A power cable fault diagnosis method and system based on nonlinear feature extraction

Abstract

This application discloses a method and system for fault diagnosis of power cables based on nonlinear feature extraction, belonging to the field of power equipment technology. The method includes: acquiring real-time status data through multi-channel trajectory weaving and constructing an association map after filtering out environmental interference; subsequently quantifying microscopic distortions through feature manifold comparison, identifying status fingerprints using clustering evolution, and adaptively adjusting trigger thresholds accordingly; finally, fusing basic operating data and real-time features, and completing qualitative judgment of the power cable's operating status, defect type identification, and fault location through causal chain analysis, ultimately deriving a complete fault diagnosis result. This scheme effectively eliminates interference from complex operating conditions while significantly improving the sensitivity of microscopic damage capture, thus providing efficient and robust cross-modal causal diagnostic support for early warning, accurate location, and life prediction of insulation defects.

Description

Technical Field

[0001] This application belongs to the field of power equipment technology, specifically relating to a method and system for diagnosing power cable faults based on nonlinear feature extraction. Background Technology

[0002] As the core transmission carrier of urban power grids, the slight changes in the internal insulation state of power cables are often precursor signals of faults. However, such signals are usually buried in complex electromagnetic background noise and exhibit nonlinear characteristics of multi-physical field coupling and multi-time scale interweaving, making them difficult to capture effectively through a single monitoring method.

[0003] Currently, a single sensor is typically used to collect local signals from power cables. Anomalies are then determined by using a fixed threshold, or features are extracted using frequency domain analysis methods such as Fourier transform and compared with preset standards. Some solutions introduce multi-sensor data, perform comprehensive evaluation through weighted fusion, and extrapolate and predict fault trends by combining historical statistical patterns.

[0004] However, existing technologies lack spontaneous clustering and adaptive feature capture mechanisms for signal modal structures, which makes it impossible to dynamically adjust the acquisition trigger threshold according to environmental interference and real-time changes in signal characteristics. Ultimately, this leads to a break in the causal chain between microscopic topological changes and macroscopic evolution trends, resulting in insufficient sensitivity and accuracy in fault diagnosis. Summary of the Invention

[0005] To overcome the above-mentioned deficiencies, this invention is proposed to provide a solution or at least a partial solution to the technical problem that the prior art lacks a spontaneous clustering and adaptive feature capture mechanism for signal modal structures, which makes it impossible to dynamically adjust the acquisition trigger threshold with environmental interference and real-time changes in signal characteristics, ultimately causing the causal chain between microscopic topological changes and macroscopic evolution trends to break, resulting in insufficient sensitivity and accuracy of fault diagnosis.

[0006] In a first aspect, the present invention provides a method for diagnosing power cable faults based on nonlinear feature extraction, the method comprising: Multimodal signal data of power cables are acquired, and multi-channel time delay dynamic mapping and multi-dimensional spatial trajectory weaving are performed based on the multimodal signal data to obtain multi-dimensional state space trajectory data that characterizes the real-time operation of power cables. Based on multidimensional state space trajectory data, blind identification and non-contact spatial weighted filtering of the distribution density of random and disordered environmental interference are performed to obtain a multidimensional state correlation map after filtering out electromagnetic background noise. Based on the multidimensional state correlation map and the preset benchmark state correlation map, dynamic alignment of multidimensional feature manifolds and nonlinear spatial overlap comparison are performed to obtain real-time morphological distortion rate data characterizing the microscopic anomalies of the internal insulation state of power cables. Based on real-time morphological distortion rate data, spontaneous association of affinity and cluster boundary iterative evolution under unlabeled state is carried out to obtain state classification characterization clusters that characterize the manifold distribution state of power cable state signal group. Based on state classification characterization clusters, online evaluation of the severity of feature mutations and adaptive adjustment of feature capture sensitivity are performed to obtain dynamic adjustment command data for correcting the trigger threshold of power cable signal acquisition channels. By acquiring basic operating data of power cables, and based on dynamic adjustment command data and basic operating data, cross-modal causal chain fusion diagnosis of micro-correlation distortion characteristics and macro-evolution trend is performed to obtain fault diagnosis results of power cables.

[0007] In a second aspect, the present invention provides a power cable fault diagnosis system based on nonlinear feature extraction, the system comprising: The multi-dimensional trajectory construction module is used to acquire multi-mode signal data of power cables, and to perform multi-channel time delay dynamic mapping and multi-dimensional spatial trajectory weaving based on the multi-mode signal data to obtain multi-dimensional state space trajectory data that characterizes the real-time operation of power cables. The noise interference filtering module is used to perform blind identification and non-contact spatial weighted filtering of the random and disordered interference distribution density of the environment based on multidimensional state space trajectory data, so as to obtain a multidimensional state correlation map after filtering out electromagnetic background noise. The dynamic distortion comparison module is used to perform dynamic alignment and nonlinear spatial overlap comparison of multidimensional feature manifolds based on multidimensional state correlation maps and preset benchmark state correlation maps, so as to obtain real-time morphological distortion rate data that characterizes the microscopic anomalies of the internal insulation state of power cables. The state clustering evolution module is used to perform spontaneous association of affinity and cluster boundary iterative evolution under unlabeled state based on real-time morphological distortion rate data, so as to obtain state classification characterization clusters that characterize the manifold distribution state of power cable state signal group. The dynamic instruction generation module is used to perform online evaluation of the severity of feature mutations and adaptive adjustment of feature capture sensitivity based on state classification characterization clusters, so as to obtain dynamic adjustment instruction data for correcting the trigger threshold of the power cable signal acquisition channel. The diagnostic module is used to acquire basic operating data of power cables. Based on dynamic adjustment command data and basic operating data, it performs cross-modal causal chain fusion diagnosis of micro-correlation distortion characteristics and macro-evolution trends to obtain fault diagnosis results of power cables.

[0008] In a third aspect, an electronic device is provided, comprising a processor, a memory, and a program or instructions stored in the memory and executable on the processor, the program or instructions being loaded and run by the processor to perform the steps of the aforementioned power cable fault diagnosis method based on nonlinear feature extraction.

[0009] In a fourth aspect, a computer-readable storage medium is provided, wherein a plurality of program codes are stored therein, the program codes being adapted to be loaded and run by a processor to perform the steps of the above-described power cable fault diagnosis method based on nonlinear feature extraction.

[0010] The above-described technical solutions of the present invention have at least one or more of the following beneficial effects: In implementing the technical solution of this invention, intelligent monitoring of power cables is realized throughout the entire process, from multi-dimensional data acquisition, blind noise reduction, manifold evolution to threshold closed-loop adjustment. While effectively eliminating interference from complex operating conditions, it significantly improves the sensitivity of micro-damage capture, thereby providing efficient and robust cross-modal causal diagnostic support for early warning, accurate location and life prediction of insulation defects. Attached Figure Description

[0011] The disclosure of this invention will become more readily understood with reference to the accompanying drawings. It will be readily understood by those skilled in the art that these drawings are for illustrative purposes only and are not intended to limit the scope of protection of this invention. Furthermore, similar numbers in the drawings are used to denote similar components, wherein: Figure 1 This is a schematic flowchart of the first main steps of a power cable fault diagnosis method based on nonlinear feature extraction according to an embodiment of the present invention. Figure 2 This is a schematic diagram of the second main steps of a power cable fault diagnosis method based on nonlinear feature extraction according to an embodiment of the present invention; Figure 3 This is a schematic diagram of the main structure of a power cable fault diagnosis system based on nonlinear feature extraction according to an embodiment of the present invention; Figure 4 This is a schematic diagram of the structure of an electronic device according to an embodiment of the present invention. Detailed Implementation

[0012] Some embodiments of the present invention will now be described with reference to the accompanying drawings. Those skilled in the art should understand that these embodiments are merely illustrative of the technical principles of the present invention and are not intended to limit the scope of protection of the present invention.

[0013] In the description of this invention, "module" and "processor" can include hardware, software, or a combination of both. A module can include hardware circuitry, various suitable sensors, communication ports, memory, and may also include software components, such as program code, or a combination of software and hardware. A processor can be a central processing unit, microprocessor, image processor, digital signal processor, or any other suitable processor. The processor has data and / or signal processing capabilities. The processor can be implemented in software, in hardware, or a combination of both. Non-transitory computer-readable storage media includes any suitable medium capable of storing program code, such as magnetic disks, hard disks, optical disks, flash memory, read-only memory, random access memory, etc. The term "A and / or B" means all possible combinations of A and B, such as only A, only B, or A and B. The terms "at least one A or B" or "at least one of A and B" have a similar meaning to "A and / or B" and can include only A, only B, or A and B. The singular terms "a" or "this" can also include plural forms.

[0014] See appendix Figure 1 , Figure 1 This is a schematic flowchart of the first main steps of a power cable fault diagnosis method based on nonlinear feature extraction according to an embodiment of the present invention. Figure 1 As shown, a power cable fault diagnosis method based on nonlinear feature extraction in an embodiment of the present invention mainly includes the following steps S101-S106.

[0015] Step S101: Acquire multi-mode signal data of the power cable, perform multi-channel time delay dynamic mapping and multi-dimensional spatial trajectory weaving based on the multi-mode signal data, and obtain multi-dimensional state space trajectory data characterizing the real-time operation of the power cable.

[0016] Power cables are industrial devices used to transmit and distribute electrical energy. They are mainly composed of conductors, insulation layers, shielding layers, and protective layers, and their function is to achieve the safe and stable transmission of high-voltage or low-voltage electrical energy.

[0017] Multimodal signal data is a collection of monitoring information acquired from different physical dimensions of power cables and possessing multiple forms of expression. It includes synchronous measurement data from multiple sources and of various types, such as partial discharge current pulses, ultrasonic vibrations, high-frequency electromagnetic radiation, cable core temperature changes, and grounding current.

[0018] Multidimensional state-space trajectory data uses mathematical mapping and data weaving techniques to transform multidimensional monitoring signals into a series of discrete coordinate points in a high-dimensional abstract space. These points evolve over time in space to form continuous lines, which map the dynamic changes and health evolution trends of the cable's internal operating status.

[0019] To acquire multimodal signal data from power cables, multiple sensors must first be installed simultaneously at specific physical locations along the cable. High-frequency current transformers are used to synchronously collect partial discharge current pulses on the grounding wire; ultrasonic sensors are used to synchronously capture ultrasonic vibrations on the cable joint surface; ultrasonic high-frequency antennas are used to synchronously sense high-frequency electromagnetic radiation in the terminal field; distributed fiber optic sensors are laid along the cable core to synchronously measure core temperature changes; and grounding current sensors monitor the grounding current of the metal sheath in real time. These heterogeneous sensors convert the captured physical quantities into electrical signals, which are then processed by a multi-channel synchronous data acquisition system using high-sampling-rate modal synchronization and analog-to-digital conversion. The outputs of all sensors on the same time axis are aligned and digitally encoded, thereby acquiring multimodal signal data including partial discharge current pulses, ultrasonic vibrations, high-frequency electromagnetic radiation, core temperature changes, and grounding current.

[0020] Based on the acquired multimodal signal data, a multi-channel time delay dynamic mapping process is initiated. This process first uses the sampled time series to reconstruct a one-dimensional discrete time series of each channel in the multimodal signal data. Considering the differences in dynamic characteristics between fast-changing signals such as partial discharge current pulses and slow-changing signals such as cable core temperature changes, the optimal time delay step size for each independent channel signal is calculated using the mutual information method. Subsequently, based on the calculated time delay step size, the one-dimensional discrete time series of each channel is subjected to a step-wise translation and lag reassembly on the time axis, transforming the time series of a single channel into a multi-dimensional time delay vector composed of the current time value and several historical time lag values. This mathematical mapping from one-dimensional to high-dimensional using time delay can retain the potential dynamic information of each modal signal deposited on the time axis due to the evolution of the internal physical state of the cable, resulting in time delay mapping state data containing multi-channel characteristics.

