Cable insulation deterioration detection method and system

Abstract

The invention discloses a cable insulation degradation detection method and system, which are applied to the technical field of cable insulation degradation detection, and the method comprises the steps: obtaining the real-time operation current of a target cable; extracting a harmonic signal in the real-time operation current; performing feature extraction on the harmonic signal to obtain a to-be-detected harmonic feature; inputting the to-be-detected harmonic characteristics into an insulation degradation detection model for analysis and processing to obtain an insulation degradation detection result of the target cable; the construction process of the training set comprises the steps of obtaining a harmonic feature set; evaluating each harmonic characteristic in the harmonic characteristic set to obtain a multi-angle characteristic evaluation result; on the basis of difference information between the feature evaluation results, fusing the feature evaluation results, and screening harmonic features to obtain target degraded harmonic features; and constructing a training set based on the target degradation harmonic features and the degradation state. According to the cable insulation degradation detection method and system provided by the invention, the detection accuracy of cable insulation degradation is effectively improved.

Description

Technical Field

[0001] This invention relates to the field of cable insulation degradation detection technology, and in particular to a method and system for detecting cable insulation degradation. Background Technology

[0002] High-voltage cables are insulated cables used to transmit high-voltage electrical energy. They have advantages such as low loss and large transmission capacity, enabling long-distance, efficient power transmission and reducing energy waste during transmission. High-voltage cable insulation degradation detection can identify hidden defects in advance, preventing insulation breakdown and short-circuit faults, ensuring stable power supply, and reducing equipment losses and the risk of safety accidents.

[0003] Currently, the main method for detecting cable insulation degradation is temperature measurement. This method monitors temperature changes in the cable insulation layer or conductor to determine if the insulation material has deteriorated. However, temperature measurements are easily affected by environmental factors, leading to inaccurate readings and distorted initial test results. Distorted results can result in the failure to detect insulation degradation in a timely manner. These untreated degradation problems will continue to develop during operation, causing issues such as partial discharge and insulation breakdown, ultimately leading to short-circuit faults in the cable lines. This not only causes equipment damage and large-scale power outages but may also trigger fires and other safety accidents, threatening the safety of personnel and the overall power grid. Summary of the Invention

[0004] This invention provides a method and system for detecting cable insulation degradation, which solves the technical problem of distorted detection results in existing technologies, thereby improving the accuracy of cable insulation degradation detection results.

[0005] To address the aforementioned technical problems, this invention provides a method for detecting cable insulation degradation, the method comprising: Obtain the real-time operating current of the target cable; Extract the harmonic signals from the real-time operating current; Feature extraction is performed on the harmonic signal to obtain the harmonic features to be measured; The harmonic characteristics to be measured are input into the insulation degradation detection model for analysis and processing to obtain the insulation degradation detection results of the target cable, wherein; The insulation degradation detection model is obtained by training a preset initial insulation degradation detection model based on a training set; The training set construction process includes: processing the operating current of cable samples with known degradation states to obtain a harmonic feature set; evaluating each harmonic feature in the harmonic feature set based on at least two evaluation metrics to obtain multi-angle feature evaluation results; fusing the feature evaluation results based on the difference information between the feature evaluation results, and using the fusion result to filter the harmonic features to obtain target degradation harmonic features; and constructing the training set based on the target degradation harmonic features and the degradation state.

[0006] Preferably, the step of extracting features from the harmonic signal to obtain the harmonic features to be measured includes: The harmonic signal is subjected to multi-level wavelet decomposition to obtain wavelet coefficients at each scale; The wavelet coefficients are reconstructed to generate the corresponding second harmonic signal; Feature extraction is performed on the second harmonic signal to obtain the harmonic features to be measured.

[0007] Preferably, the step of extracting features from the second harmonic signal to obtain the harmonic features to be measured includes: Waveform analysis of the second harmonic signal in the time domain is performed to obtain the time-domain characteristics of the harmonic to be measured. Fourier spectrum analysis of the second harmonic signal in the frequency domain is performed to obtain the frequency domain characteristics of the harmonic to be measured. The second harmonic signal is subjected to wavelet energy distribution processing in the time-frequency domain to obtain the harmonic characteristics to be measured in the time-frequency domain. The second harmonic signal is estimated by spectral correlation function in the cyclic domain to obtain the characteristics of the harmonic to be measured in the cyclic domain. The harmonic features to be measured are obtained by fusing the time-domain harmonic features to be measured, the frequency-domain harmonic features to be measured, the time-frequency-domain harmonic features to be measured, and the cyclic domain harmonic features to be measured.

[0008] Preferably, the initial insulation degradation detection model includes at least a forward timing learning unit and a backward timing learning unit. The insulation degradation detection model is obtained by training a preset initial insulation degradation detection model using a training set, including: The forward temporal learning unit processes the training set input along the positive time direction to obtain the degradation evolution trend; The backward temporal learning unit processes the training set arranged in reverse chronological order to obtain the dependency relationship between the current degradation state and historical data. The degradation evolution trend and the dependency relationship are integrated.

[0009] Preferably, the step of fusing the feature evaluation results based on the difference information between the feature evaluation results includes: Process any two of the feature evaluation results to obtain the corresponding difference matrix; The degree of conflict between the two feature evaluation results is obtained based on the difference matrix. The feature evaluation results are fused based on the degree of conflict.

[0010] The present invention also provides a detection system for cable insulation degradation, comprising: The acquisition module is used to acquire the real-time operating current of the target cable; The extraction module is used to extract harmonic signals from the real-time operating current; The feature module is used to extract features from the harmonic signal to obtain the harmonic features to be measured; The model module is used to input the harmonic characteristics to be measured into the insulation degradation detection model for analysis and processing, so as to obtain the insulation degradation detection results of the target cable. The training module is used to train the insulation degradation detection model based on the training set to obtain the preset initial insulation degradation detection model; The training set module is used for the construction of the training set, which includes processing the operating current of the acquired cable samples with known degradation states to obtain a harmonic feature set; evaluating each harmonic feature in the harmonic feature set based on at least two evaluation indicators to obtain multi-angle feature evaluation results; fusing the feature evaluation results based on the difference information between the feature evaluation results, and using the fusion result to filter the harmonic features to obtain target degradation harmonic features; and constructing the training set based on the target degradation harmonic features and the degradation state.

