Multi-state micro-power consumption sensing system for power distribution cable

Abstract

This invention relates to the field of intelligent monitoring technology for power equipment. To address the problems of isolated cable condition monitoring parameters, difficulty in deeply correlating with fundamental insulation aging indicators, and fuzzy anomaly location, a multi-state low-power sensing system for distribution cables is disclosed. This system acquires cable sheath temperature, discharge waveform, and insulation loss data through an information acquisition module; a condition generation module processes the data to obtain comprehensive cable condition characteristics; a weight learning module uses a trained anomaly representation learning network to generate an anomaly weight spectrum reflecting the correlation strength between operating conditions and insulation loss; a source tracing and location module performs dual-source anomaly trajectory backtracking analysis on the original temperature and discharge data based on this weight spectrum, achieving precise location of the anomaly's physical location and identification of its inherent defect type; and a parameter optimization module generates and executes a control parameter adjustment scheme. This invention achieves deep correlation assessment of cable insulation status and precise physical-level diagnosis of anomaly root causes.

Description

Technical Field

[0001] This invention relates to the field of intelligent monitoring technology for power equipment, specifically a multi-state low-power sensing system for power distribution cables. Background Technology

[0002] Currently, monitoring the operating status of power distribution cables mainly relies on the independent acquisition of single or multiple physical parameters and threshold alarms. Common technical solutions include deploying temperature sensors at key points of the cable for point temperature monitoring, or using ultra-high frequency and ultrasonic sensors to capture partial discharge signals and perform simple pattern recognition and location. These methods can issue alarms for significant abnormalities such as cable overheating and severe discharge.

[0003] These conventional technical solutions have shortcomings. Parameters such as temperature and discharge are monitored and evaluated in isolation, lacking a deep and quantifiable correlation analysis between the overall cable operating condition reflected in the data and the actual aging and loss state of the insulation material. The system struggles to determine whether instantaneous temperature fluctuations or occasional discharge pulses truly constitute valid signs of insulation performance degradation, easily leading to false alarms or missed alarms. Even when an anomaly is detected, existing positioning technologies mostly rely on a single signal source for rough spatial location estimation, unable to simultaneously integrate heterogeneous data from multiple sources such as heat and electricity to trace the complete trajectory of the anomaly's generation and development. Therefore, it is difficult to accurately determine the specific physical origin of the anomaly and its corresponding microscopic defect type.

[0004] There is a need for an analysis method and system that can deeply integrate multi-state information, deeply correlate fundamental indicators of insulation aging, and achieve physical-level precise tracing of anomalies, in order to overcome the current limitations of cable status perception which is limited to surface parameter alarms. Summary of the Invention

[0005] The purpose of this invention is to provide a multi-state low-power sensing system for power distribution cables to solve the problems mentioned in the background art.

[0006] To achieve the above objectives, the present invention provides a multi-state low-power sensing system for power distribution cables, the system comprising:

[0007] The information acquisition module collects three types of raw monitoring information that reflect the real-time operating status of the cable. These three types of raw monitoring information include cable sheath temperature scanning data, transient waveforms of discharge phenomena, and insulation material loss values.

[0008] The operating condition generation module performs three-dimensional thermal field gradient calculation on the cable sheath temperature scanning data, outputs the cable surface temperature change pattern, and performs pulse event deconstruction analysis on the transient waveform of the discharge phenomenon, outputs the discharge activity distribution map. Combining the cable surface temperature change pattern and the discharge activity distribution map, the module generates comprehensive cable operating condition characteristics.

[0009] The weight learning module inputs the comprehensive working condition characteristics of the cable and the insulation material loss value into the trained anomaly representation learning network, and outputs the insulation loss associated state anomaly weight spectrum generated by the state metric layer inside the anomaly representation learning network.

[0010] The source tracing and positioning module, based on the abnormal weight spectrum of the insulation loss associated state, performs dual-source abnormal trajectory back-tracing analysis on the cable sheath temperature scanning data and the transient waveform of the discharge phenomenon to locate the physical origin of the abnormal state and identify its inherent defect category.

