DPP algorithm-based active injection type cable fault early warning positioning method and system

Abstract

The application belongs to the technical field of power cable fault detection, and provides a DPP algorithm-based active injection type cable fault early warning positioning method and system, wherein the DPP algorithm-based active injection type cable fault early warning positioning method generates an active injection signal by adaptively matching cable baseline characteristics, synchronously collects multi-measurement-point reflection signals and extracts a multi-dimensional feature set, introduces a deterministic point process (DPP) algorithm to complete de-redundancy optimization of a feature subset, performs iterative accurate re-measurement on a suspected defect section in combination with an initial positioning result, and finally outputs a fault position and an early warning level; the application can realize early warning and accurate positioning of cable insulation deterioration, is suitable for power cables of different lengths and different aging degrees, reduces the misjudgment risk caused by environmental interference and data redundancy, and improves positioning resolution and early warning timeliness.

Description

Technical Field

[0001] This invention belongs to the field of power cable fault detection technology, and particularly relates to a method and system for active injection-based cable fault early warning and location based on the DPP algorithm. Background Technology

[0002] With the continuous increase in the cable coverage rate of urban power distribution networks, the operational status of underground power cables directly affects the reliability of power supply. In existing technologies, active injection detection is one of the mainstream technical approaches for cable fault location. For example, Chinese Patent (Publication No.: CN120161280A) discloses an online method for locating local defects in high-voltage cables based on active excitation reflection spectra. This method determines the cable length through online frequency sweeping, generates an active excitation signal after matching the optimal test frequency, injects it into the cable core via electromagnetic induction, and then determines the location of the local defect through time-frequency analysis of the reflected signal. This scheme achieves uninterrupted power supply detection, but it still has several technical limitations in practical applications.

[0003] The problems with existing technologies are mainly reflected in the following four aspects:

[0004] Insufficient adaptability of injected signal parameters and limited ability to detect early defects: Existing active injection methods mostly adopt fixed frequency or single-round frequency sweeping strategies, which cannot dynamically optimize injection parameters for cables with different service years and different aging degrees. For weak impedance discontinuities formed by early insulation degradation, the injection signal with fixed parameters is prone to excessive attenuation or insufficient excitation, resulting in a low signal-to-noise ratio of reflected signals, making it difficult to distinguish between environmental noise and real defect responses, and resulting in a high rate of missed detection of early hidden dangers.

[0005] High redundancy of multi-point data and interference of feature coupling with positioning accuracy: In order to improve coverage, multiple acquisition points are usually set up in the field to output multi-dimensional features such as time domain and frequency domain. However, there is a strong correlation and redundancy between features of different measurement points and different dimensions. Traditional threshold method or simple weighted fusion cannot effectively decouple feature coupling. It is easy to misjudge the signal offset caused by load fluctuation and environmental temperature change as defects, resulting in deviation of positioning results and difficulty in controlling the misjudgment rate to a low level.

[0006] It mainly focuses on post-fault fault location and lacks early warning capabilities: Existing active injection technology focuses on location locking after the fault occurs and relies on the obvious impedance change formed at the fault point to generate reflected signals. For the intermediate stage of slow insulation deterioration, it lacks a quantitative status assessment and trend prediction mechanism, and cannot issue early warning before the fault occurs. Maintenance personnel can only passively repair, making it difficult to achieve condition-based maintenance and preventive maintenance, thus limiting the room for improvement in power supply reliability.

[0007] The single-round detection mode is fixed, and it is difficult to improve the positioning resolution as needed: Existing solutions mostly adopt the detection mode of single-round full-area scanning. The density of measurement points and signal parameters are uniform throughout the process. If the positioning accuracy of local areas is to be improved, the hardware sampling rate and measurement point density must be increased as a whole, which greatly increases the detection cost and computational load. It is impossible to perform directional fine retesting on suspected sections initially located, and it is difficult to balance positioning resolution and detection efficiency.

[0008] Therefore, a proactive cable fault early warning and location method and system based on the DPP algorithm is needed to solve the above problems. Summary of the Invention

[0009] The purpose of this invention is to provide a method and system for active injection-based cable fault early warning and location based on the DPP algorithm, so as to solve the problems mentioned in the background art.

[0010] To achieve the above objectives, the present invention provides the following technical solution: a DPP algorithm-based active injection cable fault early warning and location method, comprising the following steps:

[0011] S1. Collect the steady-state operating parameters and structural ledger parameters of the cable under test, construct the cable baseline feature library, and adaptively match the frequency, amplitude and waveform parameters of the actively injected signal based on the baseline features;

[0012] S2. Inject the matched active detection signal into the head end of the cable under test, and simultaneously collect the reflected signal at multiple test points along the cable laying path to extract a multi-dimensional initial feature set including time domain, frequency domain, and time-frequency domain.

[0013] S3. Construct the DPP kernel matrix, and use the DPP sampling algorithm to select the approximately optimal de-redundant feature subset with both discriminative power and independence from the initial feature set to generate the de-redundant feature matrix.

[0014] S4. Based on the redundancy-removing feature matrix and the cable transmission line model, calculate the location of impedance discontinuities at each measuring point, and combine the defect confidence judgment of multi-feature fusion to complete the initial fault location and mark the suspected defect section.

