Method for detecting defects of live cable equipment by using high-frequency current detection method

Abstract

The invention relates to the technical field of defect detection, and discloses a method for detecting defects of live cable equipment by using a high-frequency current detection method, which comprises the following steps of: exciting the live cable equipment to generate a high-frequency current signal, and judging the optimal frequency of the high-frequency current signal by adopting the minimum signal propagation loss; the method comprises the following steps: acquiring a current signal on cable equipment through a high-frequency current sensor, and performing multi-scale time-frequency analysis on the signal through wavelet packet transformation; carrying out noise reduction processing on the converted signal to remove an interference signal and extract a partial discharge signal; positioning the partial discharge signal based on the Bayesian theorem, and judging the position of a partial discharge source; and classifying the partial discharge signals by adopting a support vector machine. The excitation frequency of the high-frequency current signal enables the signal transmission loss to be minimized and the extraction efficiency of the partial discharge signal to be improved, and the signal transmission efficiency is improved in a complex cable environment, so that the defect detection precision and sensitivity are enhanced.

Description

Technical Field

[0001] The present invention relates to the technical field of defect detection, and in particular to a method for detecting defects in live cable equipment using a high-frequency current detection method. Background Art

[0002] As distribution networks become increasingly complex, cable lines are increasingly important in urban power systems. 10kV medium-voltage cables, in particular, are widely used in urban grid architectures. Partial discharge (PD), an early sign of cable insulation degradation, is crucial for preventing sudden cable failures and improving operational reliability. Traditional PD detection methods include pulse current, ultrasonic, and electric field coupling. Most of these methods rely on manual intervention or power outages, failing to meet the real-time assessment requirements for potential defects during live operation.

[0003] In recent years, high-frequency current detection has become an important approach for live cable defect detection due to its non-invasive, online detection, and high sensitivity to partial discharges. However, in practical engineering applications, diverse cable structures, complex installation paths, and severe electromagnetic interference result in significant propagation losses of the excitation signal within the cable. During transmission, partial discharge signals are prone to distortion, amplitude attenuation, and even being lost in noise, severely impacting the accuracy of the final identification results. Summary of the Invention

[0004] In response to the shortcomings of the existing technology, the present invention provides a method for detecting defects in live cable equipment using a high-frequency current detection method. This method solves the problem that in existing high-frequency current detection methods, the excitation frequency is usually fixed or selected by manual experience, and is not dynamically optimized according to the cable structure and working status, resulting in excessive signal loss during transmission, thereby reducing the extraction efficiency of partial discharge signals and the accuracy of defect identification.

[0005] To achieve the above objectives, the present invention is implemented through the following technical solutions: A method for detecting defects in live cable equipment using a high-frequency current detection method, comprising the following steps:

[0006] Exciting the live cable device to generate a high-frequency current signal, wherein the frequency of the high-frequency current signal is determined to be an optimal frequency by minimizing signal propagation loss;

[0007] The current signal on the cable equipment is obtained by a high-frequency current sensor, and the signal is subjected to multi-scale time-frequency analysis through wavelet packet transform;

[0008] Performing noise reduction processing on the transformed signal to remove interference signals and extract partial discharge signals;

[0009] Locate the partial discharge signal based on Bayesian theorem to determine the location of the partial discharge source;

[0010] Support vector machine is used to classify partial discharge signals and determine the defect type, and the severity of the defect is evaluated based on a fuzzy logic reasoning system.

[0011] Preferably, the optimal frequency is optimized by the least squares method, and the objective function for minimizing the signal propagation loss is:

[0012]

[0013] Among them, α(f i ) is the signal propagation loss at the i-th frequency point, α ideal For the ideal frequency response.

[0014] Preferably, the wavelet packet transform decomposes the signal at multiple levels to obtain coefficients in different frequency bands, and selects the optimal wavelet packet basis function according to Kolmogorov complexity theory to maximize the signal compression rate and minimize noise interference.

[0015] Preferably, in the Bayesian positioning step, the posterior probability of the location of the local discharge source is calculated by the following Bayesian formula:

[0016]

[0017] Where P(x0|y(t)) is the posterior probability, P(y(t)|x0) is the likelihood function, which represents the probability of signal y(t) at a given location x0, P(x0) is the prior probability, and P(y(t)) is the marginal likelihood.

[0018] Preferably, the signal noise reduction processing includes selecting a subband in the signal related to partial discharge and applying a soft threshold denoising method to remove background noise.