[0011] The parameter tuning module generates a real-time tuning scheme for the cable operation control parameters based on the physical origin location and inherent defect category of the abnormal state, and delivers the real-time tuning scheme to the cable management execution terminal.

[0012] Preferably, the process of calculating the three-dimensional thermal field gradient of the cable sheath temperature scanning data and outputting the cable surface temperature change pattern includes the following steps:

[0013] A fine grid of temperature sampling points is established in the physical surface coordinate system of the cable; spatial neighborhood heat flow extrapolation is performed on the temperature reading of each point in the temperature sampling point grid to calculate the local heat flow vector of each point; the local heat flow vectors corresponding to all temperature sampling points are aggregated to form an original heat flow field map that reflects the overall heat flow direction of the cable.

[0014] By introducing a cable structure geometric model and a material thermal property parameter library, a field strength correction operation based on physical structure is performed on the original thermal flow field spectrum to generate a structure-calibrated thermal flow field spectrum.

[0015] A sliding window feature analysis in the time dimension is performed on the structure-calibrated thermal flow field spectrum to capture the temporal evolution of the thermal flow pattern and output the temperature change pattern of the cable surface.

[0016] Preferably, the step of performing pulse event deconstruction analysis on the transient waveform of the discharge phenomenon and outputting a discharge activity distribution map includes the following specific steps:

[0017] Multiple independent discharge pulse event segments are separated from the transient waveform of the discharge phenomenon; for each discharge pulse event segment, waveform morphological decomposition is performed at multiple time scales to extract the set of morphological components of the discharge pulse event at different scales.

[0018] Perform spectral energy distribution analysis on each morphological component set to map the temporal morphological features to the frequency domain energy distribution features, forming the frequency domain feature vector of the discharge pulse event;

[0019] A self-organizing feature mapping network is used to cluster and summarize the frequency domain feature vectors corresponding to all discharge pulse event segments to identify typical discharge activity pattern categories with common characteristics.

[0020] The frequency, intensity, and spatial distribution of various typical discharge activity patterns within a preset monitoring period are statistically analyzed, and the discharge activity distribution map is encoded and output in the form of a graph.

[0021] Preferably, the generation of comprehensive cable operating condition characteristics by combining the cable surface temperature change pattern and the discharge activity distribution map is achieved through the following interactive process:

[0022] Establish a time-space synchronization index between cable surface temperature change patterns and discharge activity distribution maps;

[0023] Based on the aforementioned time-space synchronization index, a correlation matrix is ​​constructed to characterize the interaction between temperature field changes and discharge activity;

[0024] Using the cross-domain feature extractor guided by the correlation matrix, a subset of temperature features related to discharge activity is extracted from the temperature change pattern of the cable surface, and a subset of discharge features related to temperature field changes is extracted from the discharge activity distribution map.

[0025] The temperature feature subset and the discharge feature subset are input into the feature fusion engine. The feature fusion engine performs iterative cross-updates and deep integration of the two subsets by simulating a physical interaction process, and finally forms a comprehensive cable condition feature that characterizes the overall operating status of the cable.

[0026] Preferably, the anomaly representation learning network includes: a feature encoding layer, an association learning layer, and a state measurement layer;

[0027] The feature encoding layer converts the comprehensive cable operating condition features into a set of high-dimensional abstract feature vectors;

[0028] The association learning layer constructs a nonlinear mapping relationship between the high-dimensional abstract feature vector and the insulation material loss value, and dynamically learns the sensitivity of each feature dimension to the insulation material loss value during the mapping process.

[0029] The state metric layer reorganizes each dimension of the high-dimensional abstract feature vector according to the contribution sensitivity to generate the insulation loss associated state anomaly weight spectrum, wherein the feature dimension corresponding to the spectral peak of the insulation loss associated state anomaly weight spectrum indicates the anomaly state that is highly associated with insulation loss.