[0015] S5. Adjust the injection signal parameters and measurement point density for suspected defective sections, perform multiple rounds of iterative injection and acquisition, correct the location results and calculate the defect deterioration trend, and output the fault location results and early warning level.

[0016] A complete closed-loop detection system has been established, encompassing signal excitation, feature processing, and location-based early warning, adaptable to the inspection needs of power cables of varying lengths and aging levels. Through a progressive combination of adaptive injection, feature redundancy removal, and iterative retesting, it can effectively identify weak impedance discontinuities in early insulation degradation, reducing the risk of misjudgments and missed detections caused by environmental interference and data redundancy. Simultaneously, it improves location resolution while controlling detection costs, and outputs quantified degradation trends and early warning levels, extending traditional post-fault location to pre-fault status warning, significantly enhancing the timeliness and proactiveness of cable maintenance.

[0017] Furthermore, the cable baseline characteristics in step S1 include the nominal cable length, conductor material, insulation medium type, service life, load current value, and historical fault records; wherein the historical fault records are used to correct the initial weight of baseline degradation; the actively injected signal adopts a frequency sweep pulse sequence, the frequency covers a wide frequency band, and the amplitude is dynamically adjusted according to the cable insulation tolerance threshold and the end signal-to-noise ratio constraint.

[0018] Furthermore, the multi-source acquisition step synchronously acquires multi-dimensional electrical characteristics through distributed measurement points. This step adopts a multi-point synchronous acquisition architecture, which can simultaneously acquire reflected signals from different locations on the cable. By extracting multi-dimensional features, it covers information in the time domain, frequency domain, and time-frequency domain, providing a sufficient data foundation for subsequent feature screening and location calculation, and reducing the detection risk caused by anomalies in single measurement point signals.

[0019] By dynamically mapping the baseline characteristics across all dimensions with the injected signal, the frequency and amplitude of the excitation signal can be precisely adjusted according to the actual service condition of the cable, ensuring that the signal remains within the optimal signal-to-noise ratio range throughout the entire process. The degradation weight correction mechanism of historical fault records can specifically strengthen the signal excitation intensity in historical defect sections, further improving the detection capability of early minor degradation; the amplitude, combined with the insulation tolerance threshold and the end signal-to-noise ratio dual constraints, ensures detection safety while also taking into account the attenuation compensation requirements of long-distance cables, significantly improving the detection adaptability of cables in different service conditions.

[0020] Furthermore, in step S2, multiple measuring points are arranged along the cable axis, with denser arrangement in high-fault sections and equidistant arrangement in stable sections; each measuring point synchronously acquires voltage reflection waveform, current traveling wave waveform, and partial discharge pulse signal; the extracted initial features include reflected wave delay, amplitude attenuation rate, harmonic distortion rate, wavelet packet energy entropy, and phase offset.

[0021] By employing a differentiated deployment strategy of increasing the density of sensors in high-fault sections and maintaining equal spacing in stable sections, the sensing density at vulnerable locations such as cable joints and pipe crossings can be increased while controlling hardware investment costs. Simultaneous acquisition of multiple electrical signals and multi-dimensional time-frequency domain feature extraction provide a comprehensive data foundation for subsequent feature screening and location calculations, reducing the risk of detection failure caused by anomalies in single-point signals. Furthermore, the rich feature dimensions provide ample data support for subsequent defect type identification and condition assessment.

[0022] Furthermore, the DPP kernel matrix in step S3 consists of a quality term and a similarity term. The quality term corresponds to the fault discrimination of a single-dimensional feature, and the similarity term corresponds to the Pearson correlation coefficient between different features. DPP sampling is achieved through eigenvalue decomposition combined with a greedy strategy, and the output dimension is... The approximate optimal feature subset; The value is adaptively adjusted based on the initial feature dimension and the degree of anomaly detection, and the value ranges from 30% to 60% of the initial feature dimension.

[0023] The DPP kernel matrix, constructed based on quality and similarity terms, can effectively eliminate redundant features and decouple interference between features while preserving defect discrimination capabilities. Compared to traditional screening methods, it can simultaneously ensure feature independence and representativeness. The mechanism of adaptive adjustment of the k value according to the degree of anomaly can improve computational efficiency in low-anomaly scenarios and preserve detailed information in high-anomaly scenarios. The 30%~60% dimensionality constraint can balance the dimensionality reduction magnitude and information integrity, significantly reducing misjudgments caused by environmental and load fluctuations and improving the robustness of the localization results.

[0024] Furthermore, the iterative retesting in step S5 adopts a three-level strategy of gradually narrowing the range: after the initial positioning in the first round, the length of the suspected section is compressed to 1 / 4 to 1 / 3 of the length of the section covered by the previous round of detection; in the second round, the density of the measuring points is increased and the frequency of the injected signal is increased; in the third round, narrowband pulses are introduced to accurately lock the defect boundary; at the same time, the defect deterioration rate is fitted based on the multi-round detection data, and the warning level threshold is matched.