[0011] Preferably, the feature module includes: Wavelet decomposition unit is used to perform multi-level wavelet decomposition on the harmonic signal to obtain wavelet coefficients at each scale. The reconstruction unit is used to reconstruct the wavelet coefficients and generate the corresponding second harmonic signal; The feature extraction unit is used to extract features from the second harmonic signal to obtain the harmonic features to be measured.

[0012] Preferably, the feature extraction unit includes: The time-domain unit is used to perform waveform analysis on the second harmonic signal in the time domain to obtain the time-domain characteristics of the harmonic to be measured. The frequency domain unit is used to perform Fourier spectrum analysis on the second harmonic signal in the frequency domain to obtain the frequency domain characteristics of the harmonic to be measured. The time-frequency domain unit is used to perform wavelet energy distribution processing on the second harmonic signal in the time-frequency domain to obtain the time-frequency domain harmonic characteristics to be measured. A cyclic domain unit is used to estimate the spectral correlation function of the second harmonic signal in the cyclic domain to obtain the cyclic domain harmonic characteristics to be measured. The fusion unit is used to fuse the time-domain harmonic features to be measured, the frequency-domain harmonic features to be measured, the time-frequency-domain harmonic features to be measured, and the cyclic domain harmonic features to be measured to obtain the harmonic features to be measured.

[0013] Preferably, the initial insulation degradation detection model includes at least a forward timing learning unit and a backward timing learning unit. The training module includes: A forward unit is used by the forward temporal learning unit to process the training set input along the positive time direction to obtain the degradation evolution trend. The backward unit is used by the backward temporal learning unit to process the training set arranged in reverse chronological order to obtain the dependency relationship between the current degradation state and historical data. The fusion unit is used to fuse the degradation evolution trend and the dependency relationship.

[0014] Preferably, the step of fusing the feature evaluation results based on the difference information between the feature evaluation results includes: The difference unit is used to process any two of the feature evaluation results to obtain the corresponding difference matrix; A conflict unit is used to obtain the degree of conflict between two feature evaluation results based on the difference matrix. The evaluation result unit is used to fuse the feature evaluation results based on the degree of conflict.

[0015] Compared with the prior art, the beneficial effects of the present invention are at least one of the following: This invention acquires the real-time operating current of the target cable, extracts the harmonic signals, performs feature extraction to obtain the harmonic features to be measured, and inputs them into an insulation degradation detection model trained on a training set, outputting the insulation degradation detection results. The training set is obtained by processing the operating current of cable samples with known degradation states to obtain a harmonic feature set. Each harmonic feature is evaluated based on at least two evaluation indicators to obtain multi-angle feature evaluation results. Then, the differences between the feature evaluation results are fused, and the target degradation harmonic features are selected based on the fusion results. Combined with the corresponding degradation state, a training set is constructed. This avoids the detection distortion problem caused by environmental interference in traditional temperature measurement methods, and achieves high precision, anti-interference, and early identification of insulation degradation. It effectively prevents safety accidents such as partial discharge, insulation breakdown, and short circuit caused by missed detection, and significantly improves the safety and reliability of high-voltage cable operation. Attached Figure Description

[0016] Figure 1 This is a flowchart illustrating a method for detecting cable insulation degradation in one embodiment of the present invention. Figure 2 This is a structural block diagram of a cable insulation degradation detection model in one embodiment of the present invention; Figure 3 This is a schematic diagram of the cable insulation degradation detection system in one embodiment of the present invention; Figure label: Among them, 11 is the acquisition module; 12 is the extraction module; 13 is the feature module; 14 is the model module; 15 is the training module; and 16 is the training set module. Detailed Implementation

[0017] 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. The purpose of providing these embodiments is to make the disclosure of the present invention more thorough and comprehensive. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0018] In the description of this invention, the terms "first," "second," "third," etc., are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Thus, a feature defined with "first," "second," "third," etc., may explicitly or implicitly include one or more of that feature. In the description of this invention, unless otherwise stated, "a plurality of" means two or more.

[0019] In the description of this invention, it should be noted that, unless otherwise explicitly specified and limited, the terms "installation," "connection," and "linking" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; and they can refer to the internal communication between two components. The term "and / or" as used herein includes any and all combinations of one or more of the associated listed items. Those skilled in the art will understand the specific meaning of the above terms in this invention based on the specific circumstances.

[0020] In the description of this invention, it should be noted that, unless otherwise defined, all technical and scientific terms used in this invention have the same meaning as commonly understood by one of ordinary skill in the art. The terminology used in this specification is for the purpose of describing specific embodiments only and is not intended to limit the invention. Those skilled in the art can understand the specific meaning of the above terms in this invention based on the specific circumstances.

[0021] High-voltage cables have advantages such as low loss and large transmission capacity, and are widely used for long-distance, high-efficiency power transmission. Effective detection of insulation degradation can identify hidden defects in advance, prevent faults such as breakdown and short circuits, and ensure power supply safety.

[0022] The current mainstream temperature measurement method judges the insulation status by monitoring the temperature change of the cable, but it is easily affected by environmental interference, which leads to detection distortion and makes it difficult to identify early deterioration in time. Undetected deterioration will continue to develop, causing serious accidents such as partial discharge, insulation breakdown, short circuit, and fire, threatening the power grid and personal safety.

