cable fault prevention. Attached Figure Description

[0017] Figure 1 This is a flowchart of the evaluation method of the present invention. Detailed Implementation

[0018] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0019] Please see Figure 1 , This embodiment provides: a method for assessing the aging state of cable insulation based on operational data analysis, including: S1: Data Acquisition Cable operation data is acquired in real time using a heterogeneous sensor array. Electrical parameters include the time-frequency characteristics of partial discharge signals in the 50kHz-1MHz frequency band, three-phase current harmonic distortion rate, and dielectric loss tangent. A differential sensor array is used for partial discharge signal acquisition, and the spatial arrangement meets certain requirements. The sampling theorem states that the distance between adjacent sensors is ≤ the shortest wavelength of the signal / 2; non-electrical parameters include distributed optical fiber temperature with a spatial resolution ≤ 1m, mechanical vibration acceleration in the 0.1-10kHz frequency band, and ambient dew point temperature.

[0020] For example, on a 110kV high-voltage cable, a differential sensor array is installed to collect partial discharge signals, while distributed fiber optic temperature sensors, mechanical vibration acceleration sensors, and ambient dew point temperature sensors are deployed to acquire various data during the cable's operation in real time.

[0021] It should be noted that the method for obtaining the three-phase current harmonic distortion rate is as follows: The three-phase current signal is acquired by using a current sensor, and then the time domain signal is converted to the frequency domain using signal processing techniques such as fast Fourier transform. The fundamental wave and each harmonic are separated, and the ratio of the root mean square value of each harmonic content to the root mean square value of the fundamental wave component is calculated according to the formula, thereby obtaining the harmonic distortion rate of the three-phase current. Method for obtaining the dielectric loss tangent: Using a dielectric loss tester based on the Schering bridge principle, an AC voltage is applied to the cable, and the current passing through the insulating medium and the phase difference between the voltage and the current are measured. The dielectric loss tangent is then calculated based on the relationship between active power and reactive power.

[0022] Furthermore, the acquisition of multi-source data ensures comprehensive monitoring of the cable's operating status. Different types of data complement each other, enabling more accurate detection of early signs and potential problems of cable insulation aging.

[0023] S2: Edge-side preprocessing A layered preprocessing strategy is adopted at the edge.

[0024] The first layer: Processing partial discharge signals based on an improved wavelet thresholding denoising algorithm. First, the threshold is adaptively determined; the threshold calculation formula is: ; in, The standard deviation of noise. For signal length, The calculated wavelet threshold; Nonlinear compression is applied to the high-frequency coefficients using the following formula: ; in, These are the high-frequency wavelet coefficients after nonlinear compression processing. These are the original high-frequency wavelet coefficients. To obtain The larger of the two values ​​is 0. Through these processes, frequency bands with an energy percentage of ≥8% are retained.

[0025] For example: Suppose a partial discharge signal is acquired, and the noise standard deviation is calculated. =0.3, signal length =500.

[0026] First calculate =ln500≈6.215.

[0027] Then ≈12.43, ≈3.526.

[0028] Finally, calculate the threshold. =0.3 × 3.526 = 1.0578, this calculated threshold This will be used to determine whether wavelet coefficients are noise coefficients, and thus decide whether to process them. Suppose a certain high-frequency wavelet coefficient =1.5, the threshold calculated earlier =1.0578.

[0029] first, =1 (because) >0); then, =1.5-1.0578=0.4422>0, therefore =0.4422.

[0030] at last, =1×0.4422=0.4422, which means that when the absolute value of the high-frequency wavelet coefficients is greater than the threshold, the compression process using this formula retains some effective information and removes some components that may be noise. If another high-frequency wavelet coefficient =0.8, =0.8-1.0578=-0.2578<0, then =0, =1×0=0 indicates that when the absolute value of the high-frequency wavelet coefficient is less than the threshold, the formula treats it as a noise coefficient and sets it directly to 0, thereby removing noise interference.