[0025] The three-stage iterative retesting strategy, which gradually shrinks the measurement range, sets a segment shrinkage ratio of 1 / 4 to 1 / 3 based on the propagation characteristics of traveling wave positioning errors. This achieves an optimal balance between positioning accuracy and detection efficiency. Without requiring a full-domain hardware upgrade, simply targeting suspected segments with increased measurement points and signal frequency can reduce positioning errors from the meter level to the decimeter level. Simultaneously, by fitting the degradation rate to multiple rounds of data and matching warning thresholds, early warning of insulation degradation can be achieved, supporting the implementation of condition-based maintenance. This solves the problem of balancing accuracy and cost in traditional single-round detection.

[0026] The active injection cable fault early warning and location system based on the DPP algorithm is applied to any of the above-mentioned active injection cable fault early warning and location methods based on the DPP algorithm, including an active signal injection unit, a multi-channel sensor acquisition unit, a baseline feature matching unit, a DPP feature optimization unit, and a fault location and early warning unit.

[0027] The output of the baseline feature matching unit is connected to the control terminal of the active signal injection unit for sending injection signal parameters.

[0028] The output of the active signal injection unit is coupled to the power cable under test and is used to inject an active detection signal into the cable.

[0029] The acquisition end of the multi-channel sensing acquisition unit is coupled to each measurement point of the power cable under test, and the output end is connected to the input end of the DPP feature optimization unit for uploading the acquired waveform and initial features.

[0030] The output of the DPP feature optimization unit is connected to the input of the fault location and early warning unit, and is used to output the redundancy removal feature matrix.

[0031] The feedback output terminal of the fault location and early warning unit is connected to the control terminals of the active signal injection unit and the multi-channel sensor acquisition unit, respectively, and is used to issue parameter adjustment instructions for iterative retesting.

[0032] This system adopts a closed-loop control architecture. Each unit forms a complete link according to the signal flow, from parameter matching, signal injection, data acquisition, feature processing to positioning and early warning. Simultaneously, it automatically issues iterative retesting commands through the feedback link, completing the entire detection process without requiring repeated manual parameter adjustments. The system is adaptable to various field operation scenarios such as live-line inspections and periodic checks. The coordinated operation of each unit can stably output accurate positioning results and early warning information. The system boasts a high degree of automation and standardized testing procedures, effectively reducing human error and improving the standardization level of cable testing.

[0033] Furthermore, the active signal injection unit includes a cascaded signal generator, a power amplifier, and a coupled injection probe; the coupled injection probe adopts a clamp-on electromagnetic coupling structure for fitting onto the outside of the cable phase wire; the active signal injection unit has a built-in impedance matching network for achieving dynamic matching between the signal output impedance and the cable characteristic impedance;

[0034] The cascaded signal generation, amplification, and injection structure ensures the stability and adjustability of the output signal. The clamp-on electromagnetic coupling non-contact injection method allows signal injection without power interruption or disassembly of the cable insulation, making it suitable for uninterrupted online testing scenarios. The built-in impedance matching network dynamically matches the cable's characteristic impedance, significantly reducing interference from reflections at the injection end, improving signal coupling efficiency, ensuring the signal-to-noise ratio of weak defect reflections, and further enhancing the testing adaptability for cables of different lengths.

[0035] Furthermore, the multi-channel sensing acquisition unit includes multiple fixed distributed acquisition terminals, a mobile acquisition terminal, and a synchronization clock module; the synchronization clock module is connected to each acquisition terminal in a star configuration to provide a unified time reference; each acquisition terminal has a built-in high-frequency current sensor and voltage sensor for acquiring electrical signals; the synchronization clock module supports both fiber optic time synchronization and GPS time synchronization modes, with fiber optic time synchronization corresponding to the fixed acquisition terminals and GPS time synchronization corresponding to the mobile acquisition terminals.

[0036] The architecture combining fixed and mobile terminals balances the continuity of routine monitoring across the entire area with the flexibility of iterative retesting and encryption. The fiber optic + GPS dual-time synchronization solution achieves sub-microsecond time synchronization, with fiber optic time synchronization ensuring long-term stable operation of fixed terminals and GPS time synchronization meeting the temporary and rapid networking needs of mobile terminals, providing an accurate time reference for traveling wave delay calculation. Combined with differentiated deployment and edge preprocessing capabilities, it can control transmission costs while ensuring sensing density and data timeliness in high-fault areas.

[0037] Furthermore, the fault location and early warning unit includes a transmission line simulation module and a degradation trend fitting module connected in series; the transmission line simulation module is used to correct signal attenuation and delay errors, and the degradation trend fitting module is used to output four levels of early warning results: normal, attention, early warning, and alarm.

[0038] The transmission line simulation module can correct signal attenuation and delay errors based on transmission line theory, effectively improving the accuracy of location calculations. The degradation trend fitting module can quantify the insulation degradation rate by combining multiple rounds of historical data, outputting a four-level warning result, with each level corresponding to a clear degree of degradation. The architecture of the two modules connected in series enables seamless integration of location calculations and condition assessment. The output results can be directly linked to the operation and maintenance process, providing clear handling guidelines for operation and maintenance personnel and improving the operability and implementation of cable condition-based maintenance.

[0039] Furthermore, it also includes a cloud interaction unit; the cloud interaction unit is bidirectionally connected to the fault location and early warning unit, and is used to store historical detection data, update the baseline feature library, and push location results and early warning information to the operation and maintenance terminal.