[0030] Preferably, the feature encoding layer converts the comprehensive operating condition features of the cable into a set of high-dimensional abstract feature vectors, which is achieved through the following processing:

[0031] The comprehensive operating conditions of the cable are input into a feature encoder with a multilayer perceptron structure.

[0032] In the first hidden layer of the feature encoder, the comprehensive working condition features of the cable are subjected to linear transformation and nonlinear activation processing to generate a primary abstract feature representation.

[0033] The primary abstract feature representation is passed to the second hidden layer, and local feature combinations are extracted through the convolution kernel sliding window operation to form the intermediate abstract feature representation;

[0034] The intermediate-level abstract feature representation is input into the output layer of the feature encoder. Global feature integration and dimensional expansion are performed through a fully connected network, and finally a set of high-dimensional abstract feature vectors with dimensions much higher than the original input are output.

[0035] Preferably, the execution steps for performing dual-source anomaly trajectory backtracking analysis on the cable sheath temperature scanning data and the transient waveform of the discharge phenomenon are as follows:

[0036] Guided by the insulation loss associated state anomaly weight spectrum, the starting point and diffusion path of temperature anomaly evolution are traced in reverse in the temperature field formed by the cable sheath temperature scanning data to form a temperature field anomaly trajectory. Simultaneously, guided by the insulation loss associated state anomaly weight spectrum, the initial moment and evolution process of discharge mode distortion are located in reverse in the discharge activity sequence contained in the transient waveform of the discharge phenomenon to form a discharge activity anomaly trajectory.

[0037] A spatiotemporal coupling analysis model is established between the abnormal trajectory of the temperature field and the abnormal trajectory of the discharge activity. The physical coordinate point where the two trajectories intersect in space and time is solved by the spatiotemporal coupling analysis model, and the physical coordinate point is determined as the physical origin of the abnormal state.

[0038] The coupling pattern between the abnormal temperature field trajectory and the abnormal discharge activity trajectory at the physical origin location is analyzed, and the coupling pattern is matched with a pre-stored defect pattern library to identify the intrinsic defect category.

[0039] Preferably, the establishment of the spatiotemporal coupling analysis model between the abnormal temperature field trajectory and the abnormal discharge activity trajectory specifically includes:

[0040] Define a spatiotemporal parameterized equation to describe the abnormal trajectory of the temperature field, and define a spatiotemporal parameterized equation to describe the abnormal trajectory of the discharge activity.

[0041] A coupled objective function is constructed with the goal of minimizing the spatiotemporal distance between two trajectories. The coupled objective function is solved using a numerical optimization algorithm to obtain the set of coupling parameters that make the two trajectories closest in spatiotemporal space.

[0042] Based on the set of coupling parameters, the intersection point of the abnormal temperature field trajectory and the abnormal discharge activity trajectory is calculated, and the intersection point is the physical coordinate point output by the spatiotemporal coupling analysis model.

[0043] Preferably, the analysis of the coupling mode between the abnormal temperature field trajectory and the abnormal discharge activity trajectory at the physical origin location includes:

[0044] Extract the temperature gradient change feature sequence near the intersection point from the abnormal temperature field trajectory;

[0045] Extract the discharge pulse intensity and repetition rate feature sequences near the intersection point from the abnormal discharge activity trajectory; synchronously compare the temperature gradient change feature sequence with the discharge pulse intensity and repetition rate feature sequence to form a set of coupled feature pairs;

[0046] The coupled feature set is input into the trained defect classifier, which outputs the determination result of the intrinsic defect category based on the internally stored mapping relationship between coupling patterns and defect categories.