[0031] The second layer: eliminates the delay in multi-source data acquisition through the spatiotemporal alignment module, ensuring that the synchronization error is ≤1ms.

[0032] Furthermore, the first layer processes the partial discharge signal based on an improved wavelet threshold denoising algorithm, removing noise interference, retaining useful signal components, and improving the feature extraction accuracy of the signal; the second layer eliminates the acquisition delay of multi-source data through a spatiotemporal alignment module, ensuring the consistency of data in time and space, which facilitates comprehensive analysis.

[0033] S3: Building a Hybrid Deep Learning Model Construct a hybrid deep learning model of parallel-operating 3D convolutional neural networks and gated recurrent units.

[0034] 3D-CNN employs a multi-scale feature extraction structure. The first convolutional layer has a kernel size of 7×7×7 and a stride of 2, which is used to extract macroscopic defect features. The second convolutional layer has a kernel size of 3×3×3 and a stride of 1, which captures microscopic discharge pulse clusters.

[0035] GRU analysis of nonlinear temporal correlations between temperature and vibration data.

[0036] The model enhances the training data using a Generative Adversarial Network (GAN). The generator's input is the latent space encoding of the actual samples, and its output must satisfy the following: ; in, Represents actual samples With generated samples The 2-norm distance between two samples is used to measure the degree of difference between them. This is a sample of actual cable operation data collected. These are virtual samples generated by a generator to augment training data and simulate data conditions under extreme circumstances. The standard deviation of the actual sample reflects the dispersion of the actual sample data. For example, suppose the actual sample This is a set of cable temperature data, and its standard deviation was calculated. =1.5, the generator produced a virtual sample. Calculate the 2-norm distance between the two. =0.12, at this time =0.1×1.5=0.15. Since 0.12≤0.15, it means that the generated sample meets the requirements and can be used to augment the training data. The discriminator output simultaneously uses Distance optimization, gradient penalty coefficient λ=10.

[0037] Furthermore, the multi-scale feature extraction structure of 3D-CNN can effectively capture features of different scales in partial discharge signals, while GRU's ability to analyze time-series data helps predict the changing trend of cable insulation status over time. The virtual samples generated by GAN can simulate data under extreme working conditions, enabling the model to perform well under various conditions. This allows the hybrid model to more comprehensively and deeply mine information in the data, improving the ability to identify and predict the aging status of cable insulation.

[0038] S4: Output of the Improved Evidence Theory Fusion Model The output is obtained using an improved evidence theory fusion model.

[0039] The credibility weight of each piece of evidence is dynamically calculated, and the weight update frequency is positively correlated with the cable load change rate; the formula for calculating the electrical parameter weights is as follows: ,in, This indicates the weight of electrical parameters in the cable insulation aging assessment. The current operating time of the cable is in years; the formula for calculating the weights of non-electrical parameters is as follows: .

[0040] Bayesian network recalibration is triggered when the conflict factor K > 0.5.

[0041] The basic probability assignment function satisfies: ; in, It is the basic probability assignment function, which represents the allocation of probabilities to the body of evidence. The degree of support for the assessment results is used to measure the extent to which different pieces of evidence support the assessment results in the fusion of evidence theories for cable insulation aging assessment.

[0042] For the first In cable insulation aging assessment, this type of evidence includes different types of evidence such as electrical parameter evidence, temperature evidence, and vibration data evidence.

[0043] For the first Cross-entropy loss is a measure of the difference between the predicted outcome and the actual outcome of evidence. The smaller the loss, the more accurate the prediction. The temperature coefficient acts as a moderating element in the formula, affecting the distribution of probability allocation. =0.1, It is about all the evidence. The summation is used to normalize the probability distributions, so that the sum of the probability distributions of all evidence is 1.

[0044] Furthermore, when there are conflicts between data, Bayesian network recalibration can adjust the model in a timely manner to ensure the stability of the evaluation results. For example, dynamic weight adjustment can make more reasonable use of the information of each parameter based on the actual operation of the cable, and avoid the deviation of a single parameter from having a significant impact on the evaluation results.