[0040] The cloud-based interactive unit and the fault location and early warning unit communicate bidirectionally, enabling the aggregation of multi-cable inspection data to build a regional status database. Through big data analysis, common degradation patterns of cables in the same batch and environment are identified, which in turn feeds back into the optimization of baseline matching and feature selection models, achieving continuous iteration of system performance. Simultaneously, it supports multi-terminal early warning push notifications and remote management, allowing maintenance personnel to view inspection reports and receive early warning information at any time, effectively improving the management efficiency and response speed of large-scale cable maintenance.

[0041] Compared with the prior art, the beneficial effects of the present invention are:

[0042] This invention improves the adaptability of injected signals and enhances the ability to detect early weak defects: By combining the material, length, aging degree and operating load of the power cable under test with the baseline feature matching unit, the output frequency, amplitude and waveform of the active signal injection unit are dynamically calculated. This mechanism can accurately match the injected signal with the current transmission characteristics of the cable, maximize the amplitude of the reflected signal at the weak defect, improve the signal-to-noise ratio, reduce the probability of missing early insulation degradation, and adapt to the testing needs of cables in different service states. Compared with fixed parameter injection schemes, it has a significantly improved ability to detect weak impedance discontinuities in the early stage of insulation degradation. Combined with targeted excitation of historical fault records, the detection rate of weak degradation in historical defect sections can be significantly improved.

[0043] This invention reduces feature redundancy and coupling interference, improving the stability of positioning results: By constructing a kernel matrix including quality and similarity terms through the DPP feature optimization unit, a near-optimal subset with both discriminative power and independence is selected from the high-dimensional feature set output by the multi-channel sensing acquisition unit. This can eliminate redundant features, decouple feature coupling, and retain the feature dimensions that are most sensitive to defects and do not overlap, reducing misjudgments caused by environmental disturbances and load fluctuations, and improving the robustness and accuracy of positioning results. Compared with the traditional feature weighted fusion scheme, it can effectively reduce the positioning deviation caused by high-dimensional feature coupling. After redundancy removal by DPP, the feature dimension can be significantly reduced, and the defect misjudgment rate is significantly reduced.

[0044] This invention constructs a degradation trend assessment system to achieve early warning of faults: by using the degradation trend fitting module of the fault location and early warning unit, and combining multiple rounds of historical detection data to fit the insulation degradation rate, multiple early warning levels are divided. It can issue early warning signals based on trend data before the insulation degradation develops into an obvious fault, supporting maintenance personnel to carry out condition-based maintenance, transforming fault handling from passive emergency repair to proactive prevention, improving the operational reliability of the power supply system, and making up for the functional shortcomings of traditional detection technology that can only locate faults that have already occurred.

[0045] This invention employs an iterative partitioned retesting approach, balancing detection efficiency and positioning resolution. Through a range-shrinking strategy for fault location and early warning units, after the initial full-domain positioning, the density of measurement points and signal resolution are increased only for suspected defective sections. Multiple rounds of refined retesting are then conducted in a targeted manner. Without requiring a full-domain hardware upgrade, the positioning resolution of local defects can be significantly improved while controlling overall detection time and computational load. This achieves a balance between detection efficiency and positioning accuracy, resolving the contradiction between accuracy and cost in traditional single-round full-domain detection. A 1 / 4 to 1 / 3 section shrinkage ratio can converge the positioning error from the meter level to the decimeter level within a limited number of iterations, and the detection time is far shorter than that of high-precision full-domain detection schemes.

[0046] To more clearly illustrate the structural features and effects of the present invention, the present invention will be described in detail below with reference to the accompanying drawings and specific embodiments. Attached Figure Description

[0047] Figure 1 This is a schematic diagram of the overall process of the present invention;

[0048] Figure 2 This is a block diagram of the overall architecture of the present invention.

[0049] In the figure: 1. Active signal injection unit; 2. Multi-channel sensor acquisition unit; 3. Baseline feature matching unit; 4. DPP feature optimization unit; 5. Fault location and early warning unit; 6. Cloud interaction unit; 7. Power cable under test; 8. Coupled injection probe; 9. Distributed acquisition terminal; 10. Synchronous clock module; 11. Mobile acquisition terminal. Detailed Implementation

[0050] The present invention will be further described below with reference to embodiments.

[0051] The following embodiments are used to illustrate the present invention, but should not be used to limit the scope of protection of the present invention. The conditions in the embodiments can be further adjusted according to specific conditions, and simple improvements to the method of the present invention under the premise of the concept of the present invention are all within the scope of protection claimed by the present invention.

[0052] Example 1

[0053] like Figure 1-2 As shown, this embodiment of the invention provides an active injection cable fault early warning and location system based on the DPP algorithm, which is applied to the online detection of the power cable 7 under test. The system includes an active signal injection unit 1, a multi-channel sensor acquisition unit 2, a baseline feature matching unit 3, a DPP feature optimization unit 4, a fault location and early warning unit 5, and a cloud interaction unit 6.

[0054] In this embodiment, the power cable 7 under test is a 10kV cross-linked polyethylene insulated cable, laid underground. The active signal injection unit 1 is installed in the ring network cabinet at the beginning of the cable and is connected to the phase wire of the cable in a non-contact manner through the coupling injection probe 8; the multi-channel sensing acquisition unit 2 includes multiple fixed distributed acquisition terminals 9, several mobile acquisition terminals 11 and a set of synchronous clock modules 10, with the fixed acquisition terminals deployed along the cable path at preset intervals; the baseline feature matching unit 3, the DPP feature optimization unit 4, and the fault location and early warning unit 5 are integrated into the field detection host, and the cloud interaction unit 6 is deployed on the power operation and maintenance cloud platform.