[0047] Preferably, the real-time optimization scheme for generating cable operation control parameters includes the following decision-making steps:

[0048] Analyze the physical origin of the abnormal state to determine the specific affected cable line section; combine the inherent defect category with the operation and maintenance knowledge base to obtain a set of standard intervention measures for the inherent defect category;

[0049] By integrating the current load data, environmental parameters, and historical operation records of the specific cable line section, an adaptability and feasibility assessment is conducted on each measure in the set of standard intervention measures;

[0050] Based on the evaluation results, a multi-objective decision-making method is used to generate a composite operation instruction set that includes voltage adjustment, current limiting, scheduling switching, and alarm threshold resetting, forming a real-time optimization scheme for the cable operation control parameters.

[0051] Compared with the prior art, the beneficial effects of the present invention are:

[0052] A pre-trained anomaly representation learning network is used to conduct a deep coupling analysis of the comprehensive operating conditions of cables and the measured values ​​of insulation material loss. The state metric layer within the network calculates a dynamic quantitative index reflecting the correlation strength between different operating conditions and insulation loss through a nonlinear mapping relationship—the insulation loss correlation state anomaly weight spectrum. The core of this process lies in uniformly mapping multi-dimensional surface monitoring data to the evaluation dimension of insulation material aging, a core performance indicator, thus transforming the process from discrete parameter alarms to insulation state correlation assessment. The insulation loss correlation state anomaly weight spectrum can clearly quantify the influence of specific temperature distribution patterns or discharge activity characteristics on the current insulation loss state, making the state assessment result no longer an isolated judgment, but a risk distribution map with clear physical direction.

[0053] Guided by the anomaly weight spectrum associated with insulation loss, the system performs a dual-source fusion anomaly trajectory backtracking analysis on cable sheath temperature scanning data and transient waveforms of discharge phenomena. This analysis process does not treat the two types of data independently, but rather uses the associated weights as clues to simultaneously and inversely deduce the evolution paths of thermal gradient anomalies and discharge pulse events in both time and space dimensions. This method effectively overcomes the ambiguity of single signal source localization, accurately determining the physical spatial coordinates that trigger the apparent anomaly through the spatiotemporal coupling characteristics of thermal and electrical signals on the anomaly trajectory. By analyzing the signal pattern combinations corresponding to the backtracking trajectory at these coordinate points, the system can distinguish the intrinsic physical causes of the anomaly, thus achieving simultaneous physical localization of the anomaly origin and identification of the defect category. Attached Figure Description

[0054] Figure 1 This is a timing diagram of the multi-state low-power sensing system for power distribution cables described in this invention.

[0055] Figure 2 A flowchart generated for the cable surface temperature change pattern;

[0056] Figure 3 A flowchart for generating a discharge activity distribution map;

[0057] Figure 4 A trend analysis diagram of coupling characteristics for cable condition monitoring;

[0058] Figure 5 A rating chart for evaluating cable fault intervention measures. Detailed Implementation

[0059] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0060] Please see Figure 1 This invention provides a multi-state low-power sensing system for power distribution cables. The system includes an information acquisition module, a condition generation module, a weight learning module, a source tracing and positioning module, and a parameter optimization module. The information acquisition module collects raw monitoring information such as cable sheath temperature scanning data, transient waveforms of discharge phenomena, and insulation material loss values ​​through distributed low-power sensors. The sensors employ a low-power design, share signal acquisition and processing circuits through multi-sensor parameter module multiplexing technology, and upload data to a gateway via LoRa wireless communication. The condition generation module performs three-dimensional thermal field gradient calculation on the cable sheath temperature scanning data to generate a cable surface temperature change pattern. Simultaneously, it performs pulse event deconstruction analysis on the transient waveforms of discharge phenomena to generate a discharge activity distribution map. The two are combined to form a comprehensive cable condition characteristic. The weight learning module maps the comprehensive condition characteristic and insulation material loss values ​​to an insulation loss-related state anomaly weight spectrum through an anomaly representation learning network. The source tracing and positioning module performs dual-source anomaly trajectory backtracking analysis on the cable sheath temperature scanning data and transient waveforms of discharge phenomena based on this insulation loss-related state anomaly weight spectrum to locate the physical origin of the anomaly and the defect category. The parameter optimization module generates a real-time optimization scheme for cable operation control parameters based on the positioning results, and realizes closed-loop control through the cable management execution terminal.