[0021] After obtaining the time delay mapping state data, multi-dimensional spatial trajectory weaving continues to integrate the scattered time delay vectors into an organic whole. This process employs phase space reconstruction technology, using the time delay mapping state data representing partial discharge current pulses, ultrasonic vibrations, high-frequency electromagnetic radiation, cable core temperature changes, and grounding current as different independent dimensional coordinate axes. Reconstruction is performed in a unified high-dimensional Euclidean geometric coordinate system, allowing the time delay vectors of each channel to undergo spatial cross-projection and matrix weaving on the corresponding coordinate axes. This integrates the monitoring values ​​of multiple channels at the same discrete moment into a discrete coordinate point with a clear physical coordinate within this high-dimensional abstract space. As the time series data is collected, the system smoothly connects these discrete coordinate points, continuously generated in a high-dimensional abstract space, according to their chronological order, using a cubic spline interpolation algorithm. This forms a three-dimensional geometric line that continuously evolves in space over time. Through stretching, twisting, or clustering of the spatial geometric topology, this line visually transforms the microscopic physical state of the cable's internal insulation layer and other structures into the geometric deformation of a macroscopic spatial trajectory. Ultimately, this yields multidimensional state-space trajectory data that maps the dynamic changes and health evolution trends of the cable's internal operating state. For example, if the cable joint is in an ideal healthy state, its discharge intensity, temperature value, and current fluctuations remain in a constant dynamic equilibrium. The corresponding trajectory line will appear as a stable "ring" or "flat ellipse" in the coordinate system, symbolizing the periodic recurrence of the cable's internal physical state. When the joint begins to absorb moisture or the insulation layer ages, due to intensified partial discharge and abnormal temperature rise, the distribution of the signal vector in the system on each coordinate axis will deviate. At this time, the originally stable "closed loop" trajectory will exhibit stretching or twisting in a specific spatial direction, and even the center point of the trajectory may drift, and the line thickness may expand. This continuous curve, formed by geometric deformation, intuitively presents the evolutionary path from a healthy state to microscopic damage, which is multidimensional state-space trajectory data.

[0022] Based on the above technical solution, optionally, multi-channel time delay dynamic mapping and multi-dimensional spatial trajectory weaving can be performed based on multi-modal signal data to obtain multi-dimensional state-space trajectory data characterizing the real-time operation of power cables, including: Based on the multi-mode signal data, dynamic mapping of multi-channel time delay is performed to obtain time delay mapping state data characterizing the coupling timing state of each phase of the power cable. Based on the time delay mapping state data, trajectory cross weaving is performed in the high-dimensional phase space to obtain a multi-dimensional reconstructed phase space trajectory that characterizes the coordinated changes of power cable pulse interference and normal operating load. Based on the multidimensional reconstructed phase space trajectory, continuous time slice segmentation of the dynamic trajectory is performed to obtain time slice trajectory data characterizing the continuous monitoring time sequence of the power cable. Based on the time slice trajectory data, feature reconstruction of multidimensional feature manifold morphology is performed to obtain feature spatial distribution contour data that characterizes the internal state evolution law of power cable. Based on the feature space distribution contour data, weighted feature extraction of manifold trajectory distribution density is performed to obtain manifold density weight data characterizing the evolution law of power cable operating condition distribution characteristics; Based on the manifold density weight data and the multidimensional reconstructed phase space trajectory, the correlation and fusion calibration of the intrinsic dynamic characteristics are performed to obtain multidimensional state space trajectory data characterizing the real-time operation of the power cable.

[0023] In this scheme, the time delay mapping state data is a set of time-series data obtained by dynamically mapping the multi-channel monitoring signals of the power cable. It is used to quantify the cross-correlation characteristics, electromagnetic wave propagation phase difference, time lag, and coupling relationship between the signals of each channel.

[0024] High-dimensional phase space is a technique that maps univariate monitoring signals to a high-dimensional Euclidean geometric space using phase space reconstruction. It utilizes delayed coordinate vectors to fully reconstruct the system's evolution trajectory, revealing the deterministic dynamic characteristics hidden in complex time series.

[0025] The multidimensional reconstructed phase space trajectory is a complex streamlined structure formed by the cross-weaving and fusion of the reconstructed phase spaces of each channel. It intuitively represents the nonlinear evolution path and cooperative mechanism of pulse interference and conventional load in power cables under complex electromagnetic background.

[0026] Time-slice trajectory data is local state geometric path data extracted by dynamically segmenting the multidimensional reconstructed phase space trajectory on the time axis. It discretizes the long-span continuous evolution process into high-density state segments to achieve segmented quantitative analysis of micro-features.

[0027] Feature spatial distribution contour data is a set of geometric morphological descriptions reconstructed from time slice trajectory data. Through spatial density distribution, curvature change and boundary circumference, it accurately depicts the evolution trend of the physical operating state of power cables in the feature dimension.

[0028] Manifold density weight data is a weighted scalar of the distribution density of each working condition in the feature space, reflecting the contribution ratio and importance of different feature subspaces in the evolution of the entire working condition.

[0029] Based on the multimodal signal data, a dynamic mapping of multi-channel time delay is performed. First, a time delay estimation technique based on cross-correlation function is introduced. This involves performing sliding cross-correlation calculations on synchronous measurement data collected from different physical dimensions of the power cable, including partial discharge current pulses, ultrasonic vibrations, high-frequency electromagnetic radiation, cable core temperature changes, and grounding current, to capture the relative time difference of signal arrival between different phases and channels. Next, a nonlinear deviation compensation technique is introduced. Combining the physical location of each channel sensor, adaptive gain correction is applied to the relative time difference to eliminate weak phase drift, obtaining a time-series data set that quantifies the time lag and coupling relationship between channels, i.e., time delay mapping state data.

[0030] Based on the time-delay mapping state data, trajectory cross-weaving in a high-dimensional phase space is performed. First, a topology topology reconstruction technique based on delay coordinates is introduced to project the time-delay mapping state data into a high-dimensional Euclidean geometric space. The geometric topology of the single-channel signal evolution trajectory is completely reconstructed through the delay coordinate vector, thus constructing a high-dimensional phase space. Next, a multidimensional tensor product nonlinear fusion technique is introduced to perform spatial cross-weaving and streamline alignment of the reconstructed trajectories of each channel. This allows high-frequency pulse interference and low-frequency conventional operating loads to synchronously couple in the same geometric space, mapping the coordinated changes in the state motion path and obtaining a complex streamlined structure characterizing its synchronous dynamic behavior, i.e., the multidimensional reconstructed phase space trajectory.

[0031] Based on the multidimensional reconstructed phase space trajectory, continuous time-slice segmentation of the dynamic trajectory is performed. First, a temporal sliding window segmentation technique is introduced, which calls the multidimensional reconstructed phase space trajectory and performs equidistant geometric cutting along a fixed-width spatial sliding window set along the time axis. Next, a boundary smoothing extraction technique is introduced to perform continuity correction at the beginning and end of the local streamline segments within each slice window, discretizing the long-span continuous evolution process into sequential, spatially independent high-density state segments. While maintaining the integrity of microscopic dynamic features, local state geometric path data within a specific time window, i.e., time-slice trajectory data, is obtained.

[0032] Based on the time-slice trajectory data, a feature reconstruction of the multidimensional feature manifold morphology is performed. First, a high-dimensional feature geometric projection technique is introduced to read the time-slice trajectory data and calculate the rate of change of the normal vector and the local curvature of the surface for each local geometric path in space, extracting its high-dimensional bending and twisting features. Next, a feature manifold boundary circumference extraction technique is introduced to convert the above curvature and normal vector characteristics into a geometric topological boundary representing the trajectory's distribution range, accurately depicting the morphological evolution trend and spatial distribution density of the running state within the feature dimension, and obtaining a set of geometric morphological descriptions representing the state evolution law, i.e., feature spatial distribution contour data.

[0033] Based on the aforementioned feature space distribution contour data, weighted feature extraction of manifold trajectory distribution density is performed. First, Gaussian kernel density estimation is introduced. The feature space distribution contour data is read, and Gaussian weighted superposition is performed centered on each local feature region. The trajectory clustering degree at each coordinate is calculated online, and its proportion of the overall working condition evolution volume is quantified. Next, a normalized weight allocation technique is introduced. The clustering degree is subjected to full-space integral normalization processing to calculate the contribution ratio of each local region to the overall topological change, forming scalar distribution data. This yields manifold density weight data characterizing the evolution law of working condition features.

[0034] Based on the manifold density weight data and the multidimensional reconstructed phase space trajectory, the intrinsic dynamic characteristics are correlated and fused for calibration. First, a dynamic matrix weighted correction technique is introduced. The manifold density weight data and the multidimensional reconstructed phase space trajectory are called, and the spatial coordinate weighting adjustment of the multidimensional reconstructed phase space trajectory is performed using the contribution ratio scalar of each feature subspace, enhancing key dynamic trajectories and suppressing occasional clutter interference. Next, a cross-modal causal chain fusion diagnostic technique for multidimensional manifold space is introduced. The weighted trajectory is reconstructed and its lines are woven in multidimensional space, fusing discrete multi-channel dynamic characteristics into a continuously evolving geometric entity. This achieves the correlation and alignment between local microscopic laws and macroscopic trends, ultimately obtaining multidimensional state-space trajectory data that maps the cable's operating status and health evolution trend.

[0035] This solution enables dynamic monitoring of the entire power cable chain. Through multi-dimensional reconstruction and cross-modal fusion, noise interference is accurately eliminated, significantly improving the accuracy of capturing early microscopic insulation damage and providing efficient and robust support for fault tracing and life prediction.

[0036] Step S102: Based on the multidimensional state space trajectory data, perform blind identification of the random disordered interference distribution density of the environment and non-contact spatial weighted filtering to obtain a multidimensional state association map after filtering out electromagnetic background noise.

[0037] The multidimensional state correlation map is a dynamic distribution manifold constructed in the parameter space by using high-dimensional data weaving and manifold evolution technology to denoise and align multimodal monitoring signals. It is used as a set of high-order mathematical mappings to characterize the operating characteristics of power cables. By quantifying the correlation weights in the feature space to eliminate interference, and based on the parameter linkage evolution logic, it accurately characterizes the degradation mechanism of key cable components.

[0038] Based on multidimensional state-space trajectory data, this method utilizes high-dimensional data weaving and manifold evolution techniques to achieve blind identification and non-contact spatial weighted filtering of random and disordered environmental interference, thereby constructing a multidimensional state correlation map. The process first introduces a non-parametric clustering technique based on neighborhood density, using a continuous line composed of a series of discrete coordinate points in a high-dimensional abstract space as the evaluation subject. Without relying on prior labels, it automatically identifies scattered interference points deviating from the cable's healthy evolution trend by searching the clustering degree of nearest neighbors in the local feature space, and constructs a density reference field describing the spatial distribution of disordered environmental disturbances, thus obtaining interference distribution density data characterizing the interference characteristics of the power cable's on-site operating environment. Subsequently, adaptive boundary tracking technology is used to extract gradient variation features of probability values, accurately delineating the boundary between noise distribution and core operating states, and obtaining interference boundary data.

[0039] By employing the spatial distance inverse attenuation technique and the manifold weighting factor solution method, the closed interface established by the interference boundary data is transformed into a dynamic multidimensional weighted coefficient matrix. This matrix serves as a weight lever to perform spatial lattice weighted mapping on the original trajectory data. While preserving core state characteristics, it non-contactly suppresses the signal amplitude of environmental stray interference, causing continuous lines to spontaneously and smoothly shift towards high-trust weight regions, thus obtaining a weighted attenuation feature sequence. The sequence is then compared for temporal evolution coherence using dynamic sliding window anomaly detection technology and isolated point directional stripping methods to completely eliminate residual microscopic abrupt noise, obtaining a pure state evolution sequence.

[0040] By employing the simple complex construction technique in algebraic topology and the manifold alignment method in multidimensional feature space, the discrete coordinate points in the pure state evolution sequence are solidified and their spatial causal linkage logic is calibrated for consistency. This process reconstructs the trajectory lines that originally flowed along the time axis into a static high-order set distributed in the parameter space. By quantifying the correlation weights in the feature space, the nonlinear interference of electromagnetic background noise is completely removed. Based on the linkage evolution logic between parameters, the degradation mechanism and distribution law of key components such as the internal insulation layer of power cables under complex operating conditions are accurately mapped, ultimately deriving a multidimensional state correlation map that can characterize the real-time operating characteristics of power cables.