[0019] Preferably, the decision function of the support vector machine classifier is:

[0020]

[0021] Among them, x i is the training data, y i is the category label, K(x i ,x) is the kernel function and b is the bias term.

[0022] Preferably, the fuzzy logic inference system evaluates the severity of cable defects based on input signal characteristics, and the fuzzy rules are expressed as:

[0023] Severity(x)=IF x1is A1AND x2is A2THEN severity is S;

[0024] Among them, x1, x2 are signal features, A1, A2 are fuzzy sets, and S is the severity assessment result.

[0025] Preferably, after evaluating the severity of the cable defect, the following steps are performed:

[0026] The real-time monitoring and data feedback steps include dynamically adjusting detection parameters to adapt to different working conditions of live cable equipment and automatically updating the parameters of the signal processing algorithm during the detection process.

[0027] Preferably, the signal extraction and processing system includes a high-frequency current sensor, a data acquisition device, a signal processing unit and a display terminal, wherein the signal processing unit dynamically adjusts the frequency range and signal processing algorithm of the high-frequency current signal according to real-time data.

[0028] Preferably, the live cable equipment is a 10kV distribution line, and the method is used to detect and diagnose partial discharge of the cable to determine the operating safety and stability of the cable.

[0029] The present invention provides a method for detecting defects of live cable equipment by utilizing a high-frequency current detection method.

[0030] It has the following beneficial effects:

[0031] 1. The present invention minimizes signal transmission loss by optimizing the excitation frequency of high-frequency current signals, thereby improving the extraction efficiency of partial discharge signals. During use, even in complex cable environments, the signal transmission efficiency is improved, thereby enhancing the accuracy and sensitivity of defect detection.

[0032] 2. The present invention uses wavelet packet transform to perform multi-scale time-frequency analysis on high-frequency current signals, thereby accurately extracting the local discharge characteristics in the signal and significantly reducing noise interference. Wavelet packet transform can perform detailed analysis of signals in different frequency bands, removing environmental noise and other irrelevant signals, making local discharge signals easier to identify.

[0033] 3. The present invention locates the source of partial discharge through the Bayesian optimal estimation method, achieving precise defect source location and improving positioning accuracy. The Bayesian reasoning method combines prior information and real-time signals to accurately determine the location of the partial discharge source in real time, significantly improving positioning accuracy.

[0034] 4. The present invention uses support vector machines to classify defects and fuzzy logic reasoning to evaluate defect severity, thereby realizing intelligent discrimination of defect type and severity, and obtaining more accurate and comprehensive defect analysis results. It can automatically classify and evaluate the risk level of defects according to the characteristics of cable signals, providing a scientific basis for subsequent cable maintenance decisions. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] Figure 1 This is a flow chart of a method for detecting defects in live cable equipment using a high-frequency current detection method according to the present invention;

[0036] Figure 2 A wiring diagram of live detection equipment for a method of detecting defects in live cable equipment using a high-frequency current detection method according to the present invention;

[0037] Figure 3 This is a schematic diagram of the principle of high-frequency partial discharge detection of electric power equipment, which is a method for detecting defects in live cable equipment using a high-frequency current detection method according to the present invention;

[0038] Figure 4 This is a wiring diagram of a high-frequency partial discharge signal detection circuit for a method of detecting defects in live cable equipment using a high-frequency current detection method according to the present invention;

[0039] Figure 5 A schematic diagram of the structure of a Rogowski coil in a method for detecting defects in live cable equipment using a high-frequency current detection method according to the present invention;

[0040] Figure 6 This is an equivalent circuit diagram of a Rogowski coil in a method for detecting defects in live cable equipment using a high-frequency current detection method according to the present invention. DETAILED DESCRIPTION

[0041] The following will clearly and completely describe the technical solution of the present invention in conjunction with the accompanying drawings. Obviously, the embodiments described are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0042] Please see the attached Figure 1 The embodiment of the present invention provides a method for detecting defects in live cable equipment using a high-frequency current detection method, comprising the following steps:

[0043] Exciting the live cable equipment to generate a high-frequency current signal, the frequency of the high-frequency current signal is determined to be the optimal frequency by minimizing the signal propagation loss;

[0044] The current signal on the cable equipment is obtained through a high-frequency current sensor, and the signal is subjected to multi-scale time-frequency analysis through wavelet packet transform;

[0045] Performing noise reduction processing on the transformed signal to remove interference signals and extract partial discharge signals;

[0046] Locate the partial discharge signal based on Bayesian theorem to determine the location of the partial discharge source;

[0047] Support vector machine is used to classify partial discharge signals and determine the defect type, and the severity of the defect is evaluated based on a fuzzy logic reasoning system.