[0023] One embodiment of the present invention provides a method for detecting cable insulation degradation. For details, please refer to [link to relevant documentation]. Figure 1 , Figure 1 The diagram shown is a flowchart illustrating a method for detecting cable insulation degradation in one embodiment of the present invention. Specifically: S1. Obtain the real-time operating current of the target cable; S2. Extract harmonic signals from the real-time operating current; S3. Extract features from the harmonic signal to obtain the characteristics of the harmonic to be measured; S4. Input the harmonic characteristics to be measured into the insulation degradation detection model for analysis and processing to obtain the insulation degradation detection results of the target cable, wherein; S5. The insulation degradation detection model is obtained by training the preset initial insulation degradation detection model based on the training set. S6. The training set construction process includes processing the operating current of the cable samples with known degradation states to obtain a harmonic feature set; evaluating each harmonic feature in the harmonic feature set based on at least two evaluation indicators to obtain multi-angle feature evaluation results; fusing the feature evaluation results based on the difference information between the feature evaluation results, and using the fusion results to filter the harmonic features to obtain the target degradation harmonic features; and constructing a training set based on the target degradation harmonic features and degradation state.

[0024] Preferably, the real-time operating current of the target cable is obtained. The target cable refers to a specific power cable that needs to be tested for insulation degradation. The real-time operating current is the current flowing in the core of the cable at the current moment under normal power supply conditions. Its value and waveform will dynamically change with the cable load and its own condition. Obtaining the real-time operating current of the target cable is important because the current flowing through the cable core will cause the dielectric to be magnetized, which in turn will induce a magnetization current. When the cable experiences abnormal conditions such as insulation degradation, the magnetic dipoles inside the dielectric will change, causing the magnetic moments to realign in an orderly manner under the influence of the magnetic field of the cable core current. This change will generate specific harmonic currents in the operating current. By analyzing these harmonic signals, the abnormal condition of the cable can be diagnosed. Therefore, obtaining the real-time operating current is the basis for subsequent testing.

[0025] High-order harmonic sensors are employed. These sensors can accurately capture high-frequency harmonic components in the current. Installed near the core of the target cable or at specific detection points, the sensors continuously sense changes in the current within the cable core, converting the current signal into an acquireable electrical signal. This signal is then transmitted to a data acquisition device via a data transmission line. The data acquisition device performs preliminary processing on the received signal, such as filtering and amplification, to ensure its accuracy and integrity. Ultimately, real-time operating current data of the target cable, usable for subsequent analysis, is obtained. Acquiring real-time operating current in this way directly captures current changes related to the cable insulation condition, providing a reliable data source for subsequent harmonic signal extraction and feature analysis, ensuring the accuracy and timeliness of insulation degradation detection.

[0026] Preferably, harmonic signals are extracted from the real-time operating current. Harmonic signals refer to current components in the real-time operating current whose frequencies are integer multiples of the fundamental frequency. The fundamental frequency is the standard frequency of power grid supply; in my country, the fundamental frequency is 50 Hz. Common harmonic signals include the 3rd, 5th, and 7th harmonics. The amplitude and phase changes of these harmonic signals can reflect the insulation state inside the cable. Harmonic signals are extracted from the real-time operating current because specific changes caused by cable insulation degradation are reflected in the harmonic signals. Only by separating these harmonic signals can their characteristics be further analyzed to determine whether the cable insulation has deteriorated. This is a crucial step connecting real-time operating current acquisition and insulation degradation detection.

[0027] First, the acquired real-time operating current data is preprocessed. A combination of low-pass and high-pass filters removes high-frequency noise and low-frequency interference, ensuring the purity of the original current data. Then, a Fourier transform algorithm is used to convert the preprocessed time-domain current signal into a frequency-domain signal. This algorithm decomposes complex current waveforms into sinusoidal components of different frequencies, with components whose frequencies are integer multiples of the fundamental frequency being harmonic signals. Next, spectrum analysis is used to plot the spectrum and calculate the amplitude of the frequency-domain signal, clearly distinguishing between the fundamental and harmonic signals, and extracting key parameters such as amplitude, frequency, and phase of each harmonic signal. Finally, the extracted harmonic signal parameters are stored in a database to prepare data for subsequent feature extraction and insulation degradation detection model analysis. This method of extracting harmonic signals accurately separates key signal components reflecting the cable insulation state, providing accurate data support for subsequent detection, while reducing interference from irrelevant signals and improving the accuracy and efficiency of insulation degradation detection.

[0028] Next, feature extraction is performed on the harmonic signal to obtain the harmonic features to be measured. Specifically, multi-level wavelet decomposition is performed on the harmonic signal to obtain wavelet coefficients at each scale; the wavelet coefficients are reconstructed to generate the corresponding second harmonic signal; feature extraction is performed on the second harmonic signal to obtain the harmonic features to be measured. Further, waveform analysis is performed on the second harmonic signal in the time domain to obtain the harmonic features to be measured in the time domain; Fourier spectrum analysis is performed on the second harmonic signal in the frequency domain to obtain the harmonic features to be measured in the frequency domain; wavelet energy distribution processing is performed on the second harmonic signal in the time-frequency domain to obtain the harmonic features to be measured in the time-frequency domain; spectral correlation function estimation is performed on the second harmonic signal in the cyclic domain to obtain the harmonic features to be measured in the cyclic domain; the harmonic features to be measured in the time domain, frequency domain, time-frequency domain, and cyclic domain are fused to obtain the harmonic features to be measured.

[0029] The harmonic features to be measured refer to the key information extracted from harmonic signals that reflects the insulation status of cables. These features include multi-dimensional characteristics in the time, frequency, and cyclic domains, and their variation patterns are directly related to the degree of cable insulation degradation. Feature extraction of harmonic signals is necessary because simple harmonic signals cannot be directly used for insulation degradation detection model analysis. Only by extracting representative features can the model accurately determine the insulation status; this is the core step in transforming harmonic signals into detection criteria.