[0041] Based on the above technical solution, optionally, blind identification and non-contact spatial weighted filtering of the random and disordered environmental interference distribution density are performed based on multidimensional state space trajectory data to obtain a multidimensional state correlation map after filtering out electromagnetic background noise, including: Based on the multidimensional state space trajectory data, the spatial disordered interference distribution density is automatically identified to obtain interference distribution density data that characterizes the interference characteristics of the power cable field operation environment. Based on the interference distribution density data, the blind zone boundary of the environmental disordered random noise is delineated to obtain interference boundary data characterizing the coverage range of environmental noise for power cable monitoring. Based on the interference boundary data, a non-contact multidimensional weighting matrix is ​​calculated online in real time to obtain a multidimensional weighting coefficient matrix for suppressing interference in the power cable acquisition channel. Based on the multidimensional weighted coefficient matrix and multidimensional state space trajectory data, multidimensional feature space weighted mapping and noise amplitude suppression are performed to obtain a weighted attenuation feature sequence after achieving stray interference attenuation of power cables. Based on the weighted attenuation feature sequence, the abnormal mutation state features are selectively stripped and eliminated to obtain a pure state evolution sequence that characterizes the operating condition of the power cable. Based on the pure state evolution sequence, a parameter correlation network is constructed and its consistency is calibrated to obtain a multidimensional state correlation map after filtering out electromagnetic background noise.

[0042] In this scheme, the interference distribution density data is the spatiotemporal distribution data that quantitatively characterizes the degree of spatial aggregation of non-qualitative interference signals such as external electromagnetic radiation and environmental electric field fluctuations in the field operation environment of power cables.

[0043] Interference boundary data defines the extent to which random environmental noise can affect the system in both the spatial and temporal domains. By delineating the blind zone boundary of environmental noise, it is used to define the critical region affected by environmental electromagnetic interference in the power cable monitoring system.

[0044] The multidimensional weighted coefficient matrix is ​​a set of mathematical operators used for interference suppression in multi-channel acquisition systems for power cables. By dynamically adjusting the amplitude and phase weights of different signal channels, it can accurately cancel out complex stray electromagnetic interference around the cable.

[0045] The weighted attenuation feature sequence is a signal sequence that removes environmental background interference and random noise while retaining the evolution trend of core features, so as to reveal the weak insulation defect state characteristics after background stray interference attenuation.

[0046] The pure state evolution sequence is a pure data sequence that removes abnormal outliers and state jumps caused by sudden external electromagnetic pulses or instantaneous sensor jitter, and contains only the true physical conditions and state evolution laws of power cables.

[0047] Automatic identification of spatial disordered interference distribution density based on the multidimensional state-space trajectory data is achieved. First, high-dimensional local density clustering technology is introduced to read the multidimensional state-space trajectory data, which evolves over time into continuous lines within an abstract space. The clustering density of discrete coordinate points in various spatial regions is calculated online, capturing external electromagnetic radiation and environmental electric field fluctuations that deviate from the core evolution trajectory. Next, spatiotemporal grid mapping technology is introduced to project the identified non-qualitative interference sources onto a unified spatiotemporal grid coordinate system for frequency statistics. This yields spatiotemporal distribution data, i.e., interference distribution density data, which quantitatively characterizes the degree of spatial clustering of interference signals under the power cable field operating environment.

[0048] Based on the interference distribution density data, the blind zone boundary of environmental disordered random noise is delineated. First, spatial gradient operator edge detection technology is introduced to read the interference distribution density data and calculate its density change gradient at the intersection of each spatiotemporal grid, accurately locating the geometric abrupt changes where the interference density decreases sharply. Next, convex hull envelope topology partitioning technology is introduced, using these geometric abrupt changes as topological vertices to continuously delineate the boundary, forming a closed geometric boundary in both the spatial and temporal domains to define the blind zone boundary of environmental noise. This effectively defines the critical region affected by environmental electromagnetic interference in the power cable monitoring system, thereby obtaining interference boundary data characterizing the coverage area of ​​environmental noise in power cable monitoring.

[0049] Based on the interference boundary data, a non-contact multidimensional weighted matrix is ​​calculated online in real time. First, a spatial distance-inverse sensor sensitivity modeling technique is introduced. The interference boundary data is read, and combined with the physical location of each channel sensor, the relative geometric distance from each acquisition channel to the aforementioned environmental noise blind zone boundary is calculated in real time. Next, an adaptive amplitude-phase inverse cancellation transformation technique is introduced. Based on the calculated relative geometric distance, a set of complex attenuation factors is calculated online for each acquisition dimension to match the spatial noise level. The amplitude and phase weights of different signal channels are dynamically adjusted, and these mathematical operators are recombined into a multidimensional matrix to obtain a multidimensional weighted coefficient matrix used to suppress interference in the power cable acquisition channels.

[0050] Based on the multidimensional weighted coefficient matrix and multidimensional state space trajectory data, multidimensional feature space weighted mapping and noise amplitude suppression are performed. First, a high-dimensional matrix projection dot product technique is introduced. The multidimensional weighted coefficient matrix and the multidimensional state space trajectory data are called, and each discrete coordinate point in the trajectory is mapped by a dot product with its corresponding weighted coefficient, thus directionally reducing the amplitude of the disturbed dimension. Next, an inverse electromagnetic field cancellation synthesis technique is introduced. The phase weights in the matrix are used to perform real-time phase offsetting on the mapped trajectory, directly achieving precise cancellation and suppression of complex stray electromagnetic interference in the cable's surrounding environment at the amplitude level. This eliminates mixed environmental background interference and random noise, revealing weak insulation defect state characteristics, and ultimately obtaining a weighted attenuation feature sequence after stray interference attenuation in the power cable.

[0051] Based on the weighted attenuation feature sequence, anomalous abrupt state characteristics are selectively stripped and eliminated. First, high-frequency time-varying wavelet singularity detection technology is introduced to read the weighted attenuation feature sequence point by point. By analyzing the instantaneous curvature and first-order differential rate of change of the sequence waveform on the time axis, anomalous outliers and state jumps caused by sudden external electromagnetic pulses or instantaneous sensor jitter are accurately identified. Next, nonlinear spline trajectory smoothing elimination technology is introduced to selectively remove the identified anomalous outliers and state jumps from the sequence. Interpolation and smoothing fitting are then performed using adjacent normal operating points to ensure that the sequence only contains the actual physical operating conditions and state evolution laws of the power cable, thus obtaining a pure state evolution sequence characterizing the operating state characteristics of the power cable.

[0052] Based on the pure state evolution sequence, a parameter association network is constructed and its consistency is calibrated. First, a manifold topology reconstruction technique is introduced. The pure state evolution sequence is read, and the physical dimension data, after removing anomalous mutations, are used as topology nodes for multidimensional weaving. A dynamic state distribution manifold reflecting the system's linkage evolution logic is constructed in the parameter space. Next, a consistency closed-loop alignment calibration technique is introduced. The constructed dynamic state distribution manifold is compared with the standard cable degradation mechanism law, and the global manifold curvature is corrected and aligned with physical constraints. This completely eliminates the interference of residual noise on core operating characteristics, accurately characterizes the distribution law and degradation mechanism of key components such as the internal insulation layer of the cable under different operating loads and environmental parameters, and finally obtains a multidimensional state association map as a high-order mathematical mapping set characterizing the operating characteristics of power cables.

[0053] In this solution, through multi-level spatial mapping and dynamic calibration, environmental noise and defect characteristics are accurately separated, complex interference is effectively eliminated, the noise immunity of power cable operation status monitoring is significantly improved, and the accuracy of insulation defect identification and the long-term stability of diagnostic results are ensured.

[0054] Step S103: Based on the multidimensional state correlation map and the preset benchmark state correlation map, perform dynamic alignment of multidimensional feature manifolds and nonlinear spatial overlap comparison to obtain real-time morphological distortion rate data characterizing the microscopic anomalies of the internal insulation state of the power cable.

[0055] The preset baseline state association map refers to the baseline distribution manifold formed after the power cable has passed factory acceptance or is in a standardized operating condition, based on the high-dimensional data features extracted from it. This manifold is then processed by manifold space standardization and serves as a zero reference baseline for judging whether there are any abnormalities in the current operating state of the power cable.

[0056] Real-time morphological distortion rate data is a quantitative indicator obtained by comparing the high-dimensional manifold alignment and overlap between the currently running correlation map and the benchmark correlation map. It is used to characterize the deviation of the parametric network manifold structure, depict the difference in geometric distortion and distribution density of the current cable data relative to the healthy benchmark, and reveal the cumulative evolution rate of microscopic deterioration or structural damage of the insulation layer under complex stress.

[0057] Based on a multidimensional state correlation map and a preset benchmark state correlation map, dynamic calibration and deviation quantification of the characteristic manifold are achieved through high-dimensional manifold alignment and nonlinear overlap comparison techniques, thereby deriving real-time morphological distortion rate data. This process first introduces principal component feature extraction technology for multidimensional characteristic manifolds. Simultaneously, the current multidimensional state correlation map is used as a set of high-order mathematical mappings characterizing the operating characteristics of the power cable. This, along with the benchmark distribution manifold fixed by the preset benchmark state correlation map, serves as the input source. Without relying on the spatial geometric structure of entities, the center positions of the two sets of high-dimensional maps in the feature space are calculated and translated to coincide, completing a rough registration. This yields the initial registration correlation features characterizing the initial operating condition alignment state of the power cable. By introducing multidimensional manifold dynamic nonlinear affine transformation technology and nearest neighbor adaptive matching algorithm, such as nonrigid iterative nearest neighbor algorithm, and using a preset baseline state correlation map as a rigid reference, the dynamic state distribution manifold generated in real time is driven to undergo nonlinear morphological adjustment, eliminating the pseudo-distortion trend caused by environmental and load fluctuations, so that the current operating data and standardized operating conditions are highly consistent, and dynamic alignment network data that characterizes the spatiotemporal consistency features of the multidimensional manifold of power cable is obtained.

[0058] This study employs high-dimensional manifold space orthogonal matrix projection technology to deeply analyze dynamically aligned network data. The nonlinear overlap features between the manifold morphologies corresponding to two sets of maps are mapped onto a projection matrix composed of eigenvector groups through orthogonal transformation. This transforms the complex manifold overlap states and spatial distribution density evolution differences into a directly computable digital array, thus obtaining overlap degree projection feature data. Subsequently, a parameter space differential geometric deviation calculation method is introduced to compare the differences between corresponding nodes in the overlap degree projection feature data one by one. This captures the geometric distortion and local warping of the parameter network relative to the healthy baseline in the feature space, locating the anomaly features of the inherent links deviating from the zero reference baseline, and obtaining associated local deviation data.

[0059] By utilizing high-dimensional characteristic manifold density integral quantization, nonlinear distortion regions in correlated local deviation data caused by microscopic insulation degradation or partial discharge precursors are cumulatively superimposed, transforming them into a scalar index reflecting the degree of microscopic state evolution, thus obtaining correlated distortion integral data. Subsequently, a continuous time series sliding window mapping technique is introduced to integrate the correlated distortion integral data with the dynamic monitoring time axis, analyzing the dynamic perturbation ratio of the correlated distortion integral data over time online. This accurately characterizes the cumulative evolution rate of structural damage to the insulation layer under dynamic electric and thermal pressure, ultimately obtaining real-time morphological distortion rate data to characterize the degree of deviation of the parametric network manifold structure and measure the health of the internal insulating medium.