[0055] The early warning and location method corresponding to this embodiment is executed according to the following steps:

[0056] The first step is baseline feature matching and parameter injection generation. Baseline feature matching unit 3 extracts parameters from cable ledgers and real-time operational data: nominal cable length. conductor resistivity Relative permittivity of insulating medium Operating life Real-time load current Historical fault records. The characteristic impedance of the cable is calculated based on the above parameters. With signal attenuation coefficient The baseline value and attenuation coefficient satisfy the editable calculation formula:

[0057]

[0058] in The reference attenuation coefficient for the new cable. The aging attenuation increment coefficient is determined by the type of insulation medium and the operating environment; if there are historical fault records, the baseline attenuation coefficient of the corresponding section is multiplied by the preset degradation weight coefficient.

[0059] Then, based on the attenuation coefficient, the center frequency of the injected signal is reverse-matched. With amplitude The matching rules satisfy the editable calculation formula:

[0060]

[0061] in Reference frequency coefficient; amplitude satisfy , To determine the minimum detectable signal amplitude at the end, the voltage value corresponding to the signal-to-noise ratio threshold is selected to ensure that the signal remains detectable when it reaches the end of the cable.

[0062] The second step is multi-point synchronous signal acquisition and feature extraction. The active signal injection unit 1 injects a sweep frequency pulse sequence into the cable. Each fixed distributed acquisition terminal 9, triggered by the synchronous clock module 10, synchronously acquires voltage and current waveforms using a high sampling rate mode. The waveforms at each measurement point are processed to extract multi-dimensional initial features including time domain, frequency domain, and time-frequency domain: first-arrival delay of reflected wave, reflection coefficient amplitude, rising edge slope, harmonic distortion rate, wavelet packet energy entropy, phase offset, signal attenuation rate, pulse width, number of partial discharge pulses, root mean square value, peak factor, etc., forming an initial feature matrix.

[0063] The third step is DPP feature subset selection. DPP feature selection unit 4 receives the initial feature matrix and constructs the DPP kernel matrix. The kernel matrix elements satisfy an editable calculation formula:

[0064]

[0065] in, For the first The quality term for a feature is the inter-class dispersion of that feature between normal and defective samples. A higher value indicates a stronger ability of the feature to distinguish defects. This value is obtained through statistical analysis of historical labeled data. In the case of a cold start scenario without historical labeled data, the physical sensitivity method is used to calculate the initial quality term. ,in Features The amount of change with the degree of insulation degradation The increment of the degree of degradation is derived from the physical mechanism of cable insulation degradation. For the first Wei and Di The Pearson correlation distance of a feature is used to measure the similarity between features; This is a bandwidth parameter used to control the decay rate of similarity items, and it is adaptively adjusted according to the number of feature dimensions.

[0066] Kernel matrix Eigenvalue decomposition is performed, and a greedy sampling strategy is used to output an approximately optimal feature subset of a preset dimension. Core features such as the first arrival delay of the reflected wave, the amplitude of the reflection coefficient, the wavelet packet energy entropy, the phase offset, and the number of partial discharge pulses are retained to generate a deredundant feature matrix. The purpose of this formula is to maximize the differences between features while ensuring the ability to distinguish feature defects, and to reduce the interference of redundant information on subsequent localization.

[0067] Step 4: Initial Fault Location. The fault location and early warning unit 5, based on the redundancy removal feature matrix and combined with the transmission line model, calculates the distance from the starting end to the impedance discontinuity point corresponding to each measurement point. The location calculation formula is in an editable format:

[0068]

[0069] in, This represents the distance from the fault point to the beginning. The propagation speed of the signal in the cable is calculated from the relative permittivity of the cable insulation medium, and the propagation speed satisfies the editable calculation formula:

[0070]

[0071] The speed of light in a vacuum; The time difference between the injected signal and the reflected signal is obtained by phase correction of the first arrival delay in the redundancy removal feature.

[0072] A defect confidence scoring model is constructed based on the remaining multidimensional features: the reflection coefficient amplitude, wavelet packet energy entropy, phase offset, and partial discharge pulse number are normalized and then weighted and summed according to preset weights to obtain the defect confidence score. The value ranges from 0 to 1; only when the confidence level is... When the impedance discontinuity is greater than or equal to a preset threshold, the corresponding point is considered a suspected defect. Based on the calculation results from multiple measurement points, a signal-to-noise ratio weighted fusion method is used to obtain the final initial localization result, with the weights satisfying the following conditions: ,in For the first The signal-to-noise ratio of each measuring point is used to mark the corresponding section as a suspected defect section.