[0048] The optimal frequency is optimized by the least squares method, and the objective function of minimizing the signal propagation loss is:

[0049]

[0050] Among them, α(f i ) is the signal propagation loss at the i-th frequency point, α ideal For the ideal frequency response.

[0051] Specifically, the system sends an excitation signal through the control terminal, applying a high-frequency excitation current to the cable equipment under test. This excitation is not selected arbitrarily; rather, the frequency is quantitatively optimized based on the criterion of minimum propagation loss. To obtain the optimal operating frequency, a signal propagation loss evaluation model is designed, and the objective function is established based on the least squares criterion. The goal is to minimize the sum of the squares of the difference between the actual signal propagation loss and the ideal target response at a set of discrete frequency sampling points. The optimization problem is formally described as follows:

[0052]

[0053] in:

[0054] α(f i ) represents the propagation loss value at the i-th frequency point, α ideal Ideal frequency response index for reference;

[0055] f is the variable of the excitation frequency, and m represents the total number of frequency sampling points involved in the calculation.

[0056] By approximating an ideal transmission state, the selected frequency not only has penetration capabilities but also improves signal integrity and identifiability. Frequency optimization does not directly act on the cable load, but is performed in the offline calculation module. The system will preset a set of candidate frequency intervals, sample them one by one, calculate the corresponding loss level, and finally select a minimum error frequency. This frequency will be set as the excitation frequency f for this test. opt ;

[0057] In order to prevent local minima from interfering with the optimization results, the present invention uses smoothing filtering to pre-process the loss curve and adds an adaptive adjustment factor to the objective function to correct the error drift caused by environmental noise;

[0058] This frequency selection mechanism avoids the signal distortion caused by arbitrary frequency settings in traditional methods, further improving the efficiency of subsequent partial discharge feature extraction. It also mitigates the interference of uneven cable load distribution on frequency transmission behavior to a certain extent. Specifically, the excitation module includes a high-frequency current source, a frequency controller, and an overcurrent protection circuit. The control system can adjust frequency setting parameters through a communication interface and automatically load the optimal frequency configuration. The frequency selection strategy fully considers the differences in cable system characteristics, such as model, voltage level, and installation method. Instead of static settings, an empirical model is established based on historical test data to narrow the optimal search range. This data-driven frequency prediction mechanism provides greater adaptability to the overall test process.

[0059] Wavelet packet transform decomposes the signal at multiple levels to obtain coefficients in different frequency bands, and selects the optimal wavelet packet basis function according to Kolmogorov complexity theory to maximize the signal compression rate and minimize noise interference.

[0060] Specifically, the wavelet packet analysis module is placed at the front end of the signal processing chain to perform fine-grained frequency band decomposition of the collected high-frequency current response signal. Compared with ordinary wavelet transform, wavelet packet decomposition has bidirectional recursive capabilities, which not only further decomposes the approximation part, but also decomposes the detail part. This can form a more detailed frequency division, making the target signal characteristics more prominent in a specific frequency band. The system first sets the number of decomposition layers L, and the default range is between 2-5. Too high a number of layers will introduce a large number of invalid coefficients, resulting in computational redundancy, and too low a number of layers will not be conducive to capturing local structures. After each layer of decomposition, the original signal is projected into multiple frequency bands, and the system retains the coefficient set under each frequency band, recorded as C l,k , where l represents the decomposition layer and k represents the corresponding frequency band number in the layer;

[0061] When choosing which set of wavelet basis functions (or "wavelet packet basis") to use, traditional methods often rely on experience or simple energy concentration standards. However, the present invention does not adopt this fixed strategy, but introduces Kolmogorov complexity theory as a guiding principle. Kolmogorov complexity is used to measure the coding length required for the simplest description of an object, reflecting the "structural order" of the signal. The more structured the signal, the lower its complexity and the easier it is to compress. In the present invention, a wavelet packet selection function is constructed using this theory. The goal is to select the basis function that maximizes the signal compression rate and minimizes redundancy from all possible wavelet basis candidate sets. The compression rate can be defined as:

[0062]

[0063] In order to balance signal fidelity and anti-interference ability, the present invention also introduces a cost function balance term in the basis function selection to avoid the phenomenon of feature leakage caused by pursuing only compression rate. In actual operation, the system will traverse several commonly used wavelet packet bases (such as dbN, symN, coifN, biorN, etc.), reconstruct and encode each set of coefficients, and calculate its Kolmogorov estimated length. Finally, the group with the smallest complexity and the smallest reconstruction error is selected as the standard wavelet basis of the detection process. After the above-mentioned multi-level decomposition and screening, the wavelet packet transform can extract the impulse response component representing the local discharge characteristics, and significantly enhance its amplitude performance in multiple frequency bands, while suppressing redundant information in other bandwidths. The denoising effect does not rely on the forced elimination of filters, but eliminates invalid components through structural recognition, fundamentally reducing the possibility of false detection.