[0060] Based on the above technical solution, optionally, dynamic alignment and nonlinear spatial overlap comparison of multidimensional feature manifolds are performed based on multidimensional state correlation maps and preset benchmark state correlation maps to obtain real-time morphological distortion rate data characterizing the microscopic anomalies of the internal insulation state of power cables, including: Based on the multidimensional state association map and the preset benchmark state association map, multidimensional feature space feature extraction and initial registration of association centroid are performed to obtain the initial registration association features characterizing the initial working condition alignment state of the power cable. Based on the initial registration association features, dynamic affine transformation and manifold matching alignment of the parameter association network are performed to obtain dynamic alignment network data that characterizes the spatiotemporal consistency features of the multidimensional manifold of the power cable. Based on the dynamic alignment network data, feature extraction is performed on the nonlinear coincident manifold direction matrix projection to obtain coincidence projection feature data characterizing the spatial mapping characteristics of power cable parameters. Based on the overlap projection feature data, the correlation difference solution of the local deviation variation trend is performed to obtain the correlation local deviation data characterizing the spatial distribution of the power cable parameter network. Based on the associated local deviation data, the associated evolution density integral quantization of the nonlinear distortion region is performed to obtain associated distortion integral data characterizing the degree of microstate evolution of the power cable. Based on the associated distortion integral data, a continuous time-series mapping of the dynamic disturbance ratio is performed to obtain real-time morphological distortion rate data characterizing the microscopic anomalies in the internal insulation state of the power cable.

[0061] In this scheme, the initial registration association feature is the data formed after the current operating condition map of the power cable is aligned with the preset health benchmark. This data is used to eliminate environmental acquisition differences and characterize the degree of benchmark alignment of the operating status.

[0062] Dynamically aligned network data is a network model data that, after affine transformation and manifold matching, unifies multidimensional state data under different operating conditions into the same spatiotemporal coordinate system, and is used to characterize the consistency of the operating characteristics of power cables.

[0063] Overlap projection feature data is obtained by mapping a multidimensional network to a feature matrix space. It is used to quantitatively characterize the degree of overlap between the current actual operating state of the cable and the standard health evolution trajectory.

[0064] The associated local deviation data is a spatial variation index calculated based on the overlap characteristics. It is used to locate local abnormal fluctuation areas in the internal parameter network of power cables that deviate from the overall healthy distribution pattern.

[0065] The associated distortion integral data is a quantitative value obtained by performing spatial evolution density integration on the microscopic anomaly deviation area of ​​the power cable, reflecting the cumulative degree of insulation state deviation from the healthy trajectory.

[0066] During multidimensional feature space feature extraction and initial registration of associated centroids, the dynamic state distribution manifold in the multidimensional state association map and the baseline distribution manifold in the preset baseline state association map are read. The higher-order mathematical mapping sets of both are extracted, and the geometric centroid coordinates of the current dynamic state distribution manifold and the baseline distribution manifold are calculated. Principal component analysis is then introduced to calculate the principal axis directions and association weights in their respective feature spaces online to lock rotation and translation biases. Next, a rigid transformation matrix calculation technique is introduced, using the geometric centroid of the baseline distribution manifold as a zero reference datum. The geometric centroid and principal axis directions of the dynamic state distribution manifold are translated and aligned to this zero reference datum, eliminating environmental acquisition differences and background noise interference, and generating initial registration association features to characterize the degree of alignment of the operating state benchmark.

[0067] When performing dynamic affine transformation and manifold matching alignment of the parameter association network based on the initial registration association features, the generated initial registration association features are directly used as the basis for manifold stretching. A high-dimensional multi-point affine transformation technique is introduced, and based on the local deformation parameters in the initial registration association features, shearing, scaling, and translation transformation matrices are calculated online to perform nonlinear spatial dynamic morphological stretching on the parameter association network. Then, an iterative nearest-point topology alignment technique is introduced to perform iterative calculations to minimize the nearest-neighbor distance between the stretched dynamic state distribution manifold and the reference distribution manifold. This unifies the multidimensional state data under different operating conditions into the same spatiotemporal coordinate system, eliminating parameter trajectory misalignment caused by operating load and external environmental disturbances, and obtaining dynamically aligned network data characterizing the consistency of power cable operating characteristics.

[0068] When performing feature extraction for nonlinear overlapping manifold projection based on the aforementioned dynamic alignment network data, the generated dynamic alignment network data is read, and singular value decomposition (SVD) dimensionality reduction technology is introduced. This involves orthogonally decomposing the high-dimensional dynamic alignment network data, unified in the same spatiotemporal coordinate system, to extract the skeleton components reflecting the core evolution path of the manifold topology. Subsequently, kernel matrix projection technology is introduced, projecting these skeleton components onto a preset low-dimensional feature matrix space through nonlinear mapping. The overlapping area and boundary overlap ratio of the current actual operating state and the standard healthy evolution trajectory on the low-dimensional projection surface are calculated, quantitatively characterizing the degree of overlap between the current actual cable operating state and the standard healthy evolution trajectory, and outputting the corresponding overlap projection feature data.

[0069] When calculating the correlation difference of local deviation variation trends based on the aforementioned coincidence projection feature data, the generated coincidence projection feature data is read point by point. A high-order spatial geometric distance deviation calculation technique is introduced, performing point-by-point subtraction difference operations between the feature values ​​of each sampling point in the coincidence projection feature data and the standard coincidence values ​​at the corresponding coordinates of the baseline distribution manifold to extract spatial variation indicators. Next, grid gradient divergence analysis technology is introduced to calculate the local gradient change rate of these differences in the spatial distribution of the parameter network, capturing abnormal divergence trends that deviate from the overall healthy distribution pattern, accurately locating nonlinear offset positions, and generating correlated local deviation data for locating local abnormal fluctuation regions.

[0070] When performing correlation evolution density integral quantization of nonlinear distortion regions based on the aforementioned correlated local deviation data, the generated correlated local deviation data is used as the integration kernel source. A Gaussian neighborhood density aggregation technique is introduced, using each locked local anomalous fluctuation region in the correlated local deviation data as the center, to calculate the neighborhood anomalous evolution density of each variation point in the parameter space. Then, a spatial geometric numerical integration technique is introduced, performing multidimensional spatial integration accumulation along the manifold surface boundaries of these nonlinear distortion regions. This transforms the scattered local fluctuation regions into quantitative values ​​that intuitively reflect the cumulative degree of insulation state deviation from the healthy trajectory, resulting in correlated distortion integral data characterizing the degree of microstate evolution.

[0071] When performing continuous time-series mapping of the dynamic perturbation ratio based on the aforementioned correlated distortion integral data, the generated correlated distortion integral data is arranged as a continuous time series according to the acquisition timestamp. A time-varying Markov dynamic transition probability analysis technique is introduced to calculate the rate of change of the quantified value within adjacent time windows, and to solve for the evolutionary expansion ratio of microscopic anomalies on the time axis. Subsequently, a continuous time-series state mapping technique is introduced to correlate and normalize this rate of change with the microscopic insulation degradation rate of the cable under dynamic electric and thermal pressure. The geometric distortion and distribution density evolution differences of the parametric network manifold structure deviating from the zero reference are quantitatively calculated, and real-time morphological distortion rate data characterizing the microscopic anomalies and cumulative evolution rate of internal insulation state degradation damage are output.

[0072] In this scheme, by constructing a high-dimensional dynamic state distribution manifold, electromagnetic background noise interference is effectively eliminated. Furthermore, by utilizing manifold alignment and density integration techniques, the accumulation rate of insulation damage can be predicted in real time, thereby achieving accurate quantification of the degree of degradation of the cable insulation layer.

[0073] Step S104: Based on real-time morphological distortion rate data, perform spontaneous association of affinity and cluster boundary iteration in the unlabeled state to obtain state classification characterization clusters that represent the manifold distribution state of the power cable state signal group.

[0074] The state classification characterization cluster is a discrete mathematical set automatically formed by manifold clustering and boundary iteration techniques based on the nonlinear evolution characteristics of real-time morphological distortion rate data. It is used to map continuous morphological trajectories into logical classifications with clear boundaries in the parameter space, thereby realizing the automatic identification of cable health level, fault precursors and environmental impact categories. It serves as a high-dimensional distributed fingerprint of various potential state modes of power cables under complex operating conditions.

[0075] Based on real-time morphological distortion rate data, an adaptive logical classification of cable state signal groups is achieved through unlabeled manifold spontaneous association and cluster boundary iterative evolution techniques, thereby deriving state classification characterization clusters. This process first introduces a manifold space affinity measurement technique based on high-dimensional feature Euclidean distance and cosine similarity. Using continuously generated real-time morphological distortion rate data from online monitoring as input, and under unlabeled monitoring conditions requiring no manual labels or prior knowledge, the geometric distance and orientation angle of morphological distortion rate data points from each historical period and the currently collected data are calculated in the high-dimensional parameter space. This quantifies the nonlinear evolutionary similarity between various cable state signals online, thereby obtaining distortion affinity correlation matrix data representing the initial topological proximity relationship between signals. Next, a density peak spontaneous search technique is introduced, which directly calls the distortion affinity correlation matrix data to retrieve the core feature nodes with local maximum distribution density in the parameter space. These core nodes are then established as the topological centers of their respective potential state modes. The similar micro-variation patterns are automatically correlated with the geometric distribution topology in a centripetal manner, thereby obtaining spontaneously agglomerated kernel set data that characterizes the core distribution trend of the cable's operating state.

[0076] This study employs a manifold topological boundary iterative evolution technique to deeply analyze spontaneously agglomerated kernel data and distorted affinity matrix data. Starting from each topological center, it adaptively expands outwards to the sparser regions with an adaptive step size. Using nonparametric gradient ascent algorithms, such as the mean-shift algorithm, it adjusts the edge interfaces of each state clustering region in real time. This eliminates boundary ambiguity caused by uneven evolution rates of partial discharge precursors or structural damage accumulation under dynamic electric and thermal pressures, accurately defining the dynamic watershed between noise signals and health level classifications. This yields iterative topological data representing clear morphological boundaries. Subsequently, a high-dimensional manifold space overlap interference stripping method is introduced. The overlapping boundary regions of each interface in the iterative topological data are compared one by one. By performing a secondary optimization of the topological assignment probabilities of fuzzy cross-feature points, the feature interleaving interference caused by environmental fluctuations is completely removed, resulting in pure association classification boundary data representing completely independent and isolated state patterns.

[0077] By utilizing the discrete mathematical set convergence and solidification technique, the independent topological regions determined by the pure correlation classification boundary data are transformed into spatially closed logical classification entities. The continuous manifold morphological trajectories flowing in the parameter space are spontaneously clustered and locked, realizing a fully automatic fingerprint mapping from the original state data to the cable health level, fault precursor morphology, or operating environment influence category. Finally, a state classification characterization cluster is derived to characterize the manifold distribution state of the power cable state signal group.

[0078] Step S105: Based on the state classification characterization cluster, perform online evaluation of the severity of feature mutations and adaptive adjustment of feature capture sensitivity to obtain dynamic adjustment instruction data for correcting the trigger threshold of the power cable signal acquisition channel.

[0079] The dynamic adjustment command data is based on a real-time state classification characterization cluster. It is a set of configuration parameters generated by online evaluation of the degree of change in characteristic signals and adaptive sensitivity. This set is used to correct the trigger threshold of the signal acquisition channel in real time. By mapping the micro-variation risk of the operating state with the acquisition sensitivity in a closed loop, the system can dynamically scale the capture threshold in different state modes, effectively filtering out environmental interference while ensuring the capture of key mutation features.

[0080] Based on state classification characterization clusters, a closed-loop optimization of data acquisition accuracy is achieved through online feature mutation assessment and channel threshold adaptive feedback adjustment technology, thereby deriving dynamic adjustment command data. This process first introduces a manifold evolution rate calculation technique based on the spatiotemporal derivative matrix. Using the real-time generated state classification characterization clusters as input sources, the instantaneous tangent slope and curvature variation characteristics of the continuous manifold trajectory of the state signal group in the parameter space are directly extracted. The spatial distance jump frequency of the state classification characterization clusters on the time axis is calculated online, thereby quantifying the transition severity of health level transitions, fault precursor morphologies, or operating environment influence categories implicit in the original state data. This yields feature mutation rate index data characterizing the severity of feature signal mutations. Next, a nonlinear sensitivity bias mapping method is introduced, directly calling this feature mutation rate index data and combining it with the current potential state mode for hazard level weighting. While excluding normal operating load fluctuation interference, the channel sensitivity adjustment gain caused by microscopic anomaly risks is calculated, thus obtaining the capture sensitivity gain matrix data characterizing the cable feature capture sensitivity requirements.