[0073] Step 5: Iterative Retesting and Early Warning Output. For suspected defective sections, the mobile acquisition terminals 11 are deployed with reduced spacing, compressing the length of the suspected sections to a preset proportion of the length of the sections detected in the first round. The center frequency of the injected signal is increased, and a second round of testing is performed. Based on the second round of data, a more accurate defect location is calculated. Simultaneously, historical detection data for that location is retrieved, and an insulation degradation trend curve is fitted. An exponential fitting model is used to calculate the degradation rate, which satisfies the editable calculation formula:

[0074]

[0075] in, Insulation degradation rate; , These are the equivalent insulation resistance values ​​from the two tests conducted before and after the test, respectively, which are calculated from the reflection coefficient. This represents the time interval between two detections. Based on the degradation rate, a warning threshold is matched, the corresponding warning level is output, and the results are synchronized to the cloud interaction unit 6.

[0076] In this embodiment, the synergistic effect of the three progressive technologies is manifested as follows: adaptive injection ensures effective signal excitation, which is the foundation for accurate acquisition; DPP feature optimization eliminates redundant interference, improves the input quality of the localization algorithm, and the selected multi-dimensional features jointly participate in the defect confidence judgment, directly reducing the false judgment rate caused by environmental noise; iterative retesting improves the resolution in a targeted manner based on the initial localization, while realizing trend early warning. The three progressively advance to jointly realize a complete closed loop from signal excitation to feature processing to early warning localization. Compared with single-round fixed parameter detection, it improves the early defect detection rate, localization accuracy, and early warning capability.

[0077] Example 2

[0078] The difference between this embodiment and Embodiment 1 lies in the refinement of the structure of the coupled injection probe 8 in the active signal injection unit 1. The coupled injection probe 8 adopts an open-close clamp-like structure, internally including a ring-shaped magnetic core and an excitation winding. The magnetic core is made of a high-permeability soft magnetic alloy. The probe is sleeved on the outside of the phase wire of the cable, and the detection signal is coupled to the cable conductor through electromagnetic induction without the need to disconnect the power and strip the cable insulation layer.

[0079] In this embodiment, the active signal injection unit 1 also has a built-in impedance matching network, which can automatically adjust the output impedance according to the cable baseline characteristics, so that the probe output impedance matches the cable characteristic impedance. The deviation is controlled within a preset range to reduce reflection interference at the signal injection end. For short-distance cables, the injected signal is mainly a narrow pulse to improve distance resolution; for long-distance cables, the amplitude of the injected signal is appropriately increased and the center frequency is reduced to compensate for signal attenuation over long distances. This embodiment further improves the coupling efficiency and impedance matching accuracy of the injected signal, reduces the interference of self-reflection at the injection end on the detection results, and enhances the adaptability to cables of different lengths.

[0080] Example 3

[0081] The difference between this embodiment and Embodiment 1 lies in the optimization of the synchronization and deployment method of the multi-channel sensing acquisition unit 2. The synchronization clock module 10 adopts a fiber optic + GPS dual timing scheme to provide a unified time reference for all distributed acquisition terminals 9, achieving a synchronization accuracy of sub-microsecond level. Fixed acquisition terminals use fiber optic timing to ensure long-term operational stability; mobile acquisition terminals 11 use GPS timing to meet the needs of rapid networking for temporary on-site deployment. The acquisition terminals are pole-mounted and incorporate high-frequency current transformers and capacitive voltage divider sensors to acquire traveling wave current and ground voltage signals, respectively.

[0082] In this embodiment, the distributed acquisition terminals 9 are not deployed at equal intervals throughout the entire route. Instead, they are densely deployed in high-fault sections such as cable joints, bends, and pipe crossings, while the spacing is appropriately widened in straight sections and environmentally stable areas. Each acquisition terminal has a built-in edge computing module, which can perform preliminary feature extraction and anomaly screening locally, uploading only waveform segments suspected of being abnormal to the detection host, thus reducing data transmission volume. This embodiment improves the sensing density in high-fault sections while controlling hardware costs through differentiated deployment and edge preprocessing, while also reducing data transmission and processing latency.

[0083] Example 4

[0084] The difference between this embodiment and Embodiment 1 is that the calculation method for the mass term of the DPP kernel matrix is ​​optimized. Mass Term Instead of relying solely on historical statistical data, a real-time operating condition correction factor is introduced. The correction factor is calculated based on the load current, ambient temperature, and cable conductor temperature at the time of detection. When the load fluctuates significantly or the ambient temperature is abnormal, the weight of the quality items for feature dimensions that are more affected by the operating conditions is reduced, while the weight of features that are robust to the operating conditions is increased.

[0085] In this embodiment, the distance metric for similarity terms is expanded from a single Pearson correlation coefficient to a maximum information coefficient, enabling the capture of non-linear correlations between features and providing a more comprehensive measure of feature redundancy. (DPP sampling subset dimension) It can adaptively adjust according to the degree of abnormality in the initial screening: when the overall abnormality of the signal is low, Take a smaller value to improve computation speed; when the anomaly degree is high, Take a larger value to retain more detailed features. The adjustment range of the values ​​is strictly constrained to 30% to 60% of the initial feature dimension, avoiding redundancy due to excessively high dimensionality and preventing the loss of key information due to excessively low dimensionality. This embodiment improves the adaptability of DPP feature optimization to complex operating conditions and further reduces the probability of misjudgment caused by load and environmental fluctuations.