[0064] In the Bayesian localization step, the posterior probability of the PD source location is calculated using the following Bayesian formula:

[0065]

[0066] Where P(x0|y(t)) is the posterior probability, P(y(t)|x0) is the likelihood function, which represents the probability of signal y(t) at a given location x0, P(x0) is the prior probability, and P(y(t)) is the marginal likelihood.

[0067] Specifically, the Bayesian positioning module takes the time series signal y(t) as the input variable and the possible discharge source location, x0, as the conditional variable to construct the target posterior probability distribution function:

[0068]

[0069] P(x0|y(t)) represents the posterior probability, that is, the probability that the PD source is located at position x0 under the condition that the signal y(t) is observed;

[0070] P(y(t)|x0) is the likelihood function, which indicates the probability that the system should observe the current signal y(t) if the PD source is indeed at position x0;

[0071] P(x0) is the prior probability, which is the initial estimate of the PD source position when there is no observed signal;

[0072] P(y(t)) is the marginal likelihood, which normalizes the signal observation probability at all possible positions and is used as a normalization factor. The construction of the likelihood function is a key step. The system pre-establishes a series of reference response libraries based on the preset cable structure model (including material parameters, segment length, electromagnetic propagation characteristics, etc.). Each set of response libraries corresponds to a specific discharge position x i, obtained by high-frequency signal simulation. The system converts the residual square sum of the actual collected signal and each reference response into a likelihood value;

[0073] In terms of prior probability, if there is no clear prior information, a uniform distribution model can be used, that is, all possible locations are equally likely to be distributed. If historical data in the system indicates that certain locations are more prone to partial discharges, such as intermediate joints or bends, the prior distribution will be weighted to reflect the actual occurrence tendency. After the posterior probability values ​​are calculated for all possible locations, the system will output the location corresponding to the maximum posterior probability as the positioning result. For situations where there are multiple partial discharge sources, a multi-peak fitting model can also be used, while retaining multiple maximum values ​​to form multi-point discrimination. It is worth noting that the marginal likelihood P(y(t)) is not explicitly calculated in most actual calculations, but is processed through proportional normalization so that the entire posterior probability curve numerically meets the probability density distribution requirements. Compared with traditional methods, this method has the following advantages: it does not rely on precise synchronous clocks, only requires high-frequency current signals, is less sensitive to noise, and has stable inference results. It can be extended to two-dimensional or multi-segment cable networks and can be nested with deep neural networks to form a structured prior model. The present invention also combines the moving window strategy y(t) to divide the time period, uses short-term observation signals for independent positioning, and then integrates the inference results of multiple time periods to further improve stability and avoid misjudgment caused by occasional interference.

[0074] Signal denoising involves selecting subbands in the signal related to partial discharges and applying a soft threshold denoising method to remove background noise.

[0075] Specifically, the system first performs energy characteristic statistics on the wavelet packet coefficients in each frequency band, and automatically selects several sub-bands that are highly correlated with partial discharge activities by comparing the degree of energy concentration, frequency distribution density, and peak response. This screening process combines historical sample comparison and empirical model evaluation, has a certain structural adaptability, and can maintain a high judgment accuracy under different cable types. After selecting the target sub-band, the system introduces a soft threshold denoising strategy to compress its coefficients. The soft threshold method is a signal cleaning method based on amplitude clipping, and does not introduce sudden jumps at the signal boundary like the hard threshold method. It sets a noise reference threshold to compress the fluctuation coefficients less than the value to zero as a whole, and performs amplitude backoff on the part greater than the value, thereby achieving smooth suppression of background noise;

[0076] To accommodate the time-varying interference intensity of actual cable signals, the present invention employs a dynamic calculation mechanism to determine the noise reduction threshold for each subband. This mechanism adjusts the cleaning amplitude based on factors such as the discreteness, variation intensity, and signal length of each set of wavelet coefficients. This significantly increases the probability of retaining low-amplitude discharge signals and avoids signal attenuation caused by excessive cleaning. Subband coefficients, after soft thresholding, are then resynthesized into a time-domain signal using an inverse wavelet packet reconstruction method. Compared to the original input signal, this reconstruction significantly reduces background noise in the low-frequency interference band while preserving the high-frequency spike structure, resulting in clearer local discharge characteristics. This signal serves as direct input for the subsequent positioning and identification modules.