[0081] A closed-loop feedback control tuning algorithm is invoked to deeply analyze the acquisition sensitivity gain matrix data, and the current sensitivity requirement is matched with the inherent threshold characteristics of the underlying hardware acquisition channel. The differential prediction term in the algorithm is used to quantize the characteristics in real time to capture the deviation amplitude of the sensitivity from the safe redundancy range. Combined with the integral accumulation term, the steady-state error caused by environmental background interference is dynamically offset, ultimately generating a set of closed-loop control correction parameters that can accurately offset this deviation amplitude. This yields channel trigger reference correction data characterizing the physical trigger reference bias of the acquisition channel. Subsequently, signal transmission bus instruction format encoding technology is introduced to read each bias parameter in the channel trigger reference correction data one by one. Following the standard frame structure of the underlying control bus of the power cable monitoring system, the physical trigger reference bias is encapsulated and converted into a hexadecimal standard configuration message containing the channel number, trigger threshold modification amplitude, and checksum, resulting in threshold configuration control message data characterizing the physical link of the threshold configuration parameters.

[0082] By utilizing continuous flow control feedback execution technology, threshold configuration control message data is injected as control pulses into the storage register of the cable signal acquisition channel in real time. This enables the acquisition channel to dynamically scale the capture threshold in real time according to different potential state modes such as normal operation, insulation degradation, or fault precursors. This ensures that the monitoring system maintains high sensitivity to key abrupt changes and effectively filters out environmental background interference. Finally, dynamic adjustment command data is exported to correct the trigger threshold of the signal acquisition channel in real time.

[0083] Based on the above technical solution, optionally, based on the state classification characterization cluster, online evaluation of the severity of feature mutations and adaptive adjustment of feature capture sensitivity are performed to obtain dynamic adjustment command data for correcting the trigger threshold of the power cable signal acquisition channel, including: Based on the state classification characterization cluster, online calculation of the centroid drift evolution of the feature space manifold is performed to obtain feature mutation data characterizing the degree of characteristic mode shift of the power cable. Based on the aforementioned characteristic mutation data, a quantitative assessment of the diffusion rate of the multidimensional manifold of the time-series evolution trajectory is performed to obtain mutation assessment data characterizing the characteristic mutation trend of power cables. Based on the mutation assessment data, adaptive matching of the channel trigger threshold parameter control boundary is performed to obtain threshold matching data characterizing the initial scale of the power cable channel trigger threshold. Based on the threshold matching data and feature mutation data, closed-loop correction of the channel trigger threshold adjustment sensitivity is performed to obtain sensitivity correction data characterizing the dynamic compensation state of the characteristic amplitude of the power cable. Based on the sensitivity correction data, a sliding window iterative optimization calculation of the dynamic correction amount of the channel trigger threshold is performed to obtain the threshold distribution adjustment vector characterizing the distribution of the multi-channel trigger threshold array of the power cable. Based on the threshold distribution adjustment vector and the state classification characterization cluster, the dynamic fine-tuning parameters of the multi-channel trigger threshold are mapped and fused to obtain dynamic adjustment command data for correcting the trigger threshold of the power cable signal acquisition channel.

[0084] In this scheme, the characteristic mutation data is an index obtained by online calculation of the degree of manifold centroid drift in the characteristic space based on the state classification characterization cluster. It is used to quantitatively characterize the significant shift and abnormal mutation intensity of the power cable operating mode in the parameter space.

[0085] The mutation assessment data is a conclusion obtained by quantitatively assessing the multidimensional manifold diffusion rate in the time-series evolution trajectory based on characteristic mutation data. It is used to characterize the trend and potential rate of the evolution of the current operating state of power cables into the fault state.

[0086] Threshold matching data is a parameter obtained by adaptively matching the control boundary of the trigger threshold of the monitoring channel based on the mutation assessment data. It is used to determine the initial scale of the trigger threshold of each signal channel of the power cable under the current operating conditions.

[0087] Sensitivity correction data is a dynamic compensation parameter calculated by combining threshold matching and feature mutation data through closed-loop correction. It is used to finely adjust the sensing sensitivity and dynamic adaptability of the power cable monitoring system to the feature amplitude.

[0088] The threshold distribution adjustment vector is a set of dynamic correction quantities obtained through sliding window iterative optimization, used to comprehensively characterize the optimal spatial distribution and adjustment of the multi-channel trigger threshold array of power cables in the parameter space.

[0089] When performing online calculations of the centroid drift evolution of the feature space manifold based on the state classification representation cluster, the state classification representation cluster, which serves as the high-dimensional distribution fingerprint of the power cable, is first read in real time, and the newly integrated discrete mathematical set within the current time window is extracted. Then, instantaneous centroid tracking technology is introduced to calculate the instantaneous geometric centroid of each group in the current state classification representation cluster in the high-dimensional space online. Next, Euclidean distance sliding calculation technology is introduced to real-time align the instantaneous geometric centroid with the reference manifold centroid under historical standard operating conditions, and calculate the geometric displacement vector between the two spatial centroids. Finally, matrix norm extraction technology is introduced to calculate the second norm of this vector, quantitatively characterizing the significant shifts and anomalous mutation in the power cable operating modes within the parameter space, and outputting characteristic mutation data.

[0090] When quantitatively evaluating the multidimensional manifold diffusion rate of the time-series evolution trajectory based on the aforementioned characteristic mutation data, the characteristic mutation data is retrieved and arranged continuously according to the acquisition time series. Then, a spatiotemporal evolution trajectory tangent rate calculation technique is introduced to perform a first-order numerical differentiation on the time axis of the continuous characteristic mutation data, calculating the instantaneous movement velocity of the manifold centroid deviation in the characteristic space. Next, a multidimensional manifold diffusion coefficient estimation technique is introduced to calculate the phase space divergence range and spatial diffusion velocity of the characteristic mutation data deviating from the standard healthy boundary within a sliding time window, assessing the potential rate of evolution from the current characteristic mode shift to the fault state, and directly generating mutation evaluation data to characterize the trend and potential rate of evolution from the current operating state of the power cable to the fault state.

[0091] When adaptively matching the control boundary of the channel trigger threshold parameter based on the mutation assessment data, the mutation assessment data is read, and a multi-objective boundary adaptive matching technique is introduced, using the evolution trend and potential rate in the mutation assessment data as input variables. Subsequently, a mapping control technique based on boundary extreme value constraints is introduced to calculate the dynamic safety control boundary of the corresponding acquisition channel under normal operation, insulation degradation, or fault precursor conditions according to the magnitude of the evolution trend. Then, a Lagrange multiplier optimization algorithm is introduced to calculate the amplitude adjustment range and safety threshold boundary of each channel trigger threshold under the current operating conditions, while satisfying the false alarm rate constraint, thus obtaining threshold matching data.

[0092] When performing closed-loop correction of channel trigger threshold adjustment sensitivity based on the threshold matching data and feature mutation data, both the threshold matching data and the feature mutation data are simultaneously invoked. Then, a proportional-integral-derivative feedback closed-loop correction technique is introduced, using the intensity of abnormal mutations in the feature mutation data as the input perturbation and the initial scale of the trigger threshold in the threshold matching data as the reference. Next, a dynamic error compensation calculation technique is introduced, calculating the dynamic gain compensation coefficient online based on the instantaneous change in mutation intensity, thereby achieving precise gain control of the multi-channel acquisition system's response speed and obtaining sensitivity correction data.

[0093] When performing sliding window iterative optimization calculation of the dynamic correction amount of the channel trigger threshold based on the sensitivity correction data, the sensitivity correction data is read, and a time sliding window iterative update technique is introduced to establish an optimization operator for the channel threshold configuration within a preset sampling time window. Subsequently, a gradient descent optimization technique is introduced, using the dynamic gain compensation coefficient in the sensitivity correction data as a constraint, to iteratively search for the dynamic correction amount of the trigger threshold for each channel within the sliding window. Next, an array topology matrix optimization technique is introduced to spatially arrange and align the optimal dynamic correction amount calculated for each channel according to the physical topology of the hardware channels, generating a threshold distribution adjustment vector.

[0094] When mapping and fusing the multi-channel trigger threshold dynamic fine-tuning parameters based on the threshold distribution adjustment vector and the state classification representation cluster, both the threshold distribution adjustment vector and the state classification representation cluster are invoked simultaneously. Subsequently, a high-dimensional space mapping fusion technique is introduced to link and couple the multi-channel threshold adjustment amounts in the threshold distribution adjustment vector with the discrete mathematical set in the state classification representation cluster, thus locking the acquisition threshold of the acquisition channel with the current operating state of the cable in a closed loop. Next, a configuration parameter packaging technique is introduced to transform the mapped and fused fine-tuning parameters into a standard feedback control logic configuration sequence, outputting dynamic adjustment command data for real-time correction of the signal acquisition channel trigger threshold.

[0095] This solution establishes a closed-loop feedback mechanism encompassing state perception, trend prediction, and adaptive adjustment. Upon detecting insulation state deviations, it automatically and accurately corrects the acquisition threshold. This significantly improves the efficiency of detecting early signs of cable insulation damage and the accuracy of fault warnings while suppressing environmental interference.

[0096] Step S106: Obtain basic operating data of the power cable, and based on the dynamic adjustment command data and basic operating data, perform cross-modal causal chain fusion diagnosis of micro-correlation distortion characteristics and macro-evolution trend to obtain the fault diagnosis results of the power cable.

[0097] Basic operating condition data is a set of original multi-dimensional environmental and electrical parameters collected by sensors after the power cable is connected to the power system and after noise reduction preprocessing. It covers multi-modal monitoring values ​​such as load current, surface temperature, operating voltage, ambient humidity, vibration frequency and partial discharge pulse during the standardized operation of the cable. It serves as a benchmark measured base for describing the physical state of the cable.

[0098] The fault diagnosis result is a conclusion drawn from a comprehensive assessment of the health status of power cables based on cross-modal causal chain fusion diagnostic technology, which correlates real-time morphological distortion rate data, state classification characterization clusters, and corrected feature information. The result first determines the cable's operating status. If it is determined to be abnormal, it further characterizes specific defect types within the insulation layer, such as dendritic discharge, electrochemical corrosion, or localized breakdown, and outputs the corresponding fault spatial location information. If it is determined to be normal, the fault diagnosis result is "normal with no abnormalities."

[0099] To obtain basic operating data as a benchmark for describing the physical state of the cable, a hardware-synchronized, time-series multi-dimensional data acquisition technology is first introduced. After the power cable is connected to the power system, high-precision current transformers, fiber optic distributed temperature sensors, capacitive voltage dividers, capacitive micro-water sensors, piezoelectric vibration accelerometers, and high-frequency current sensors deployed on key cable nodes are used to capture multi-modal monitoring values ​​such as load current, surface temperature, operating voltage, ambient humidity, vibration frequency, and partial discharge pulses of the power cable during standardized operation, driven by the same time synchronization pulse. Then, wavelet packet decomposition denoising technology and a weighted moving average smoothing algorithm based on the latest sampling point weight gain are introduced to perform fine cleaning on the captured multi-modal monitoring values. This smoothing algorithm assigns alternating weights that decay over time to the time-series signals within a continuous sliding window. By using the centripetal averaging of new and old feature data, it can reduce isolated spikes and sudden glitches in the load current or surface temperature in real time, while dynamically offsetting the signal phase lag effect caused by traditional filtering. It also removes high-frequency random white noise and baseline drift interference in the power grid environment, thereby obtaining basic operating data of the reference base that characterizes the physical state of the cable.