[0045] S5: Output aging status label and confidence level Based on the calculated cable insulation health index HI, four levels of aging status labels and confidence levels are output. A cable insulation health index HI≥0.8 outputs a health label, 0.6≤HI<0.8 outputs a warning label, 0.4≤HI<0.6 outputs a critical label, and HI<0.4 outputs a fault label. The confidence level is calculated through Bootstrap sampling, requiring a 95% confidence interval width ≤0.1.

[0046] When the condition is assessed as critical for three consecutive times, the life prediction module of the digital twin system is automatically triggered; when the mean square error between the prediction result and the measured cable insulation health index HI exceeds 0.05, the model is triggered for online fine-tuning.

[0047] Furthermore, clear aging status labels and confidence levels enable maintenance personnel to promptly understand the insulation status of cables, take appropriate measures, and trigger subsequent processing mechanisms to achieve continuous monitoring of cable insulation status and dynamic optimization of the model, thereby improving the adaptability and accuracy of the assessment method.

[0048] It should be noted that the method also includes abnormal operating condition backtracking. The cable insulation health index HI mutation point in the stored historical data is... Full-dimensional data from the first and last 10 minutes; The causal reasoning model is used to locate the causes of mutations. The causes are classified as follows: lightning overvoltage, mechanical damage, and insulation hydrolysis.

[0049] For example, when a sudden change in the cable insulation health index is detected, the system automatically stores data for the relevant time period and uses a causal reasoning model to analyze the cause of the change, providing a reference for subsequent cable maintenance and fault prevention.

[0050] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely illustrative of the principles of the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the present invention as claimed. The scope of protection of the present invention is defined by the appended claims and their equivalents.

Claims

1. A method for assessing the aging state of cable insulation based on operational data analysis, characterized in that: include: S1. Real-time acquisition of cable operation data, including electrical and non-electrical parameters, through a heterogeneous sensor array: S2. Employ a layered preprocessing strategy at the edge: First layer: Process partial discharge signals based on an improved wavelet threshold denoising algorithm, retaining frequency bands with an energy ratio of ≥8%; The second layer: eliminates the latency of multi-source data acquisition through the spatiotemporal alignment module, with a synchronization error of ≤1ms; S3. Construct a hybrid deep learning model, including: Parallel operation of a three-dimensional convolutional neural network and a gated recurrent unit, wherein the 3D-CNN processes the time-frequency-spatial three-dimensional features of partial discharge signals, and the GRU analyzes the nonlinear temporal correlation of temperature and vibration data; The model enhances the training data through an adversarial generative network to generate virtual samples under extreme conditions. S4, Adopting an improved method The output of the evidence theory fusion model includes: The credibility weight of each piece of evidence is dynamically calculated, and the weight update frequency is positively correlated with the cable load change rate. Bayesian network recalibration is triggered when the conflict factor K > 0.5; S5. Output four-level aging status labels and confidence levels: The four aging status labels include healthy, warning, critical, and faulty; The confidence level is calculated using Bootstrap sampling, requiring a 95% confidence interval width ≤ 0.

1.

2. The method for assessing the aging state of cable insulation based on operational data analysis according to claim 1, characterized in that: The electrical parameters acquired in step S1 include the time-frequency characteristics of the partial discharge signal in the frequency band of 50kHz-1MHz, the three-phase current harmonic distortion rate, and the dielectric loss tangent. The partial discharge signal acquisition uses a differential sensor array, and the spatial arrangement meets the following requirements. Sampling theorem: The distance between adjacent sensors ≤ the shortest wavelength of the signal / 2; Non-electrical parameters include distributed fiber temperature with a spatial resolution ≤1m, mechanical vibration acceleration in the 0.1-10kHz frequency band, and ambient dew point temperature.