[0077] The decision function of the support vector machine classifier is:

[0078]

[0079] Among them, x i is the training data, y i is the category label, K(x i ,x) is the kernel function and b is the bias term.

[0080] Specifically, after completing the noise reduction process and obtaining a clean, high-signal-to-noise ratio time-domain signal, the present invention further introduces a classification and recognition mechanism to determine the type of partial discharge signal. Depending on its occurrence mechanism, partial discharge may manifest as different forms such as tip discharge, creepage, holes, surface or joint degradation, and these forms have significant differences in their time and frequency domain performance. Therefore, it is difficult to accurately identify them based solely on peak amplitude or a single energy parameter. First, several statistical feature parameters are extracted from the denoised signal. The SVM classifier used has the following core discriminant function:

[0081]

[0082] Among them, x is the input feature vector to be classified; x i Represents the i-th sample vector activity y in the training set i is the category label of the corresponding sample (for example, tip discharge is 1, surface discharge is -1, etc.); α i is the Lagrange multiplier coefficient obtained by model learning; K(x i ,x) is the kernel function used to calculate sample similarity in high-dimensional space; b is the bias constant. Different from the posterior probability distribution in the aforementioned Bayesian positioning module, the discriminant model here focuses on the "maximum boundary interval" principle, that is, the hyperplane that maximizes the distance between samples of different categories is selected as the classification boundary. This strategy can effectively avoid overfitting and adapt to the nonlinear changes of signal waveforms under real working conditions. The kernel function K(x i,x), the radial basis function (RBF kernel) is preferred. This function performs well when processing high-dimensional features and nonlinear boundaries. If the data distribution in the subsequent actual detection environment shows a linear or quasi-linear trend, it can also be replaced by a linear kernel or a polynomial kernel.

[0083] During the training process, the system uses historical defect samples to construct a training set for model learning. After the training is completed, the model will be solidified for on-site detection. In order to enhance the generalization ability of the model, the present invention also introduces a cross-validation mechanism to dynamically adjust the model parameters (including penalty coefficient and kernel width, etc.) according to the accuracy of the validation set to ensure that the final deployed model is stable under different working conditions. Finally, the classification module will output the judgment result, that is, which type of defect morphology the currently detected partial discharge signal belongs to. This result will be used as one of the inputs of the subsequent severity assessment module. The SVM classifier is not an independent running module in the present invention, but has a causal relationship with the wavelet packet feature extraction and the aforementioned Bayesian positioning process. By classifying and identifying the characteristic waveforms in the positioning area, "space-type" joint modeling can be achieved, providing higher-dimensional judgment support for the subsequent diagnostic process.

[0084] The fuzzy logic reasoning system evaluates the severity of cable defects based on the input signal characteristics. The fuzzy rules are expressed as:

[0085] Severity(x)=IF x1is A1AND x2is A2THEN severity is S;

[0086] Among them, x1, x2 are signal features, A1, A2 are fuzzy sets, and S is the severity assessment result.

[0087] Specifically, after completing the classification and identification of partial discharge signals, the system does not directly output the final result. Instead, it further introduces a fuzzy logic reasoning mechanism to assess the severity level of the detected defects. This mechanism is designed to help operation and maintenance personnel determine the urgency of the defect and assist them in deciding whether to immediately shut down the system for maintenance, conduct delayed observation, or conduct regular inspections. The fuzzy logic reasoning system establishes a set of highly interpretable rule bases based on multi-dimensional input signal features to simulate the judgment logic of human experts on defect waveforms. This type of method has a strong knowledge embedding capability and can transform a large amount of previously accumulated maintenance experience into a reusable reasoning model.

[0088] Input signal characteristics include: signal peak, average amplitude, pulse repetition rate, energy distribution, duration, and frequency band distribution trend. The system has completed the acquisition and normalization of these parameters in the feature extraction stage, so they can be directly used as input to the fuzzy system. Each feature dimension is divided into multiple fuzzy sets, such as "low", "medium", "high" or "weak", "obvious", and "violent". These fuzzy sets are not divided using strict boundaries, but are modeled using membership functions so that each input value has a certain membership in multiple categories. This processing method is closer to human language judgment and helps to deal with the ambiguity and uncertainty of signal characteristics;

[0089] Taking a typical fuzzy rule as an example, the system can construct reasoning conditions in the following semantic expression form: if the peak amplitude is "high" and the signal duration is "long", the severity is "high"; if the pulse repetition rate is "medium" and the energy distribution tends to the medium frequency area, the severity is judged to be "medium"; if the signal amplitude is small and only appears instantaneously within a certain period of time, and the frequency distribution is concentrated in the high frequency band, it is judged to be "low severity" or "negligible level".