[0100] A multi-source heterogeneous data time-series alignment technique is introduced, using basic operating condition data and dynamic adjustment command data as inputs. Through threshold configuration parameters provided by the dynamic adjustment commands, spatiotemporal mapping and gain compensation are performed on multi-modal monitoring values ​​such as load current, surface temperature, and partial discharge pulses. This eliminates nonlinear distortion caused by hardware adaptive adjustment, obtaining normalized sensing feature matrix data characterizing the actual physical operating conditions of power cables. A nonlinear causal graph extrapolation method is introduced, using microscopic correlation distortion features and macroscopic evolution trends in the matrix as nodes. The probabilistic relationship of insulation anomalies occurring at each node under electrothermal stress is calculated, obtaining causal link evolution data characterizing the correlation between microscopic fault mechanisms and macroscopic performance failures.

[0101] A cross-modal dynamic feature fusion technique is introduced to analyze causal link evolution data. The geometric distortion degree of the manifold space, density clustering offset, and the amplitude of time-axis sliding window abrupt changes are multidimensionally weighted and superimposed to construct a comprehensive evaluation manifold representing the feature mapping of all operating conditions within the parameter space. Then, a particle filter algorithm based on a high-dimensional state resampling mechanism is introduced to calculate the posterior probability distribution of the comprehensive evaluation manifold deviating from the healthy baseline in real time. The state particle weights are dynamically adjusted based on the measured manifold characteristics to calculate the cumulative rate of insulation dielectric damage probability on the time axis, obtaining probabilistic evaluation index data representing the potential fault evolution state.

[0102] A rule-matching technique based on the fault mode matrix is ​​introduced. First, the evolution rates of various probabilistic evaluation indicators are compared with the normal baseline state to determine whether the cable is currently in a healthy operating state or exhibits anomalies. If anomalies are identified, the evolution characteristics are further compared and mapped with typical damage rules such as dendritic discharge, electrochemical corrosion, and localized insulation breakdown to identify the specific severity of the damage. Based on this, a positioning technique based on dual-end traveling wave ranging is introduced, combined with microscopic degradation characteristics in the causal link evolution data, to calculate the precise location of the fault. The above state determination conclusions, specific defect types, and spatial location information of the fault location are deeply integrated to derive the power cable fault diagnosis result. If the state is determined to be normal, the fault diagnosis result is normal with no anomalies.

[0103] Based on the above steps S101-S106, intelligent monitoring of power cables throughout the entire process, from multi-dimensional data acquisition, blind noise reduction, manifold evolution to threshold closed-loop adjustment, is realized. While effectively eliminating interference from complex operating conditions, the sensitivity of micro-damage capture is significantly improved, thus providing efficient and robust cross-modal causal diagnostic support for early warning, accurate location and life prediction of insulation defects.

[0104] Based on the above technical solutions, optionally, based on dynamic adjustment command data and basic operating condition data, a cross-modal causal chain fusion diagnosis of microscopic correlation distortion characteristics and macroscopic evolution trends can be performed to obtain fault diagnosis results for power cables, including: Based on the dynamic adjustment command data and the basic operating condition data, differential calculation of the correlation strength of cross-modal anomaly characteristic manifold is performed to obtain correlation strength data characterizing the degree of synchronous evolution of abnormal signals inside the power cable. Based on the correlation strength data, a causal network is constructed to describe the micro-correlation distortion characteristics and macro-evolution trend, resulting in causal correlation network data that describes the causal dependence between the micro-anomalies and the macro-operating state of power cables. Based on the causal relationship network data, quantitative extraction of the characteristic anomaly attributes of each node in the causal relationship network is performed to obtain node anomaly data describing the offset of the operating state of the nodes in the causal relationship network of power cables. Based on the node variation data, the spacing of the state evolution trajectory manifold is quantized to obtain probability boundary data characterizing the probability of power cable fault trajectory matching. Based on the probability boundary data and causal correlation network data, a logical closed-loop verification of the consistency between the fault development source and the evolution path is performed to obtain path verification data that characterizes the confidence of the power cable fault evolution path. Based on the path verification data and basic operating condition data, a comprehensive judgment is made on the abnormality type and defect attribution dimension of the power cable operating status to obtain the fault diagnosis result data of the power cable.

[0105] In this scheme, the correlation strength data is obtained by differential calculation of the correlation strength of cross-modal anomaly characteristic manifold between dynamic adjustment command data and basic operating condition data, and is used to quantitatively describe the degree of coupling of abnormal signals inside power cables during the spatiotemporal synchronous evolution process.

[0106] Causal correlation network data is a complex network formed by constructing topological connections between microscopic correlation distortion characteristics and macroscopic evolution trends based on correlation strength data. It is used to clearly describe the causal dependency logic between microscopic local anomalies and the overall macroscopic operating state of power cables.

[0107] Node variation data is an index obtained by quantitatively extracting the characteristic variation attributes of each functional node in the causal relationship network. It is used to accurately describe the deviation of the operating status of each key monitoring node of the power cable in the causal network structure from the normal baseline.

[0108] Probability boundary data is calculated by quantifying the relative distance of the state evolution trajectory in the characteristic manifold space, and is used to define the degree of matching between the current operating state evolution of the power cable and the known fault characteristic trajectory and its statistical probability interval.

[0109] Path verification data is obtained by combining probability boundary data and causal relationship networks, and by performing logical closed-loop verification of the consistency of the entire link from the source of fault development to the end of evolution. It is used to measure the authenticity of the power cable fault evolution path and the confidence of the result.

[0110] When performing differential calculations of the correlation strength of cross-modal anomaly manifolds based on the aforementioned dynamic adjustment command data and basic operating condition data, the dynamic adjustment command data reflecting the channel capture sensitivity and the basic operating condition data serving as the measurement base are first read synchronously. Then, a manifold feature alignment technique is introduced to map the two sets of data to a high-dimensional feature space to construct control parameter manifolds and physical parameter manifolds, respectively. Next, a cross-modal manifold differential technique is introduced to subtract the local tangent space geometric distances of the two manifolds point by point, extracting the feature variations caused by internal anomalies. Finally, a sliding window covariance matrix technique is used to quantify the co-evolution trend of the differential manifold across monitoring dimensions, generating correlation strength data.

[0111] When constructing a causal network based on the aforementioned correlation strength data to determine the microscopic correlation distortion characteristics and macroscopic evolutionary trends, Granger causality testing is used to analyze the leading and lagging response relationships of the correlation strength data from each acquisition channel over time. Subsequently, Bayesian network topology construction is introduced, using the tested causal response relationships as directed edges, connecting them into a directed acyclic graph with microscopic correlation distortion characteristics as the starting point and macroscopic evolutionary trends as the ending point. Then, conditional probability density estimation is introduced to assign dependency weights to each edge online, clearly describing the causal dependency logic between microscopic local anomalies and the overall macroscopic operating state, thus obtaining the causal correlation network data.

[0112] When quantitatively extracting the characteristic anomaly attributes of each node in the causal network based on the aforementioned causal network data, a graph structure feature vector extraction technique is introduced. This involves traversing each monitoring functional node in the causal network data and calculating its in-degree, out-degree, and betweenness centrality. Subsequently, a Mahalanobis distance calculation technique is introduced to calculate the relative spatial distance between the current real-time network topology feature vector of each functional node and the healthy topology feature vector under standard normal operating conditions. This directly quantifies the degree of abnormal divergence of each node in the network topology structure, yielding node anomaly data.

[0113] When quantifying the spacing between state evolution trajectory manifolds based on the node variation data, a phase space reconstruction technique is introduced to project the node variation data within a continuous time period onto a high-dimensional state feature space, thus plotting the actual temporal evolution trajectory manifold. Subsequently, a Fraser distance manifold comparison technique is introduced to calculate the minimum spatial envelope distance between this trajectory manifold and the known fault feature trajectory manifold point by point. Then, a Gaussian probability density mapping technique is introduced to transform this distance into a statistical probability interval, which is used to define the degree of matching between the current operating state evolution and the known fault feature trajectory, generating probability boundary data.

[0114] When performing logical closed-loop verification of the consistency between the source and evolution path of a fault based on the probabilistic boundary data and causal correlation network data, a causal path tracing technique is introduced. Using the potential fault characteristic trajectory locked by the probabilistic boundary data as the endpoint, the technique traces backward along the directed topological edges of the causal correlation network data to locate the micro-level source of development. Next, a path consistency evaluation technique is introduced to calculate the overlap probability between the forward evolution path and the backward tracing path. Combined with the probability intervals of the probabilistic boundary data, a logical closed-loop verification of end-to-end consistency is performed to generate path verification data.

[0115] When comprehensively determining the types of abnormal operating conditions and defect attribution dimensions of power cables based on the path verification data and basic operating condition data, a multimodal expert system fusion reasoning technology is introduced. After the confidence level of the path verification data reaches the threshold, the locked fault evolution path is first cross-mapped with the load current, surface temperature, and partial discharge pulses in the basic operating condition data to determine whether the cable is currently in a healthy operating state or has an abnormality. If it is determined to be healthy, a normal diagnosis conclusion is output; if it is determined to be abnormal, the decoupling cause-based technology is further used to comprehensively assess the health of the insulation medium in the multi-dimensional mapping space, confirm the specific defect types such as dendritic discharge, electrochemical corrosion, or local insulation breakdown, and output the spatial location information of the fault location by combining the dual-end traveling wave ranging technology, thus deriving the final power cable fault diagnosis result data.

[0116] This solution achieves precise source tracing of faults through cross-modal causal networks and logical closed-loop verification. Its core advantage lies in its ability to extract the true degradation mechanism from massive amounts of monitoring data, significantly improving the qualitative and rate-of-flight diagnostic capability for complex insulation defects.

[0117] See appendix Figure 2 , Figure 2 This is a schematic flowchart of the second main step of a power cable fault diagnosis method based on nonlinear feature extraction according to an embodiment of the present invention. Figure 2 As shown, a power cable fault diagnosis method based on nonlinear feature extraction in an embodiment of the present invention mainly includes the following steps S201-S210.

[0118] Step S201: Acquire multi-mode signal data of the power cable, perform multi-channel time delay dynamic mapping and multi-dimensional spatial trajectory weaving based on the multi-mode signal data, and obtain multi-dimensional state space trajectory data characterizing the real-time operation of the power cable.

[0119] Step S202: Based on the multidimensional state space trajectory data, perform blind identification and non-contact spatial weighted filtering of the random disordered interference distribution density of the environment to obtain a multidimensional state association map after filtering out electromagnetic background noise.

[0120] Step S203: Based on the multidimensional state correlation map and the preset benchmark state correlation map, perform dynamic alignment of multidimensional feature manifolds and nonlinear spatial overlap comparison to obtain real-time morphological distortion rate data characterizing the microscopic anomalies of the internal insulation state of the power cable.

[0121] Step S204: Based on the real-time morphological distortion rate data, perform dynamic response correlation degree calculation on the correlation distortion evolution trend under cross-channel monitoring time sequence to obtain state correlation matrix data characterizing the synchronicity of abnormal changes in power cable state distortion.

[0122] Step S205: Based on the state correlation matrix data, perform spontaneous affinity clustering of the feature vector group under unlabeled states to obtain an initial feature cluster that characterizes the operating mode of the power cable.

[0123] Step S206: Based on the initial feature clusters, initialize and define the distribution of the centroid of the cluster manifold and the parameter scale to obtain the manifold core parameter data characterizing the distribution center of the power cable state modes.

[0124] Step S207: Based on the manifold boundary description parameters, perform closed-loop verification of the cluster boundary convergence state and parameter manifold stability to obtain convergence verification metric data characterizing the convergence state of the power cable operating mode.

[0125] Step S208: Based on the convergence verification metric data and manifold boundary description parameters, perform consistency reconstruction and mode feature locking of the multidimensional state mode distribution map to obtain a state classification characterization cluster that represents the manifold distribution state of the power cable state signal group.

[0126] Step S209: Based on the state classification characterization cluster, perform online evaluation of the severity of feature mutations and adaptive adjustment of feature capture sensitivity to obtain dynamic adjustment instruction data for correcting the trigger threshold of the power cable signal acquisition channel.