[0061] In one embodiment of the present invention, see [reference] Figure 2 A fine-grained temperature sampling point grid was established within the physical surface coordinate system of the cable, with the grid node spacing set to 5–10 cm based on the cable diameter and thermal field resolution requirements. Spatial neighborhood heat flow extrapolation was performed on the temperature readings of each grid node, and the temperature gradient between adjacent nodes was calculated using the finite element method to derive the local heat flow vector for each node. The local heat flow vectors of all nodes were aggregated to form an original heat flow field map reflecting the overall heat flow direction of the cable. The cable structural geometric model and material thermal property parameter library were introduced to perform a field strength correction operation based on the physical structure on the original heat flow field map. The corrected heat flow field map better reflects the actual heat dissipation characteristics of the cable. A sliding window feature analysis was performed on the structurally calibrated heat flow field map in the time dimension, with a window length of 10 minutes and a sliding step of 1 minute. Principal component analysis was used to capture the temporal evolution of the heat flow pattern and output the cable surface temperature change pattern.

[0062] In practical implementation, the spacing between nodes in the fine temperature sampling point grid established in the cable physical surface coordinate system is set to 8 cm. The grid covers the entire monitoring area of ​​the cable sheath. In practice, the temperature reading of each grid node is collected by a low-power temperature sensor at a frequency of once per second. When performing spatial neighborhood heat flow extrapolation on the temperature reading of each point within the temperature sampling point grid, discrete gradient calculation based on the central difference method is used to calculate the local heat flow vector of each point. The formula for calculating the local heat flow vector is as follows:

[0063]

[0064] in: Represents coordinate points The local heat flow vector at that location, This indicates the thermal conductivity of the cable sheath material. Represents coordinate points The temperature gradient vector at each point. The local heat flow vectors corresponding to all temperature sampling points are aggregated to form an original heat flow field map reflecting the overall heat flow direction of the cable. The original heat flow field map is stored in vector field form, where the vector direction represents the heat flow direction and the vector magnitude represents the heat flow intensity.

[0065] In some embodiments, a cable structure geometric model and a material thermal property parameter library are introduced. The cable structure geometric model includes the insulation layer thickness, the metal shielding layer outline, and the outer sheath dimensions. The material thermal property parameter library includes the thermal conductivity of polyethylene insulation material, the heat capacity of copper conductor, and the environmental convective heat transfer coefficient. When performing a field strength correction operation based on the physical structure on the original heat flow field spectrum, the direction of the heat flow vector at the structural boundary is adjusted according to the cable structure geometric model, and the modulus of the heat flow vector is calibrated according to the material thermal property parameter library to generate a structure-calibrated heat flow field spectrum. The specific implementation of the cable structure geometric model includes a refined three-dimensional model of the cable's physical structure. The insulation layer thickness defines the precise radial dimensions of the insulation material, the metal shielding layer profile describes the spatial layout of the shielding layer through geometric parameters, and the outer sheath dimensions specify the outer diameter, thickness, and surface characteristics of the protective sheath. These geometric parameters are constructed based on the actual cable design drawings and installation data and integrated into the system's coordinate system. They are used to accurately adjust the propagation direction of the heat flow vector at the boundary between the insulation layer and the metal shielding layer during the physical structure-based field strength correction operation, and to calibrate the distribution modulus of the heat flow on the cable surface according to the outer sheath dimensions, thereby ensuring that the generated heat flow field spectrum truly reflects the cable's heat dissipation characteristics. In some embodiments, a comparison between the structure-calibrated heat flow field spectrum and the original heat flow field spectrum shows that the heat flow vector modulus in the metal shielding layer region increases by 15% after correction, and the angle between the heat flow vector direction in the insulation layer region and the cable axis decreases by 10 degrees after correction.