3. The method for assessing the aging state of cable insulation based on operational data analysis according to claim 1, characterized in that: The improved wavelet thresholding denoising algorithm in step S2 includes: The threshold is determined adaptively, and the threshold calculation formula is as follows: ; in, The standard deviation of noise. For signal length, The calculated wavelet threshold; Nonlinear compression is applied to the high-frequency coefficients using the following formula: ； in, These are the high-frequency wavelet coefficients after nonlinear compression processing. These are the original high-frequency wavelet coefficients. To obtain The larger of the two values ​​is 0, i.e., if If the value is greater than 0, then the value is [value]. ;if If ≤0, then the value is 0.

4. The method for assessing the aging state of cable insulation based on operational data analysis according to claim 1, characterized in that: The GAN data augmentation in step S3 includes: The generator takes the latent space encoding of the actual samples as input and outputs the following: ； in, Represents actual samples With generated samples The 2-norm distance between two samples is used to measure the degree of difference between them. This is a sample of actual cable operation data collected. These are virtual samples generated by a generator to augment training data and simulate data conditions under extreme circumstances. The standard deviation of the actual sample reflects the dispersion of the actual sample data. The discriminator output simultaneously uses Distance optimization, gradient penalty coefficient λ=10.

5. The method for assessing the aging state of cable insulation based on operational data analysis according to claim 1, characterized in that: The 3D-CNN in step S3 employs a multi-scale feature extraction structure, including: First convolutional layer: kernel size 7×7×7, stride 2, extracts macroscopic defect features; Second convolutional layer: kernel size 3×3×3, stride 1, capturing microscopic discharge pulse clusters.

6. The method for assessing the aging state of cable insulation based on operational data analysis according to claim 1, characterized in that: The confidence weight calculation in step S4 includes: The weights of electrical parameters are expressed by the following formula: ,in, This indicates the weight of electrical parameters in the cable insulation aging assessment. This represents the current operating time of the cable, in years. The weights of non-electrical parameters are expressed by the following formula: ,in, This indicates the weight of non-electrical parameters in the cable insulation aging assessment.

7. The method for assessing the aging state of cable insulation based on operational data analysis according to claim 1, characterized in that: In the evidence theory fusion of step S4, the basic probability allocation function satisfies: ； in, It is the basic probability assignment function, which represents the allocation of probabilities to the body of evidence. The degree of support, in the fusion of evidence theories for cable insulation aging assessment, is used to measure the extent to which different pieces of evidence support the assessment results. For the first This type of evidence, in cable insulation aging assessment, represents electrical parameter evidence, temperature evidence, and vibration data. For the first Cross-entropy loss is a measure of the difference between the predicted outcome and the actual outcome of evidence. The smaller the loss, the more accurate the prediction. The temperature coefficient acts as a moderating element in the formula, affecting the distribution of probability allocation. =0.1, It is about all the evidence. The summation is used to normalize the probability distributions, so that the sum of the probability distributions of all evidence is 1.

8. The method for assessing the aging state of cable insulation based on operational data analysis according to claim 1, characterized in that: In step S5, a health tag is output when the cable insulation health index HI≥0.8, a warning tag is output when 0.6≤HI<0.8, a critical tag is output when 0.4≤HI<0.6, and a fault tag is output when HI<0.

4.

9. The method for assessing the aging state of cable insulation based on operational data analysis according to claim 1, characterized in that: After the aging status label is generated in step S5, further processing is performed: When the system is assessed as being in a critical state three times in a row, the life prediction module of the digital twin system is automatically triggered. When the mean square error between the predicted result and the measured cable insulation health index HI exceeds 0.05, online model fine-tuning is triggered.

10. The method for assessing the aging state of cable insulation based on operational data analysis according to claim 1, characterized in that: The method also includes an abnormal operating condition backtracking function: The cable insulation health index HI mutation point in the stored historical data is... Full-dimensional data from the first and last 10 minutes; The causal reasoning model is used to locate the causes of mutations. The causes are classified as follows: lightning overvoltage, mechanical damage, and insulation hydrolysis.

Images