[0086] Example 5

[0087] The difference between this embodiment and Embodiment 1 is that the iterative retesting strategy is expanded to adopt a three-level iterative mechanism. The first level is a coarse measurement of the entire area, with large measurement point spacing and low injected signal frequency, used to quickly check the overall cable condition; the second level is a fine measurement of sections, which densifies the measurement points and increases the frequency for suspected sections marked by the coarse measurement; the third level is a point-by-point verification, which uses a mobile high-frequency probe to scan point by point within the small area locked by the fine measurement, injecting narrowband pulse signals to accurately mark the defect boundaries.

[0088] In this embodiment, the segment shrinkage ratio of each iteration is controlled between 1 / 4 and 1 / 3 of the segment length of the previous iteration. This ratio is derived based on the propagation characteristics of traveling wave positioning error: the positioning error in the first round is approximately 1 / 2 of the distance between measurement points. After densifying the measurement points, the error decreases proportionally. A shrinkage ratio of 1 / 4 to 1 / 3 ensures the optimal cost-effectiveness of error reduction and measurement point increment in each round. After each level of detection is completed, the confidence level of the result is automatically evaluated. When the confidence level reaches a preset threshold, the iteration can be terminated without executing all three levels of the process. Simultaneously, the signal parameters and positioning results of each round are automatically recorded during the iteration process, forming a traceability chain for the detection process. This embodiment further balances detection efficiency and positioning accuracy through a graded iteration and confidence level termination mechanism, and the detection depth can be flexibly adjusted according to the actual condition of the cable.

[0089] Example 6

[0090] The difference between this embodiment and Embodiment 1 lies in the refinement of the warning levels and handling strategies. The warning levels are divided into four levels: Normal, Attention, Warning, and Alarm. Each level corresponds to a specific degradation rate threshold and maintenance recommendations. The Normal level indicates stable insulation condition, requiring regular periodic testing; the Attention level indicates a slight degradation trend, requiring shortened testing cycles and increased monitoring; the Warning level indicates a rapid degradation rate, requiring planned power outages for maintenance; and the Alarm level indicates an immediate risk of failure, requiring immediate special inspections and the development of emergency repair plans.

[0091] In this embodiment, the fault location and early warning unit 5 also incorporates a defect type identification module. Based on a combination pattern of optimized features, it identifies different defect types such as insulation aging, moisture absorption, and mechanical damage, providing more targeted references for operation and maintenance. Defect type identification uses a subset of redundant features as input, where wavelet packet energy entropy corresponds to partial discharge features, phase offset corresponds to moisture absorption features, and reflection coefficient amplitude corresponds to mechanical damage features. Type determination is achieved through feature combination mapping. This embodiment improves the practicality of the early warning output, enabling the detection results to be directly integrated into the operation and maintenance process, enhancing the operability of condition-based maintenance.

[0092] Example 7

[0093] The difference between this embodiment and Embodiment 1 is that the system adds a cable joint temperature acquisition function. A passive temperature sensor is installed at each cable joint, and the temperature data is uploaded wirelessly to the multi-channel sensing acquisition unit 2. The temperature data is included as an auxiliary feature in the initial feature set and participates in the DPP feature optimization process.

[0094] In this embodiment, when an abnormal temperature rise occurs, the system automatically triggers an active injection detection, enabling proactive response under abnormal operating conditions without waiting for a fixed detection cycle. The fusion of temperature and electrical characteristics further distinguishes between overheating and discharge defects, improving the accuracy of defect type identification. This embodiment expands the system's perception dimensions, achieving linkage between temperature anomalies and electrical detection, and enhancing the early warning capability for thermally induced defects.

[0095] Example 8

[0096] The difference between this embodiment and Embodiment 1 is that the cloud-based interactive unit 6 has the capability to analyze regional cable groups. The cloud platform aggregates historical testing data from multiple cables, constructs a regional cable status database, and uses big data analysis to uncover common degradation patterns in cables from the same batch and laid in the same environment, feeding back into the optimization of the baseline feature matching model and DPP kernel matrix parameters.

[0097] In this embodiment, the cloud-based interactive unit 6 also supports multi-terminal access. Maintenance personnel can view inspection reports, receive early warning push notifications, and remotely issue inspection commands via mobile phones, tablets, and desktops. Simultaneously, the platform automatically generates periodic status analysis reports to assist managers in formulating maintenance plans and cable replacement plans. This embodiment extends the inspection capability of a single cable to the status management of a regional cable group, enhancing the system's engineering application value and large-scale deployment capabilities.

[0098] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A cable fault early warning and location method based on DPP algorithm, characterized in that, Includes the following steps: S1. Collect the steady-state operating parameters and structural ledger parameters of the cable under test, construct the cable baseline feature library, and adaptively match the frequency, amplitude and waveform parameters of the actively injected signal based on the baseline features; S2. Inject the matched active detection signal into the head end of the cable under test, and simultaneously collect the reflected signal at multiple test points along the cable laying path to extract a multi-dimensional initial feature set including time domain, frequency domain, and time-frequency domain. S3. Construct the DPP kernel matrix, and use the DPP sampling algorithm to select the approximately optimal de-redundant feature subset with both discriminative power and independence from the initial feature set to generate the de-redundant feature matrix. S4. Based on the redundancy-removing feature matrix and the cable transmission line model, calculate the location of impedance discontinuities at each measuring point, and combine the defect confidence judgment of multi-feature fusion to complete the initial fault location and mark the suspected defect section. S5. Adjust the injection signal parameters and measurement point density for suspected defective sections, perform multiple rounds of iterative injection and acquisition, correct the location results and calculate the defect deterioration trend, and output the fault location results and early warning level.