[0090] These rules are either manually set or automatically generated based on sample training. During system operation, all active rules are evaluated in parallel and integrated into the fuzzy synthesis module. The final output value is a continuous severity scale, which can be divided into several levels according to actual needs, such as three levels (low / medium / high) or five levels (normal / mild / moderate / severe / extreme). To adapt to actual field operations, the evaluation results can not only be used for alarm judgment, but can also be directly connected to the monitoring system or remote inspection platform for long-term trend analysis.

[0091] After assessing the severity of the cable defect, proceed as follows:

[0092] The real-time monitoring and data feedback steps include dynamically adjusting detection parameters to adapt to different working conditions of live cable equipment and automatically updating the parameters of the signal processing algorithm during the detection process.

[0093] Specifically, the system first collects the real-time operating parameters of the cable, such as the load current change trend, ambient temperature, humidity, cable surface potential change, frequent operation conditions, etc. These operating variables will affect the local discharge activity and will also indirectly interfere with the signal propagation process. In order to cope with such changes, the present invention sets a dynamic frequency adjustment mechanism in the signal excitation part. This mechanism is based on the reference frequency obtained by initial optimization. When it monitors the drastic fluctuation of the external state (such as high temperature, overload, humidity), it can fine-tune the excitation frequency appropriately through the feedback adjustment interface to ensure that the signal propagation performance remains near the optimal range;

[0094] The signal acquisition and processing process also has feedback update capabilities. For example, when the system continuously receives signal samples with unclear features and high positioning errors over a period of time, it will trigger the wavelet packet analysis module to re-evaluate whether the wavelet basis function currently in use is optimal, and whether there are problems such as energy leakage or redundant features. If the conditions are met, the system will try to replace or reinitialize the wavelet basis selection strategy to ensure that the current processing capability matches the cable environment. In terms of the support vector machine classifier, if the system detects that the current input signal characteristics have deviated significantly from the training sample distribution boundary, it may trigger a decrease in model confidence. At this time, it can enter the retraining or rapid parameter adjustment process to make local corrections to the model kernel function parameters and support vector distribution. This process is automatically executed in the background and does not affect real-time output;

[0095] In addition, the fuzzy logic system also has the ability to revise rules. The system compares and analyzes historical classification and evaluation records at regular intervals. If it finds that the same type of defect has different consequences in similar environments (such as some "medium severity" events eventually develop into "high-risk faults"), it will adjust the severity membership value or priority of this type of feature combination accordingly to ensure that the practicality of the rule base gradually increases over time. The feedback mechanism runs through the entire life cycle of the detection system, with the ability to operate continuously and adjust in real time, and is applicable to a variety of cable system structures and operating scenarios. Its goal is not to replace manual judgment, but to provide a stable and traceable auxiliary mechanism to enhance the adaptive ability of fault identification when engineering conditions are complex and failure modes are uncertain.

[0096] The signal extraction and processing system includes a high-frequency current sensor, a data acquisition device, a signal processing unit and a display terminal, wherein the signal processing unit dynamically adjusts the frequency range and signal processing algorithm of the high-frequency current signal according to real-time data.

[0097] Specifically, the signal extraction and processing system consists of the following components: a high-frequency current sensor, a data acquisition device, a signal processing unit, and a display terminal. Each module is connected via wired or wireless data channels, forming a fast-response, collaborative detection platform.

[0098] Among them, high-frequency current sensors are usually installed near the cable outlet or connection node. The sensor has the ability to respond to megahertz-level current pulses and can capture high-frequency electromagnetic leakage signals caused by local discharge activities in a non-contact manner. The sensor adopts a shielded coupling structure and an embedded front-end anti-saturation protection circuit to ensure high sensitivity and high linearity output capabilities even when the cable load fluctuates greatly. After the signal is output from the sensor, it enters the data acquisition device. The data acquisition module uses a high-speed A / D conversion chip, the sampling rate can be set according to the working frequency band range, and supports multi-channel synchronous sampling. A trigger mechanism is set inside the system to automatically start data capture when the signal suddenly changes or spikes appear, ensuring that the information is fully retained at critical moments;