[0127] Step S210: Obtain basic operating data of the power cable, and based on the dynamic adjustment command data and basic operating data, perform cross-modal causal chain fusion diagnosis of micro-correlation distortion characteristics and macro-evolution trend to obtain the fault diagnosis results of the power cable.

[0128] In this embodiment, the state correlation matrix data is a correlation index that quantifies the dynamic response degree between various distortion evolution trends under the cross-channel monitoring time sequence of power cables. It is used to intuitively characterize the synchronicity and coupling strength of insulation abnormal changes between different monitoring locations or dimensions.

[0129] The initial feature cluster is based on the spontaneous affinity clustering calculation results of the feature vector group under unlabeled state. It divides the complex operation data of power cable into several subsets with similar dynamic characteristics, so as to initially characterize the distribution groups of each state mode under the current operating conditions.

[0130] The core parameter data of the manifold is a set of parameters that are mathematically quantified and defined based on the initial feature clusters, the centroid distribution, centroid, and scale range of the clustered manifold in the high-dimensional feature space. It is used to accurately locate the center coordinates and spatial morphological distribution of various operating modes of power cables.

[0131] The manifold boundary description parameters are geometric and topological characterization parameters obtained by combining the core parameters of the manifold and adaptive iterative evolution of the hypersurface clustered in the high-dimensional feature space. They are used to define the distribution boundaries of signals under different operating conditions of power cables in the parameter space.

[0132] Convergence verification metrics are evaluation indicators obtained by performing closed-loop verification on the convergence of manifold boundaries and the stability of parametric manifolds. They are used to quantify the stability of the power cable operation mode identification model in multidimensional space and the accuracy of mode division.

[0133] When calculating the dynamic response correlation degree of the correlation distortion evolution trend under cross-channel monitoring time series based on the real-time morphological distortion rate data, the real-time morphological distortion rate data of each acquisition channel is first read, and the time series parameters are aligned along the time axis. Then, dynamic time warping technology is introduced to nonlinearly adjust the time axis correspondence of the distortion evolution trends between different channels under continuous time series, and the distance of the deviation degree of the parameter network manifold structure is calculated. Next, sliding window mutual information technology is introduced to calculate the nonlinear spatiotemporal dependence of different monitoring channels on the micro-insulation degradation rate online, quantify the synchronicity and coupling strength of insulation anomaly changes between different monitoring locations, and generate state correlation matrix data characterizing the synchronicity of anomaly changes.

[0134] When performing spontaneous affinity clustering of feature vector groups under unlabeled states based on the state association matrix data, the state association matrix data is invoked, and spectral clustering technology is introduced. The state association matrix data is used as similarity weights to construct a Laplacian matrix in a high-dimensional feature space. Subsequently, eigenvalue decomposition technology is introduced to reduce the dimensionality of the matrix and extract the feature vector groups, transforming the complex operational data into a low-dimensional manifold space. Next, self-organizing map neural network technology is introduced to perform spatial topological competition based on the spontaneous affinity of the feature vector groups, automatically merging data subsets with similar dynamic characteristics, dividing them into different state mode distribution groups, and obtaining the initial feature clusters.

[0135] When initializing the centroid distribution and parameter scale of the clustered manifold based on the initial feature clusters, the initial feature clusters are read, and a high-dimensional space centroid calculation technique is introduced. The sample points of each feature group within the initial feature clusters are traversed, and their geometrically weighted average center in the parameter space is calculated to determine the center coordinates of various operating modes. Then, a covariance matrix feature extraction technique is introduced to quantify the discrete distribution gradient of sample points relative to the center coordinates, extracting the eigenvalues ​​and eigenvectors of the high-dimensional clustered manifold to define the axial range and scale of its spatial morphological distribution, and outputting the core parameter data of the manifold.

[0136] When performing adaptive iterative evolution of the hypersurface boundary in the high-dimensional feature space based on the manifold core parameter data and the initial feature clusters, the manifold core parameter data is used as the starting point of the evolution, and the initial feature clusters are used as sample references. Support vector data description technology is introduced to establish a hypersurface that wraps around the sample points in the high-dimensional feature space. Subsequently, gradient descent adaptive optimization technology is introduced, using the center coordinates and parameter scale in the manifold core parameter data as constraints, and adaptively adjusting the boundary curvature along the normal vector direction of the hypersurface until the hypersurface accurately defines the distribution limit of the signal in the parameter space, thus obtaining the manifold boundary description parameters.

[0137] When performing closed-loop verification of the clustering boundary convergence state and parametric manifold stability based on the manifold boundary description parameters, the manifold boundary description parameters are read, and a perturbation sampling closed-loop verification technique is introduced. Weak Gaussian white noise is randomly added to the current operating data of the power cable, and the data is then remapped onto the hypersurface defined by the manifold boundary description parameters. Next, a Hess matrix stability verification technique is introduced to calculate the rate of change of the first and second partial derivatives of the hypersurface boundary under noise perturbation, quantifying the local deformation rate of the hypersurface boundary. This closed-loop evaluation assesses the stability of the boundary in multidimensional space and the accuracy of pattern partitioning, generating convergence verification metric data.

[0138] When performing consistent reconstruction and mode feature locking of multidimensional state mode distribution maps based on the convergence verification metric data and manifold boundary description parameters, the convergence verification metric data and manifold boundary description parameters are simultaneously invoked. A consistent manifold graph embedding technique is introduced. After the convergence verification metric data meets the stability index, the hypersurface geometric boundaries locked in the manifold boundary description parameters are used to cluster the continuous manifold morphological trajectories within the parameter space. Then, a high-dimensional fingerprint feature mapping technique is introduced to lock the dynamic dynamic mechanism within each closed boundary, transforming the continuous state data into a discrete mathematical set with well-defined boundaries, resulting in a state classification representation cluster that characterizes the high-dimensional distribution fingerprint of the latent state modes.

[0139] In this embodiment, by using dynamic response correlation calculation and topological clustering, the insulation characteristics under different operating conditions are accurately mapped. This not only effectively suppresses cross-channel environmental noise, but also realizes the automated convergence and locking of cable operation modes, providing a high-precision criterion for early warning of microscopic insulation degradation.

[0140] It should be noted that although the steps in the above embodiments are described in a specific order, those skilled in the art will understand that in order to achieve the effects of the present invention, different steps do not necessarily have to be executed in such an order. They can be executed simultaneously (in parallel) or in other orders, and these variations are all within the scope of protection of the present invention.

[0141] Furthermore, the present invention also provides a power cable fault diagnosis system based on nonlinear feature extraction.

[0142] See appendix Figure 3 , Figure 3 This is a main structural block diagram of a power cable fault diagnosis system based on nonlinear feature extraction according to an embodiment of the present invention. Figure 3 As shown, it specifically includes: The multi-dimensional trajectory construction module 301 is used to acquire multi-mode signal data of power cables, and perform multi-channel time delay dynamic mapping and multi-dimensional spatial trajectory weaving based on the multi-mode signal data to obtain multi-dimensional state space trajectory data that characterizes the real-time operation of power cables. The noise interference filtering module 302 is used to perform blind identification and non-contact spatial weighted filtering of the random and disordered interference distribution density of the environment based on multidimensional state space trajectory data, so as to obtain a multidimensional state correlation map after filtering out electromagnetic background noise. The dynamic distortion comparison module 303 is used to perform dynamic alignment and nonlinear spatial overlap comparison of multidimensional feature manifolds based on multidimensional state correlation maps and preset benchmark state correlation maps, so as to obtain real-time morphological distortion rate data that characterizes the microscopic anomalies of the internal insulation state of power cables. The state clustering evolution module 304 is used to perform spontaneous association of affinity and cluster boundary iterative evolution under unlabeled state based on real-time morphological distortion rate data to obtain state classification characterization clusters that characterize the manifold distribution state of power cable state signal group. The dynamic instruction generation module 305 is used to perform online evaluation of the severity of feature mutations and adaptive adjustment of feature capture sensitivity based on state classification characterization clusters, so as to obtain dynamic adjustment instruction data for correcting the trigger threshold of the power cable signal acquisition channel. The diagnostic module 306 is used to acquire basic operating data of power cables, and based on dynamic adjustment command data and basic operating data, to perform cross-modal causal chain fusion diagnosis of micro-correlation distortion characteristics and macro-evolution trend to obtain fault diagnosis results of power cables.

[0143] The power cable fault diagnosis system based on nonlinear feature extraction provided in this application embodiment can achieve... Figure 1 The various processes implemented in the method implementation examples will not be described again here to avoid repetition.

[0144] Those skilled in the art will understand that all or part of the processes in the method of the above embodiment of the present invention can also be implemented by a computer program instructing related hardware. The computer program can be stored in a computer-readable storage medium, and when executed by a processor, it can implement the steps of the various method embodiments described above. The computer program includes computer program code, which can be in the form of source code, object code, executable file, or some intermediate form. The computer-readable storage medium can include any entity or device capable of carrying the computer program code, a medium, a USB flash drive, a portable hard drive, a magnetic disk, an optical disk, a computer memory, a read-only memory, a random access memory, an electrical carrier signal, a telecommunication signal, and a software distribution medium, etc. It should be noted that the content included in the computer-readable storage medium can be appropriately added or removed according to the requirements of legislation and patent practice in the jurisdiction. For example, in some jurisdictions, according to legislation and patent practice, the computer-readable storage medium does not include electrical carrier signals and telecommunication signals.

[0145] Furthermore, the present invention also provides an electronic device 400, including a processor 401, a memory 402, and a program or instructions stored in the memory 402 and executable on the processor 401. When the program or instructions are executed by the processor 401, they implement the various processes of the above-described embodiment of a power cable fault diagnosis method based on nonlinear feature extraction and achieve the same technical effect. To avoid repetition, they will not be described again here.

[0146] It should be noted that the electronic devices in the embodiments of this application include the mobile electronic devices and non-mobile electronic devices described above.

[0147] Furthermore, the present invention also provides a computer-readable storage medium. In one embodiment of the computer-readable storage medium according to the present invention, the computer-readable storage medium can be configured to store a program for executing a power cable fault diagnosis method based on nonlinear feature extraction according to the above-described method embodiments. This program can be loaded and run by a processor to implement the above-described power cable fault diagnosis method based on nonlinear feature extraction. For ease of explanation, only the parts related to the embodiments of the present invention are shown; for specific technical details not disclosed, please refer to the method section of the embodiments of the present invention. The computer-readable storage medium can be a storage device comprising various electronic devices. Optionally, in the embodiments of the present invention, the computer-readable storage medium is a non-transitory computer-readable storage medium.

[0148] Furthermore, it should be understood that since the various modules are only provided to illustrate the functional units of the device of the present invention, the physical devices corresponding to these modules may be the processor itself, or a part of the processor's software, hardware, or a combination of software and hardware. Therefore, the number of modules shown in the figures is merely illustrative.

[0149] Those skilled in the art will understand that the various modules in the device can be adaptively split or combined. Such splitting or combining of specific modules will not cause the technical solution to deviate from the principles of the present invention; therefore, the technical solutions after splitting or combining will fall within the protection scope of the present invention.

[0150] The technical solution of the present invention has been described above with reference to the preferred embodiments shown in the accompanying drawings. However, it will be readily understood by those skilled in the art that the scope of protection of the present invention is obviously not limited to these specific embodiments. Without departing from the principles of the present invention, those skilled in the art can make equivalent changes or substitutions to the relevant technical features, and the technical solutions after such changes or substitutions will all fall within the scope of protection of the present invention.