[0030] First, multi-level wavelet decomposition is performed. This decomposition breaks down the harmonic signal into multiple signal components at different frequency scales using wavelet basis functions of different scales. Commonly used wavelet basis functions include the Daubechies 4 wavelet and the Symlets 8 wavelet. During the decomposition process, the Mallat algorithm is used to progressively decompose the harmonic signal into approximate coefficients and detail coefficients, which are collectively referred to as wavelet coefficients at each scale. Next, the wavelet coefficients at a specific scale are reconstructed based on the Mallat algorithm. The specific scale is usually a mid-to-high frequency scale that highlights the insulation degradation characteristics. The wavelet coefficients at this scale are restored to a continuous signal through inverse wavelet transform, generating the corresponding second harmonic signal. The second harmonic signal is a harmonic signal component that better reflects the changes in insulation state. Then, feature extraction is performed in different domains. In the time domain, waveform analysis is performed on the second harmonic signal to calculate parameters such as peak value, mean, and variance, obtaining the time-domain harmonic features to be measured. In the frequency domain, the second harmonic signal is further analyzed. Fourier spectrum analysis is performed on the second harmonic signal to obtain its spectrum through Fourier transform. Parameters such as peak frequency bandwidth and spectral energy are calculated to obtain the frequency domain harmonic characteristics. In the time-frequency domain, wavelet energy distribution processing is applied to the second harmonic signal to calculate the signal energy within different time-frequency windows, yielding the time-frequency domain harmonic characteristics. In the cyclic domain, spectral correlation function estimation is performed on the second harmonic signal, and the spectral correlation coefficient at different cyclic frequencies is calculated to obtain the cyclic domain harmonic characteristics. Finally, the time-domain, frequency-domain, and cyclic domain harmonic characteristics are fused using feature-level fusion methods, such as weighted average or principal component analysis, to integrate multi-dimensional features into a unified harmonic characteristic. This feature extraction method comprehensively captures information related to insulation degradation in the harmonic signal, reducing the limitations of single-dimensional features and providing more comprehensive and accurate input data for the insulation degradation detection model, thus improving the model's judgment accuracy and reliability.

[0031] Preferably, the harmonic features to be measured are input into the insulation degradation detection model for analysis and processing to obtain the insulation degradation detection results of the target cable. The insulation degradation detection model is an intelligent analysis model trained based on known cable degradation state data. Commonly used model types include support vector machine models, convolutional neural network models, and random forest models. Its core function is to determine whether the insulation state of the target cable is normal and the degree of degradation by analyzing the input harmonic features. The insulation degradation detection results are conclusions about the insulation state of the target cable output by the model after analysis, typically including categories such as normal, slight degradation, moderate degradation, and severe degradation, providing direct basis for cable maintenance. Inputting the harmonic features to be measured into the insulation degradation detection model for analysis and processing is crucial because although the harmonic features contain insulation state information, they need to be transformed into intuitive detection conclusions through intelligent calculation by the model. This is the final key step in completing the cable insulation degradation detection.

[0032] First, the input harmonic features to be measured are standardized using Min-Max or Z-Score standardization methods to adjust the feature data to a uniform numerical range, avoiding the impact of differences in feature magnitude on the accuracy of model analysis. Next, the standardized harmonic features are input into a pre-trained insulation degradation detection model. If a support vector machine model is used, the features are mapped to the corresponding insulation state category through a pre-defined classification hyperplane. If a convolutional neural network model is used, features are progressively extracted and analyzed through convolutional layers, pooling layers, and fully connected layers, outputting probability distribution results. If a random forest model is used, the final insulation state category is determined through a voting mechanism of multiple decision trees. After the model analysis is completed, the corresponding insulation degradation detection results are output, along with a result reliability assessment value, facilitating staff to judge the reliability of the results. This method of analysis and processing can quickly transform multi-dimensional harmonic features into clear insulation state conclusions, reducing errors and inefficiencies associated with manual analysis. Furthermore, leveraging mature model algorithms improves the accuracy and stability of the detection results, providing strong support for timely cable maintenance and fault prevention.

[0033] Preferably, the insulation degradation detection model is obtained by training a pre-defined initial insulation degradation detection model using a training set. The initial insulation degradation detection model includes at least a forward temporal learning unit and a backward temporal learning unit. The forward temporal learning unit processes the training set input along the forward time direction to obtain the degradation evolution trend; the backward temporal learning unit processes the training set arranged in reverse chronological order to obtain the dependency relationship between the current degradation state and historical data; the degradation evolution trend and dependency relationship are then fused. The forward temporal learning unit is the module in the initial insulation degradation detection model used to process data along the forward time direction. Common implementation structures include forward long short-term memory network units and forward gated recurrent units, etc. Its core function is to capture the pattern of features in the training set changing forward over time, such as the gradual evolution of cable degradation characteristics from slight to severe. The backward temporal learning unit is a module that processes data in reverse chronological order. It includes structures such as backward long short-term memory network units and backward gated recurrent units. Its main purpose is to uncover the correlation between the current degradation state and historical data in the training set, such as the dependency between the current moderate degradation state and the mild degradation data from the previous three months. The insulation degradation detection model is trained based on an initial model containing these two units because cable degradation is a time-varying process. One-way data processing cannot fully capture temporal correlations; bidirectional learning is necessary to more accurately grasp the degradation patterns. This is a key design feature for improving the model's temporal analysis capabilities.

[0034] First, the training set is sorted temporally according to the collection time of the cable sample data, forming a time series. Then, the training set arranged in forward time order is input into the forward temporal learning unit. Through the unit's internal memory cells and gating mechanisms, the target degradation harmonic features at each time step are processed step by step, outputting a degradation evolution trend reflecting the forward development of degradation over time, such as the monthly harmonic amplitude increase. Simultaneously, the training set arranged in reverse time order is input into the backward temporal learning unit. Similarly, the reverse-order data is processed through gating mechanisms, outputting the dependency relationship between the current degradation state and earlier historical data, such as the correlation strength between the current degradation feature and the features of the previous N time steps. Then, calculations are performed... The differences between the degradation evolution trend and the dependency relationship are analyzed using various methods. If Euclidean distance is used, the square root of the sum of the squares of the differences between the corresponding feature values ​​is calculated. If cosine similarity is used, the cosine of the angle between the feature vectors of the two is calculated. A common approach is weighted fusion, adjusting the weights of the two features. For example, if the dependency relationship differs significantly from the evolution trend, and the dependency relationship better reflects the actual degradation logic, the weight of the dependency relationship is increased, fusing the two into a unified temporal feature. Finally, the fused temporal feature is used to update the model parameters. The weight parameters of the forward and backward temporal learning units are adjusted using the backpropagation algorithm. This training process is repeated until the model achieves satisfactory performance on the validation subset, ultimately obtaining the insulation degradation detection model. This method allows the model to comprehensively learn the temporal patterns of cable degradation from both forward and backward time directions, reducing information loss in unidirectional temporal processing, improving the model's ability to capture degradation correlations across different time dimensions, and thus enhancing the accuracy and reliability of the detection results.