2. The cable fault early warning and location method based on DPP algorithm active injection according to claim 1, characterized in that, The cable baseline characteristics in step S1 include the nominal cable length, conductor material, insulation medium type, service life, load current value, and historical fault records; the historical fault records are used to correct the initial weight of baseline degradation; the actively injected signal adopts a frequency sweep pulse sequence, the frequency covers a wide frequency range, and the amplitude is dynamically adjusted according to the cable insulation tolerance threshold and the end signal-to-noise ratio constraint.

3. The cable fault early warning and location method based on DPP algorithm active injection according to claim 1, characterized in that, In step S2, multiple measuring points are arranged along the cable axis, with denser arrangement in high-fault sections and equidistant arrangement in stable sections; each measuring point synchronously collects voltage reflection waveform, current traveling wave waveform and partial discharge pulse signal. The extracted initial features include reflected wave delay, amplitude attenuation rate, harmonic distortion rate, wavelet packet energy entropy, and phase offset.

4. The active injection cable fault early warning and location method based on the DPP algorithm according to claim 1, characterized in that, The DPP kernel matrix in step S3 consists of a quality term and a similarity term. The quality term corresponds to the fault discrimination of a single-dimensional feature, and the similarity term corresponds to the Pearson correlation coefficient between different features. DPP sampling is achieved through eigenvalue decomposition combined with a greedy strategy, and the output dimension is... The approximate optimal feature subset; The value is adaptively adjusted based on the initial feature dimension and the degree of anomaly detection.

5. The active injection cable fault early warning and location method based on the DPP algorithm according to claim 1, characterized in that, The iterative retesting in step S5 adopts a three-level strategy of gradually narrowing the scope: after the initial positioning in the first round, the length of the suspected section is compressed to 1 / 4 to 1 / 3 of the length of the section covered by the previous round of detection; in the second round, the density of measurement points is increased and the frequency of the injected signal is increased; in the third round, narrowband pulses are introduced to accurately lock the defect boundary; at the same time, the defect deterioration rate is fitted based on the multi-round detection data, and the warning level threshold is matched.

6. A cable fault early warning and location system based on the DPP algorithm, applied to the early warning and location method described in any one of claims 1-5, characterized in that, It includes an active signal injection unit (1), a multi-channel sensor acquisition unit (2), a baseline feature matching unit (3), a DPP feature optimization unit (4), and a fault location and early warning unit (5). The output of the baseline feature matching unit (3) is connected to the control terminal of the active signal injection unit (1) for sending injection signal parameters. The output of the active signal injection unit (1) is coupled to the power cable under test (7) to inject an active detection signal into the cable; The acquisition end of the multi-channel sensing acquisition unit (2) is coupled to each measurement point of the power cable (7) under test, and the output end is connected to the input end of the DPP feature optimization unit (4) for uploading the acquired waveform and initial features. The output of the DPP feature optimization unit (4) is connected to the input of the fault location and early warning unit (5) and is used to output the redundancy removal feature matrix. The feedback output terminal of the fault location and early warning unit (5) is connected to the control terminals of the active signal injection unit (1) and the multi-channel sensor acquisition unit (2) respectively, and is used to issue parameter adjustment instructions for iterative retesting.

7. The cable fault early warning and location system based on the DPP algorithm active injection according to claim 6, characterized in that, The active signal injection unit (1) includes a cascaded signal generator, a power amplifier, and a coupled injection probe (8); the coupled injection probe (8) adopts a clamp-type electromagnetic coupling structure and is used to be sleeved on the outside of the phase line of the cable; the active signal injection unit (1) has a built-in impedance matching network to achieve dynamic matching between the signal output impedance and the characteristic impedance of the cable.

8. The cable fault early warning and location system based on the DPP algorithm active injection according to claim 6, characterized in that, The multi-channel sensing acquisition unit (2) includes multiple fixed distributed acquisition terminals (9), mobile acquisition terminals (11), and a synchronization clock module (10); the synchronization clock module (10) is connected to each acquisition terminal in a star configuration to provide a unified time reference; each acquisition terminal has a built-in high-frequency current sensor and voltage sensor for acquiring electrical signals; the synchronization clock module (10) supports two modes: fiber optic time synchronization and GPS time synchronization, where fiber optic time synchronization corresponds to the fixed acquisition terminal and GPS time synchronization corresponds to the mobile acquisition terminal (11).

9. The cable fault early warning and location system based on the DPP algorithm active injection according to claim 6, characterized in that, The fault location and early warning unit (5) includes a transmission line simulation module and a degradation trend fitting module connected in series. The transmission line simulation module is used to correct signal attenuation and delay errors, and the degradation trend fitting module is used to output four levels of early warning results: normal, attention, warning, and alarm.

10. The cable fault early warning and location system based on the DPP algorithm active injection according to claim 6, characterized in that, It also includes a cloud interaction unit (6); the cloud interaction unit (6) is bidirectionally connected to the fault location and early warning unit (5) for storing historical detection data, updating the baseline feature library, and pushing location results and early warning information to the operation and maintenance terminal.

Images