Claims

1. A power cable fault diagnosis method based on nonlinear feature extraction, characterized in that, The method includes: Multimodal signal data of power cables are acquired, and multi-channel time delay dynamic mapping and multi-dimensional spatial trajectory weaving are performed based on the multimodal signal data to obtain multi-dimensional state space trajectory data that characterizes the real-time operation of power cables. Based on multidimensional state space trajectory data, blind identification and non-contact spatial weighted filtering of the distribution density of random and disordered environmental interference are performed to obtain a multidimensional state correlation map after filtering out electromagnetic background noise. Based on the multidimensional state correlation map and the preset benchmark state correlation map, dynamic alignment of multidimensional feature manifolds and nonlinear spatial overlap comparison are performed to obtain real-time morphological distortion rate data characterizing the microscopic anomalies of the internal insulation state of power cables. Based on real-time morphological distortion rate data, spontaneous association of affinity and cluster boundary iterative evolution under unlabeled state is carried out to obtain state classification characterization clusters that characterize the manifold distribution state of power cable state signal group. Based on state classification characterization clusters, online evaluation of the severity of feature mutations and adaptive adjustment of feature capture sensitivity are performed to obtain dynamic adjustment command data for correcting the trigger threshold of power cable signal acquisition channels. By acquiring basic operating data of power cables, and based on dynamic adjustment command data and basic operating data, cross-modal causal chain fusion diagnosis of micro-correlation distortion characteristics and macro-evolution trend is performed to obtain fault diagnosis results of power cables.

2. The power cable fault diagnosis method based on nonlinear feature extraction according to claim 1, characterized in that, in, Based on multi-modal signal data, multi-channel time delay dynamic mapping and multi-dimensional spatial trajectory weaving are performed to obtain multi-dimensional state-space trajectory data characterizing the real-time operation of power cables, including: Based on the multi-mode signal data, dynamic mapping of multi-channel time delay is performed to obtain time delay mapping state data characterizing the coupling timing state of each phase of the power cable. Based on the time delay mapping state data, trajectory cross weaving is performed in the high-dimensional phase space to obtain a multi-dimensional reconstructed phase space trajectory that characterizes the coordinated changes of power cable pulse interference and normal operating load. Based on the multidimensional reconstructed phase space trajectory, continuous time slice segmentation of the dynamic trajectory is performed to obtain time slice trajectory data characterizing the continuous monitoring time sequence of the power cable. Based on the time slice trajectory data, feature reconstruction of multidimensional feature manifold morphology is performed to obtain feature spatial distribution contour data that characterizes the internal state evolution law of power cable. Based on the feature space distribution contour data, weighted feature extraction of manifold trajectory distribution density is performed to obtain manifold density weight data that characterizes the evolution law of power cable operating condition distribution characteristics. Based on the manifold density weight data and the multidimensional reconstructed phase space trajectory, the correlation and fusion calibration of the intrinsic dynamic characteristics are performed to obtain multidimensional state space trajectory data characterizing the real-time operation of the power cable.

3. The power cable fault diagnosis method based on nonlinear feature extraction according to claim 1, characterized in that, in, Based on multidimensional state-space trajectory data, blind identification and non-contact spatial weighted filtering of the distribution density of random and disordered environmental interference are performed to obtain a multidimensional state correlation map after filtering out electromagnetic background noise, including: Based on the multidimensional state space trajectory data, the spatial disordered interference distribution density is automatically identified to obtain interference distribution density data that characterizes the interference characteristics of the power cable field operation environment. Based on the interference distribution density data, the blind zone boundary of the environmental disordered random noise is delineated to obtain interference boundary data characterizing the coverage range of environmental noise for power cable monitoring. Based on the interference boundary data, a non-contact multidimensional weighting matrix is ​​calculated online in real time to obtain a multidimensional weighting coefficient matrix for suppressing interference in the power cable acquisition channel. Based on the multidimensional weighted coefficient matrix and multidimensional state space trajectory data, multidimensional feature space weighted mapping and noise amplitude suppression are performed to obtain a weighted attenuation feature sequence after achieving stray interference attenuation of power cables. Based on the weighted attenuation feature sequence, the abnormal mutation state features are selectively stripped and eliminated to obtain a pure state evolution sequence that characterizes the operating condition of the power cable. Based on the pure state evolution sequence, a parameter correlation network is constructed and its consistency is calibrated to obtain a multidimensional state correlation map after filtering out electromagnetic background noise.

4. The power cable fault diagnosis method based on nonlinear feature extraction according to claim 1, characterized in that, in, Based on multidimensional state correlation maps and preset benchmark state correlation maps, dynamic alignment and nonlinear spatial overlap comparison of multidimensional feature manifolds are performed to obtain real-time morphological distortion rate data characterizing the microscopic anomalies of the internal insulation state of power cables, including: Based on the multidimensional state association map and the preset benchmark state association map, multidimensional feature space feature extraction and initial registration of association centroid are performed to obtain the initial registration association features characterizing the initial working condition alignment state of the power cable. Based on the initial registration association features, dynamic affine transformation and manifold matching alignment of the parameter association network are performed to obtain dynamic alignment network data that characterizes the spatiotemporal consistency features of the multidimensional manifold of the power cable. Based on the dynamic alignment network data, feature extraction is performed on the nonlinear coincident manifold direction matrix projection to obtain coincidence projection feature data characterizing the spatial mapping characteristics of power cable parameters. Based on the overlap projection feature data, the correlation difference solution of the local deviation variation trend is performed to obtain the correlation local deviation data characterizing the spatial distribution of the power cable parameter network. Based on the associated local deviation data, the associated evolution density integral quantization of the nonlinear distortion region is performed to obtain associated distortion integral data characterizing the degree of microstate evolution of the power cable. Based on the associated distortion integral data, a continuous time-series mapping of the dynamic disturbance ratio is performed to obtain real-time morphological distortion rate data characterizing the microscopic anomalies in the internal insulation state of the power cable.

5. The power cable fault diagnosis method based on nonlinear feature extraction according to claim 1, characterized in that, in, Based on real-time morphological distortion rate data, spontaneous association and iterative evolution of cluster boundaries are performed under unlabeled conditions to obtain state classification characterization clusters representing the manifold distribution of power cable state signal groups, including: Based on the real-time morphological distortion rate data, the dynamic response correlation degree of the correlation distortion evolution trend under the cross-channel monitoring time series is calculated to obtain the state correlation matrix data characterizing the synchronicity of abnormal changes in the state distortion of power cables. Based on the state correlation matrix data, spontaneous affinity clustering of feature vector groups under unlabeled states is performed to obtain an initial feature cluster that characterizes the operating mode of power cables. Based on the initial feature clusters, the distribution of the centroid of the cluster manifold and the parameter scale are initialized and defined to obtain the manifold core parameter data characterizing the distribution center of the power cable state modes. Based on the core manifold parameter data and the initial feature clusters, an adaptive iterative evolution of the clustered hypersurface boundary in the high-dimensional feature space is performed to obtain the manifold boundary description parameters characterizing the signal distribution boundary of the power cable. Based on the manifold boundary description parameters, a closed-loop verification of the cluster boundary convergence state and the parameter manifold stability is performed to obtain convergence verification metric data characterizing the convergence state of the power cable operating mode. Based on the convergence verification metric data and manifold boundary description parameters, a consistent reconstruction and mode feature locking of the multidimensional state mode distribution map are performed to obtain a state classification characterization cluster that represents the manifold distribution state of the power cable state signal group.

6. The power cable fault diagnosis method based on nonlinear feature extraction according to claim 1, characterized in that, in, Based on state classification characterization clusters, online evaluation of the severity of feature mutations and adaptive adjustment of feature capture sensitivity are performed to obtain dynamic adjustment command data for correcting the trigger threshold of power cable signal acquisition channels, including: Based on the state classification characterization cluster, online calculation of the centroid drift evolution of the feature space manifold is performed to obtain feature mutation data characterizing the degree of characteristic mode shift of the power cable. Based on the aforementioned characteristic mutation data, a quantitative assessment of the diffusion rate of the multidimensional manifold of the time-series evolution trajectory is performed to obtain mutation assessment data characterizing the characteristic mutation trend of power cables. Based on the mutation assessment data, adaptive matching of the channel trigger threshold parameter control boundary is performed to obtain threshold matching data characterizing the initial scale of the power cable channel trigger threshold. Based on the threshold matching data and feature mutation data, closed-loop correction of the channel trigger threshold adjustment sensitivity is performed to obtain sensitivity correction data characterizing the dynamic compensation state of the characteristic amplitude of the power cable. Based on the sensitivity correction data, a sliding window iterative optimization calculation of the dynamic correction amount of the channel trigger threshold is performed to obtain the threshold distribution adjustment vector characterizing the distribution of the multi-channel trigger threshold array of the power cable. Based on the threshold distribution adjustment vector and the state classification characterization cluster, the dynamic fine-tuning parameters of the multi-channel trigger threshold are mapped and fused to obtain dynamic adjustment command data for correcting the trigger threshold of the power cable signal acquisition channel.

7. The power cable fault diagnosis method based on nonlinear feature extraction according to claim 1, characterized in that, in, Based on dynamic adjustment command data and basic operating condition data, a cross-modal causal chain fusion diagnosis of micro-correlation distortion characteristics and macro-evolutionary trends is performed to obtain fault diagnosis results for power cables, including: Based on the dynamic adjustment command data and the basic operating condition data, differential calculation of the correlation strength of cross-modal anomaly characteristic manifold is performed to obtain correlation strength data characterizing the degree of synchronous evolution of abnormal signals inside the power cable. Based on the correlation strength data, a causal network is constructed to describe the micro-correlation distortion characteristics and macro-evolution trend, resulting in causal correlation network data that describes the causal dependence between the micro-anomalies and the macro-operating state of power cables. Based on the causal relationship network data, quantitative extraction of the characteristic anomaly attributes of each node in the causal relationship network is performed to obtain node anomaly data describing the offset of the operating state of the nodes in the causal relationship network of power cables. Based on the node variation data, the spacing of the state evolution trajectory manifold is quantized to obtain probability boundary data characterizing the probability of power cable fault trajectory matching. Based on the probability boundary data and causal correlation network data, a logical closed-loop verification of the consistency between the fault development source and the evolution path is performed to obtain path verification data that characterizes the confidence of the power cable fault evolution path. Based on the path verification data and basic operating condition data, a comprehensive judgment is made on the abnormality type and defect attribution dimension of the power cable operating status to obtain the fault diagnosis result data of the power cable.

8. A power cable fault diagnosis system based on nonlinear feature extraction, characterized in that, The system includes: The multi-dimensional trajectory construction module is used to acquire multi-mode signal data of power cables, and to perform multi-channel time delay dynamic mapping and multi-dimensional spatial trajectory weaving based on the multi-mode signal data to obtain multi-dimensional state space trajectory data that characterizes the real-time operation of power cables. The noise interference filtering module is used to perform blind identification and non-contact spatial weighted filtering of the random and disordered interference distribution density of the environment based on multidimensional state space trajectory data, so as to obtain a multidimensional state correlation map after filtering out electromagnetic background noise. The dynamic distortion comparison module is used to perform dynamic alignment and nonlinear spatial overlap comparison of multidimensional feature manifolds based on multidimensional state correlation maps and preset benchmark state correlation maps, so as to obtain real-time morphological distortion rate data that characterizes the microscopic anomalies of the internal insulation state of power cables. The state clustering evolution module is used to perform spontaneous association of affinity and cluster boundary iterative evolution under unlabeled state based on real-time morphological distortion rate data, so as to obtain state classification characterization clusters that characterize the manifold distribution state of power cable state signal group. The dynamic instruction generation module is used to perform online evaluation of the severity of feature mutations and adaptive adjustment of feature capture sensitivity based on state classification characterization clusters, so as to obtain dynamic adjustment instruction data for correcting the trigger threshold of the power cable signal acquisition channel. The diagnostic module is used to acquire basic operating data of power cables. Based on dynamic adjustment command data and basic operating data, it performs cross-modal causal chain fusion diagnosis of micro-correlation distortion characteristics and macro-evolution trends to obtain fault diagnosis results of power cables.

9. An electronic device comprising a processor, a memory, and a program or instructions stored in the memory and executable on the processor, characterized in that, The program or instructions are adapted to be loaded and run by the processor to perform a power cable fault diagnosis method based on nonlinear feature extraction as described in any one of claims 1 to 7.

10. A computer-readable storage medium storing a plurality of program codes, characterized in that, The program code is adapted to be loaded and run by a processor to perform a power cable fault diagnosis method based on nonlinear feature extraction as described in any one of claims 1 to 7.

Images