[0035] Finally, the construction of the training set includes: processing the operating current of cable samples in known deterioration states to obtain a harmonic feature set; evaluating each harmonic feature in the feature set based on at least two evaluation metrics to obtain multi-angle feature evaluation results; fusing the feature evaluation results based on the differences between them, and using the fused results to filter the harmonic features to obtain the target deterioration harmonic features; and constructing a training set based on the target deterioration harmonic features and the deterioration state. Specifically, any two feature evaluation results are processed to obtain the corresponding difference matrix; the degree of conflict between the two feature evaluation results is obtained based on the difference matrix; and the feature evaluation results are fused based on the degree of conflict. The harmonic feature set is a set of features obtained after processing the operating current of cable samples in known deterioration states. It includes features such as the amplitude, phase, frequency, and energy of each harmonic extracted from the sample current, and each feature may be related to the cable deterioration state. Evaluation metrics are quantitative standards used to assess the effectiveness of each feature in a harmonic feature set. At least two are typically included, commonly feature importance and classification contribution rate. Feature importance measures the degree of influence a feature has on the assessment of degradation status, while classification contribution rate measures the effectiveness of a feature in distinguishing different degradation states. The difference matrix is ​​a matrix obtained by processing the evaluation results of any two features. Each element in the matrix represents the difference value between the two corresponding features under different evaluation metrics. For example, if feature importance and classification contribution rate are used as evaluation metrics, the matrix elements can represent the difference in scores for the same feature under these two metrics. The degree of conflict is the degree of inconsistency between two evaluation results calculated from the difference matrix, reflecting the magnitude of the difference in the assessment of feature effectiveness from different evaluation perspectives. When constructing the training set, the sample currents are processed first to obtain the harmonic feature set. Then, features are selected through multiple evaluation metrics and difference fusion because a single evaluation metric may be biased. Combining multiple metrics and fusing differences can screen for more accurate target degradation harmonic features, which is crucial to ensuring the quality of the training set.

[0036] First, obtain the operating current of a cable sample in a known deteriorated state, and extract harmonic signals and features from it to obtain a harmonic feature set containing multiple harmonic features. Next, select at least two evaluation metrics, such as feature importance and classification contribution rate, to score each feature in the harmonic feature set, obtaining multi-faceted feature evaluation results. Then, process any two feature evaluation results to construct a difference matrix. If the evaluation result is feature score data, the matrix elements can be filled by calculating the absolute difference between the corresponding feature scores to form a difference matrix. Finally, calculate the degree of conflict based on the difference matrix. A common method is to calculate the mean or variance of all elements in the matrix; the larger the mean or variance, the greater the conflict. The higher the degree of conflict, for example, if the mean of the matrix elements is 0.3, it indicates a moderate degree of conflict between the two evaluation results. Then, based on the degree of conflict, the feature evaluation results are fused. When the degree of conflict is low, a simple weighted average method is used, fusing the two evaluation results with equal weights. When the degree of conflict is high, a weighted adjustment method is used, assigning higher weights to evaluation results whose scores better reflect the actual degradation patterns. For example, if the feature importance evaluation result better matches the cable degradation mechanism, its weight is increased. After fusion, features are selected based on the fusion results, retaining features with high fusion scores as target degradation harmonic features. Finally, the target degradation harmonic features are associated with the corresponding cable sample degradation state to form a training set. Constructing a training set in this way effectively selects features more effective for degradation detection, reduces the interference of redundant and invalid features on model training, and improves the accuracy of feature selection through the differences in fused evaluation results, providing high-quality data support for subsequent model training, thereby improving the performance of the insulation degradation detection model.

[0037] In one embodiment of the present invention, discrete wavelet transform is used to perform wavelet transform on the harmonic signal. The current signal acquired from one end of the cable is decomposed into eight levels using the db4 wavelet to obtain wavelet transform coefficients at each scale. The harmonic signal is then reconstructed from the wavelet coefficients using the Mallat algorithm.

[0038] For the reconstructed harmonic signals, 30 feature indicators reflecting the harmonic characteristics of cables were extracted in the time domain, frequency domain, time-frequency domain, and cyclic domain. These features include mean, standard deviation, root mean square (RMS), peak value, peak-to-peak value, peak factor, skewness, kurtosis, impulse factor, waveform factor, margin factor, center frequency, RMS frequency, RMS frequency, frequency variance, frequency standard deviation, mean spectral kurtosis, standard deviation spectral kurtosis, skewness spectral kurtosis, energy entropy, singular entropy, instantaneous frequency peak-to-peak value, instantaneous frequency RMS value, instantaneous frequency standard deviation, average cyclic frequency, centroid cyclic frequency, cyclic frequency RMS value, cyclic frequency standard deviation, cyclic frequency skewness, and cyclic frequency energy. These feature indicators are used for the classification and aging state identification of insulation aging data. However, this approach results in a large number of redundant features, leading to the curse of dimensionality. Furthermore, directly using these 30 feature indicators as input to the classification and identification algorithms generates a significant computational load, reducing computational efficiency.

[0039] This invention proposes another feature selection method, which selects the most representative feature subset that contributes the most to model performance from the original feature set. The aim is to reduce data dimensionality. Indicators with a significant impact on aging classification are extracted from 30 feature indicators. These selected indicators are used as input to unsupervised dimensionality reduction and classification algorithms, improving computational speed and accuracy. While a single feature evaluation criterion can reflect feature quality from a certain perspective, it also has limitations. To comprehensively evaluate the impact of feature indicators on fault classification, key indicators are selected.