[0099] After data collection is completed, all raw data will be transferred to the signal processing unit for analysis and processing. The signal processing unit in the present invention is not only a static calculation module, but also has real-time feedback adjustment capabilities. The system independently determines whether the current signal processing strategy needs to be adjusted based on parameters such as the fluctuation trend of the signal in different time windows, the self-noise ratio level, and the spectrum distribution. If it is detected that the high-frequency energy component of the cable response signal continues to drift or the sideband expands, the processing unit will send a feedback command to the frequency control module to dynamically adjust the frequency range of the high-frequency current excitation to make it closer to the current system propagation characteristics. This adjustment process is not a frequency point jump, but a frequency band slip based on the step method or prediction function to ensure that the signal excitation does not cause system resonance or response dead zone.

[0100] In terms of signal processing algorithms, the processing unit will also automatically update key algorithm configurations such as wavelet packet parameters, Bayesian prior distribution, and SVM kernel function configuration based on the number of valid signals input in real time, the noise suppression effect, and the detection confidence level to maintain highly adaptive operation. After each parameter adjustment, the system will record the adjustment path and evaluate it against the recognition results. If the performance has not improved, a rollback recovery will be performed; the final processing results (such as defect type, location, severity level, etc.) will be displayed in real time on the display terminal. The display terminal can be a local touch screen or a human-machine interface system on a remote platform, supporting historical record playback, trend analysis, alarm prompts, and other functions to facilitate timely response by operation and maintenance personnel.

[0101] The live cable equipment is a 10kV distribution line, and the partial discharge of the cable is detected and diagnosed by the method to judge the operation safety and stability of the cable.

[0102] Specifically, when the equipment is energized, the high-frequency current generated during partial discharge is used to transmit signals. By analyzing the signals, the location and state of the partial discharge can be determined. This will not damage the insulation of the cable and can be detected under normal cable operation. The high-frequency current detection method is used to conduct on-site detection of the equipment status quantity, and the cable defects are detected by detecting the magnitude of the electromotive force induced by the magnetic field in the sensor. This plays a preventive role in timely discovering potential hidden dangers of the equipment and avoiding tripping faults, and can effectively improve the distribution network operation and maintenance level. Combined with the above fuzzy rules, local defects are judged. The connection of the point detection equipment is shown in Figure (2). When partial discharge occurs in power equipment, high-frequency pulse current is usually generated on its grounding lead or other ground potential connection line. The high-frequency current sensor is used to detect the high-frequency pulse current signal flowing through the grounding down conductor or other ground potential connection line to realize the live detection of partial discharge of power equipment. The principle of the high-frequency partial discharge detection technology of power equipment is shown in Figure (3). An open-type caliper positioning special sensor is installed on the shielded grounding wire of the cable terminal. The sensor can be inserted without untying the grounding wire; the sensor wiring is led out to the live detector interface, and the detector processes the information and compares it with the database for analysis and calculation to realize the discrimination and detection of partial discharge point defects;

[0103] The principle of collecting partial discharges during live detection is based on the HFCT high-frequency partial discharge sensor based on the Rogowski coil. The HFCT is clamped on the grounding wire of the cable as shown in Figure (4) to monitor the pulse current signal generated by its partial discharge, thereby obtaining partial discharge information of the monitored equipment. The HFCT partial discharge sensor consists of a magnetic core, a Rogowski coil, a filtering and sampling unit, and an electromagnetic shielding box. The coil is wound on a magnetic core with high magnetic permeability at high frequencies. The design of the filtering and sampling unit takes into account the requirements of measurement sensitivity and signal response frequency band. The partial discharge sensor also has a built-in signal conditioning module, which mainly amplifies, filters, and detects the signal coupled to the sensor, so that the high-frequency pulse signal can be effectively and completely monitored by the signal acquisition and processing unit. Through this method, the working requirements of partial discharge detection of cables during live detection are met.

[0104] Among them, the wide-band current sensor based on the Rogowski coil structure is essentially a current transformer, with a ring oxide core as the primary side and a multi-turn coil as the secondary side, as shown in Figure (5). When a high-frequency pulse flows through the coil, an alternating magnetic flux appears on the primary side, and the magnetic flux induced by each turn of the coil on the secondary side is proportional to the magnitude of the pulse current. The change in magnetic flux generates an electromotive force on the secondary side, which is proportional to the magnitude of the pulse current. The equivalent circuit diagram of the Rogowski coil is shown in Figure (6). It can be seen from the schematic diagram of the Rogowski coil structure that an integral resistor is connected in parallel to the output end of the self-integrating Rogowski coil, and a signal proportional to i in the current-carrying conductor can be obtained. Rs is the equivalent resistance of the coil, Cs is the equivalent stray capacitance of the coil, and R is the integral resistance of the coil, which forms an integral circuit with the equivalent inductance Ls of the coil.