[0040] Data on different aging states were obtained by collecting harmonic data. For each sample, 30 characteristic indicators from different domains were statistically analyzed, and then multi-criteria scores of these characteristic indicators were calculated and normalized.

[0041] For each feature score, normalization is performed to construct a basic probability allocation function, the sum of which is 1. For the feature set... After normalization, the basic probability distribution is obtained as follows: Features The basic probability, That is, the basic probability allocation function corresponding to the i-th feature score. For feature set, m represents the feature importance under the i-th feature score. j This represents the feature importance under the j-th feature score, and the distance is obtained. : It is the basic probability assignment vector. It is a matrix that measures the difference between two sets, i.e., D=D h ×h .

[0042] Based on three fault categories, the sample size for each category is 1050. Let t be the set of samples of class t. For the F in the t-th type of deteriorated sample h The mean, For all degradation category samples, F h The overall mean yields four feature scoring criteria: in, F h The weights, where N is the total number of samples. F is the nearest neighbor sample of the j-th class. h Values, F is the F of the j-th nearest neighbor sample in the t-th heterogeneous class. h For each cable sample x, the value is F. h Features, labeled as degradation type , The category discrimination score is given. The higher the score, the higher the discrimination of the underground cable's higher harmonic characteristics Fh between different categories, which is more conducive to classification. Weighted scoring, quantified features The importance of diagnosing cable degradation; The impurity of the ginnie is rated; the higher the rating, the more impure it is. The more concentrated the categories of cable degradation samples after splitting; Mutual information scoring is used to measure the strength of the association between features and category labels; The weight, This represents the number of nearest neighbor samples.

[0043] Let Gini be the name of all F1 to F2. n Impurity of the Gini after splitting. For F h Weighted Gini impurity after splitting sample set X, traversing F h Given all possible values ​​of the threshold r, divide the sample set X into two subsets. and We can obtain: For F h Mutual information with degradation type Y. Given F h back, .

[0044] According to matrix D h ×h Calculation evidence Degree of conflict with other evidence The greater the conflict, the smaller the weight should be selected during feature fusion.

[0045] Evidence calculated based on the degree of conflict weight w i : The weights are adjusted based on the degree of conflict; the higher the degree of conflict, the smaller the adjusted weight. The adjusted weights are then normalized.

[0046] Obtain the weights of different features, and perform a weighted average to fuse the scores P of the p features. i : in This represents the i-th score of the h-th feature. The scores of different features are summed according to their respective weights to obtain the final fusion result. Integrating multiple feature evaluation methods can more comprehensively reflect feature quality and select a subset of features that are more sensitive to insulation aging fault diagnosis.

[0047] Another embodiment of the present invention provides a detection model for cable insulation degradation. For details, please refer to [link to relevant documentation]. Figure 2 , Figure 2 The diagram shown is a structural block diagram of a cable insulation degradation detection model according to one embodiment of the present invention. Specifically: (1) Extracting information on higher harmonic faults Current or voltage signals are collected from operating cables, and their higher harmonic components are extracted using spectrum analysis techniques. Since cables produce specific harmonic distortion modes at different stages of fault deterioration (such as insulation aging, partial discharge, poor joint contact, etc.), the content of each higher harmonic is used as a key feature reflecting the cable's health status to construct initial fault characteristic data.

[0048] (2) Wavelet Transform Wavelet transform is applied to the original signal to achieve joint time-frequency domain analysis. Wavelet transform can effectively capture time-varying harmonic components in non-stationary signals, and is especially suitable for complex scenarios in power systems with transient interference and periodic harmonic superposition, enhancing the sensitivity and resolution of weak fault characteristics.

[0049] (3) Selection of harmonic characteristics based on multiple criteria Based on multiple evaluation criteria (such as energy contribution rate, information entropy, correlation coefficient, variance, etc.), the harmonic components of each frequency band after wavelet decomposition are screened, retaining the most discriminative harmonic features and eliminating redundant or noise-dominated components, thereby improving the discriminability and robustness of the feature vectors and providing high-quality input for subsequent modeling.

[0050] (4) Constructing harmonic eigenvectors The harmonic features, after being screened by multiple criteria, are sequentially combined into a one-dimensional or two-dimensional feature vector. This vector comprehensively represents the harmonic distribution characteristics of the cable under its current operating state and serves as input data for a deep neural network model, possessing good interpretability and modeling potential.

[0051] (5) Input → Convolutional Layer The constructed harmonic feature vectors are fed into the neural network model and first processed by convolutional layers. A one-dimensional convolutional kernel with a size of 1×5 is designed to adapt to the bandwidth of high-order harmonic frequency bands and cover the correlation features between adjacent harmonics. Local key patterns, such as abrupt changes in harmonic amplitude and spectral trend changes, are automatically extracted through sliding window convolution operations, enhancing the model's ability to perceive subtle fault signals.

[0052] (6) Pooling layer Pooling layers (such as max pooling or average pooling) are introduced after convolutional layers to reduce feature dimensionality, reduce computational complexity, while retaining key feature information, improving the model's robustness to input perturbations, and preventing overfitting.

[0053] (7) Bidirectional BiLSTM layer The pooled features are fed into a Bidirectional Long Short-Term Memory (BiLSTM) layer. This layer contains two parallel processing units: Forward unit: Processes data sequentially from the start to the end of the time series to capture the evolutionary pattern of "current harmonic characteristics → future deterioration trend"; Backward unit: Process time series data in reverse, trace the correlation between current features and historical features, and uncover long-term dependencies.

[0054] By using bidirectional modeling, we can gain a more complete understanding of the time dynamics of harmonic signals and improve our ability to identify slowly developing fault processes.

[0055] (8) Attention mechanism layer An attention mechanism layer is introduced on top of the BiLSTM output to calculate the importance weight of features at each time step. Higher weights are assigned to key features (such as significant harmonic surges and persistent distortions), and the outputs of the bidirectional temporal learning units are weighted and fused to highlight the feature representation of fault-sensitive periods, further enhancing the model's discriminative performance.

[0056] (9) Output The final output layer integrates all high-level feature representations to generate cable fault diagnosis results. This output includes classification probabilities or corresponding degradation quantification values ​​for three fault levels (mild, moderate, and severe), enabling accurate assessment and early warning of cable operating status.

[0057] Another embodiment of the present invention provides a detection system for cable insulation degradation. For details, please refer to [link to relevant documentation]. Figure 3 , Figure 3 The diagram shown is a structural schematic of a cable insulation degradation detection system according to one embodiment of the present invention. Specifically: Acquisition module 11 is used to acquire the real-time operating current of the target cable; Extraction module 12 is used to extract harmonic signals from the real-time operating current; Feature module 13 is used to extract features from harmonic signals to obtain the features of the harmonics to be measured; Model module 14 is used to input the harmonic characteristics to be measured into the insulation degradation detection model for analysis and processing, so as to obtain the insulation degradation detection results of the target cable. Training module 15 is used to train the insulation degradation detection model based on the training set to obtain the preset initial insulation degradation detection model; The training set module 16 is used for the construction of the training set, which includes processing the operating current of the cable samples with known degradation states to obtain a harmonic feature set; evaluating each harmonic feature in the harmonic feature set based on at least two evaluation indicators to obtain multi-angle feature evaluation results; fusing the feature evaluation results based on the difference information between the feature evaluation results, and using the fusion results to filter the harmonic features to obtain the target degradation harmonic features; and constructing a training set based on the target degradation harmonic features and degradation state.

[0058] Preferably, feature module 13 includes: Wavelet decomposition unit is used to perform multi-level wavelet decomposition on harmonic signals to obtain wavelet coefficients at various scales. The reconstruction unit is used to reconstruct the wavelet coefficients and generate the corresponding second harmonic signal; The feature extraction unit is used to extract features from the second harmonic signal to obtain the harmonic features to be measured.

[0059] Preferably, the feature extraction unit includes: The time-domain unit is used to perform waveform analysis on the second harmonic signal in the time domain to obtain the time-domain characteristics of the harmonic to be measured. The frequency domain unit is used to perform Fourier spectrum analysis on the second harmonic signal in the frequency domain to obtain the frequency domain characteristics of the harmonic to be measured. The time-frequency domain unit is used to perform wavelet energy distribution processing on the second harmonic signal in the time-frequency domain to obtain the time-frequency domain characteristics of the harmonic to be measured. The cyclic domain unit is used to estimate the spectral correlation function of the second harmonic signal in the cyclic domain to obtain the characteristics of the harmonic to be measured in the cyclic domain. The fusion unit is used to fuse the harmonic features to be measured in the time domain, frequency domain, time-frequency domain, and cyclic domain to obtain the harmonic features to be measured.

[0060] Preferably, the initial insulation degradation detection model includes at least a forward timing learning unit and a backward timing learning unit. Training module 15 includes: The forward unit is used by the forward temporal learning unit to process the training set input along the positive time direction to obtain the degradation evolution trend; The backward unit is used by the backward temporal learning unit to process the training set arranged in reverse chronological order to obtain the dependency relationship between the current degradation state and historical data. The fusion unit is used to fuse degradation evolution trends and dependencies.

[0061] Preferably, based on the difference information between feature evaluation results, the feature evaluation results are fused, including: The difference unit is used to process the evaluation results of any two features to obtain the corresponding difference matrix; Conflict unit, used to determine the degree of conflict between two feature evaluation results based on the difference matrix; The evaluation result unit is used to fuse feature evaluation results based on the degree of conflict.

[0062] Those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. The program can be stored in a computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. The storage medium can be a magnetic disk, optical disk, read-only memory (ROM), or random access memory (RAM), etc.

[0063] Accordingly, embodiments of the present invention provide a computer-readable storage medium comprising a stored computer program, wherein, when the computer program is executed, it controls the device containing the computer-readable storage medium to perform steps in the cable insulation degradation detection method of the above embodiments, for example... Figure 1 Steps S1 to S6 as described above.

[0064] This technical solution abandons the temperature measurement method, which is susceptible to environmental interference. Instead, it extracts harmonic signals from the real-time operating current of the target cable and generates the harmonic features to be measured based on these signals. These features are then input into a pre-trained insulation degradation detection model to output a degradation state judgment result. The training set of the model is obtained by processing the current data of cable samples with known degradation states to obtain a harmonic feature set. Multiple evaluation indicators are used to independently evaluate each feature, forming multi-angle feature evaluation results. Furthermore, adaptive fusion is performed based on the differences between these evaluation results to select the most discriminative target degradation harmonic features, which are then paired with the actual degradation state to construct the model. This method fully utilizes the inherent sensitivity of harmonics to nonlinear insulation degradation and improves feature reliability through a multi-criteria collaborative screening mechanism. It fundamentally solves the problems of misjudgment and missed detection caused by environmental temperature changes in traditional methods, achieving earlier, more accurate, and more stable insulation degradation diagnosis, and effectively preventing major operational risks such as short circuits, breakdowns, and fires.

[0065] The embodiments described above are merely illustrative of several implementations of the present invention, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of the present invention. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the present invention, and these modifications and improvements all fall within the scope of protection of the present invention. Therefore, the scope of protection of this patent should be determined by the appended claims.

Claims

1. A method for detecting cable insulation deterioration, characterized in that, include: Obtain the real-time operating current of the target cable; Extract the harmonic signals from the real-time operating current; Feature extraction is performed on the harmonic signal to obtain the harmonic features to be measured; The harmonic characteristics to be measured are input into the insulation degradation detection model for analysis and processing to obtain the insulation degradation detection results of the target cable, wherein; The insulation degradation detection model is obtained by training a preset initial insulation degradation detection model based on a training set; The process of constructing the training set includes processing the operating current of the cable samples with known degradation status to obtain a harmonic feature set; evaluating each harmonic feature in the harmonic feature set based on at least two evaluation indicators to obtain multi-angle feature evaluation results; fusing the feature evaluation results based on the difference information between the feature evaluation results, and using the fusion result to filter the harmonic features to obtain the target degradation harmonic features. The training set is constructed based on the target degraded harmonic features and the degraded state.