[0105] Therefore, the sensor's sensitivity is inversely proportional to the number of coil turns N and directly proportional to the integrating resistor R. The sensor signal's frequency response is related to the integrating resistor R, the coil inductance Ls, and the coil resistance Rs. After the core size and material are determined, the coil inductance Ls is only related to the number of coil turns N. Increasing the number of coil turns N and reducing the integrating resistor R can expand the sensor's frequency band, but this will reduce sensor sensitivity.

[0106] While embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions, and variations may be made to these embodiments without departing from the principles and spirit of the invention, and that the scope of the invention is defined by the appended claims and their equivalents.

Claims

1. A method for detecting defects in live cable equipment using a high-frequency current detection method, characterized in that: The following steps are involved: Exciting the live cable device to generate a high-frequency current signal, wherein the frequency of the high-frequency current signal is determined to be an optimal frequency by minimizing signal propagation loss; The current signal on the cable equipment is obtained by a high-frequency current sensor, and the signal is subjected to multi-scale time-frequency analysis through wavelet packet transform; Performing noise reduction processing on the transformed signal to remove interference signals and extract partial discharge signals; Locate the partial discharge signal based on Bayesian theorem to determine the location of the partial discharge source; Support vector machine is used to classify partial discharge signals and determine the defect type, and the severity of the defect is evaluated based on a fuzzy logic reasoning system.

2. The method for detecting defects in live cable equipment using a high-frequency current detection method according to claim 1, characterized in that: The optimal frequency is optimized by the least squares method, and the objective function of minimizing the signal propagation loss is: Among them, α(f i ) is the signal propagation loss at the i-th frequency point, α ideal For the ideal frequency response.

3. The method for detecting defects in live cable equipment using a high-frequency current detection method according to claim 1, characterized in that: The wavelet packet transform decomposes the signal at multiple levels to obtain coefficients in different frequency bands, and selects the optimal wavelet packet basis function according to the Kolmogorov complexity theory to maximize the signal compression rate and minimize noise interference.

4. The method for detecting defects in live cable equipment using a high-frequency current detection method according to claim 1, characterized in that: In the Bayesian positioning step, the posterior probability of the PD source location is calculated using the following Bayesian formula: Where P(x0|y(t)) is the posterior probability, P(y(t)|x0) is the likelihood function, which represents the probability of signal y(t) at a given location x0, P(x0) is the prior probability, and P(y(t)) is the marginal likelihood.

5. The method for detecting defects in live cable equipment using a high-frequency current detection method according to claim 1, characterized in that: The signal noise reduction process includes selecting a subband related to partial discharge in the signal and applying a soft threshold denoising method to remove background noise.

6. The method for detecting defects in live cable equipment using a high-frequency current detection method according to claim 1, characterized in that: The decision function of the support vector machine classifier is: Among them, x i is the training data, y i is the category label, K(x i ,x) is the kernel function and b is the bias term.

7. The method for detecting defects in live cable equipment using a high-frequency current detection method according to claim 1, characterized in that: The fuzzy logic inference system evaluates the severity of cable defects based on the input signal characteristics. The fuzzy rules are expressed as: Severity(x)=IF x1is A1AND x2isA2THEN severityis S; Among them, x1, x2 are signal features, A1, A2 are fuzzy sets, and S is the severity assessment result.

8. The method for detecting defects in live cable equipment using a high-frequency current detection method according to claim 7, characterized in that: After assessing the severity of the cable defects, proceed as follows: The real-time monitoring and data feedback steps include dynamically adjusting detection parameters to adapt to different working conditions of live cable equipment and automatically updating the parameters of the signal processing algorithm during the detection process.

9. The method for detecting defects in live cable equipment using a high-frequency current detection method according to claim 1, characterized in that: The signal extraction and processing system includes a high-frequency current sensor, a data acquisition device, a signal processing unit and a display terminal, wherein the signal processing unit dynamically adjusts the frequency range and signal processing algorithm of the high-frequency current signal according to real-time data.

10. The method for detecting defects in live cable equipment using a high-frequency current detection method according to claim 1, characterized in that: The live cable equipment is a 10kV distribution line, and the method is used to detect and diagnose partial discharge of the cable to determine the operating safety and stability of the cable.

Images