2. The method for detecting cable insulation deterioration as described in claim 1, characterized in that, The step of extracting features from the harmonic signal to obtain the harmonic features to be measured includes: The harmonic signal is subjected to multi-level wavelet decomposition to obtain wavelet coefficients at each scale; The wavelet coefficients are reconstructed to generate the corresponding second harmonic signal; Feature extraction is performed on the second harmonic signal to obtain the harmonic features to be measured.

3. The method for detecting cable insulation deterioration as described in claim 2, characterized in that, The step of extracting features from the second harmonic signal to obtain the harmonic features to be measured includes: Waveform analysis of the second harmonic signal in the time domain is performed to obtain the time-domain characteristics of the harmonic to be measured. Fourier spectrum analysis of the second harmonic signal in the frequency domain is performed to obtain the frequency domain characteristics of the harmonic to be measured. The second harmonic signal is subjected to wavelet energy distribution processing in the time-frequency domain to obtain the harmonic characteristics to be measured in the time-frequency domain. The second harmonic signal is estimated by spectral correlation function in the cyclic domain to obtain the characteristics of the harmonic to be measured in the cyclic domain. The harmonic features to be measured are obtained by fusing the time-domain harmonic features to be measured, the frequency-domain harmonic features to be measured, the time-frequency-domain harmonic features to be measured, and the cyclic domain harmonic features to be measured.

4. The method for detecting cable insulation deterioration as described in claim 1, characterized in that, The initial insulation degradation detection model includes at least a forward timing learning unit and a backward timing learning unit. The insulation degradation detection model is obtained by training a preset initial insulation degradation detection model using a training set, including: The forward temporal learning unit processes the training set input along the positive time direction to obtain the degradation evolution trend; The backward temporal learning unit processes the training set arranged in reverse chronological order to obtain the dependency relationship between the current degradation state and historical data. The degradation evolution trend and the dependency relationship are integrated.

5. The method for detecting cable insulation deterioration as described in claim 1, characterized in that, The step of fusing the feature evaluation results based on the difference information between the feature evaluation results includes: Process any two of the feature evaluation results to obtain the corresponding difference matrix; The degree of conflict between the two feature evaluation results is obtained based on the difference matrix. The feature evaluation results are fused based on the degree of conflict.

6. A detection system for cable insulation deterioration, characterized in that, include: The acquisition module is used to acquire the real-time operating current of the target cable; The extraction module is used to extract harmonic signals from the real-time operating current; The feature module is used to extract features from the harmonic signal to obtain the harmonic features to be measured; The model module is used to input the harmonic characteristics to be measured into the insulation degradation detection model for analysis and processing, so as to obtain the insulation degradation detection results of the target cable. The training module is used to train the insulation degradation detection model based on the training set to obtain the preset initial insulation degradation detection model; The training set module is used for the construction process of the training set, which includes processing the operating current of the obtained cable samples with known degradation status to obtain a harmonic feature set; evaluating each harmonic feature in the harmonic feature set based on at least two evaluation indicators to obtain multi-angle feature evaluation results; fusing the feature evaluation results based on the difference information between the feature evaluation results, and using the fusion result to filter the harmonic features to obtain target degradation harmonic features; The training set is constructed based on the target degraded harmonic features and the degraded state.

7. The cable insulation degradation detection system as described in claim 6, characterized in that, The feature module includes: Wavelet decomposition unit is used to perform multi-level wavelet decomposition on the harmonic signal to obtain wavelet coefficients at each scale. The reconstruction unit is used to reconstruct the wavelet coefficients and generate the corresponding second harmonic signal; The feature extraction unit is used to extract features from the second harmonic signal to obtain the harmonic features to be measured.

8. The cable insulation degradation detection system as described in claim 7, characterized in that, The feature extraction unit includes: The time-domain unit is used to perform waveform analysis on the second harmonic signal in the time domain to obtain the time-domain characteristics of the harmonic to be measured. The frequency domain unit is used to perform Fourier spectrum analysis on the second harmonic signal in the frequency domain to obtain the frequency domain characteristics of the harmonic to be measured. The time-frequency domain unit is used to perform wavelet energy distribution processing on the second harmonic signal in the time-frequency domain to obtain the time-frequency domain harmonic characteristics to be measured. A cyclic domain unit is used to estimate the spectral correlation function of the second harmonic signal in the cyclic domain to obtain the cyclic domain harmonic characteristics to be measured. The fusion unit is used to fuse the time-domain harmonic features to be measured, the frequency-domain harmonic features to be measured, the time-frequency-domain harmonic features to be measured, and the cyclic domain harmonic features to be measured to obtain the harmonic features to be measured.

9. The cable insulation degradation detection system as described in claim 6, characterized in that, The initial insulation degradation detection model includes at least a forward timing learning unit and a backward timing learning unit. The training module includes: A forward unit is used by the forward temporal learning unit to process the training set input along the positive time direction to obtain the degradation evolution trend. The backward unit is used by the backward temporal learning unit to process the training set arranged in reverse chronological order to obtain the dependency relationship between the current degradation state and historical data. The fusion unit is used to fuse the degradation evolution trend and the dependency relationship.

10. The cable insulation degradation detection system as described in claim 6, characterized in that, The step of fusing the feature evaluation results based on the difference information between the feature evaluation results includes: The difference unit is used to process any two of the feature evaluation results to obtain the corresponding difference matrix; A conflict unit is used to obtain the degree of conflict between two feature evaluation results based on the difference matrix. The evaluation result unit is used to fuse the feature evaluation results based on the degree of conflict.

Images