[0066] In practical implementation, a sliding window feature analysis is performed on the structure-calibrated heat flow field spectrum over time. The sliding window length is set to 10 minutes, and the sliding step size is set to 1 minute. Principal component features of the structure-calibrated heat flow field spectrum are extracted within each sliding window. These principal component features include the variance contribution rate of the first principal component and the average deflection angle of the heat flow vector, capturing the temporal evolution of the heat flow pattern and outputting the cable surface temperature change pattern. It can be understood that the cable surface temperature change pattern is encoded in time series form, with each time point in the sequence corresponding to a set of principal component feature values. Optionally, the grid node spacing of the temperature sampling points can be configured to range from 5 cm to 15 cm. A grid node spacing of 5 cm increases the number of temperature sampling points and improves the accuracy of local heat flow vector calculation; a grid node spacing of 15 cm reduces the number of temperature sampling points and decreases computational resource consumption.

[0067] In one embodiment of the present invention, see [reference] Figure 3 Multiple independent discharge pulse event segments were separated from the transient waveform of the discharge phenomenon. Threshold detection and zero-crossing point recognition algorithms were used to extract the start and end points of each pulse. For each discharge pulse event segment, waveform morphological decomposition was performed at multiple time scales, and wavelet transform was used to extract the morphological component sets of the pulse event at different scales. Spectral energy distribution analysis was performed on each morphological component set, and the temporal morphological features were mapped to frequency domain energy distribution features using Fast Fourier Transform to form the frequency domain feature vector of the pulse event. A self-organizing feature mapping network was used to cluster the frequency domain feature vectors of all discharge pulse event segments. The number of nodes in the network input layer was equal to the dimension of the feature vector, and the competition layer was the output layer structure of the self-organizing feature mapping network, set to a 3×3 topology, to identify typical discharge activity mode categories. The frequency, average intensity, and sensor location distribution of various discharge modes within a 24-hour monitoring period were statistically analyzed, and the discharge activity distribution map was encoded and output in the form of a three-dimensional map. A time-space synchronization index between the cable surface temperature change mode and the discharge activity distribution map was established, with an indexing accuracy of milliseconds. An association matrix is ​​constructed based on a synchronization index, where matrix elements represent the correlation strength between temperature field changes and discharge activity at spatiotemporal nodes. A cross-domain feature extractor guided by the association matrix extracts temperature feature subsets related to discharge activity from temperature change patterns, and simultaneously extracts discharge feature subsets related to temperature field changes from the discharge activity distribution map. These two subsets are input into a feature fusion engine, which performs deep integration through iterative cross-updates, ultimately outputting comprehensive cable operating condition features.

[0068] In practice, multiple independent discharge pulse event segments are separated from the transient waveform of the discharge phenomenon. The transient waveform of the discharge phenomenon is acquired by a high-frequency current transformer at a sampling rate of 100 MHz. The separation process adopts a combination of fixed threshold and adaptive threshold. The signal range with a voltage amplitude exceeding 5 times the standard deviation of the background noise is identified as an independent discharge pulse event segment. In a monitoring record lasting one minute, 120 independent discharge pulse event segments are separated. For each discharge pulse event segment, waveform morphological decomposition is performed at multiple time scales. Db4 wavelet is used for three-level decomposition to extract the morphological component sets of the pulse event at the microsecond, ten-microsecond and hundred-microsecond scales. After decomposition, each independent discharge pulse event segment yields a sequence of morphological components at three different time scales. For each morphological component set, spectral energy distribution analysis is performed. A Fast Fourier Transform is applied to the morphological component sequence at each scale to calculate the proportion of energy in each frequency band relative to the total energy. This maps the time-domain morphological features to the frequency-domain energy distribution features, forming a frequency-domain feature vector for the pulse event. This frequency-domain feature vector is a nine-dimensional vector containing the energy proportions of the low, mid, and high-frequency bands at each of the three time scales. The formula for calculating the energy proportion in the frequency-domain feature vector is as follows:

[0069]

[0070] in: Indicates the first The first time scale Energy percentage of each frequency band Indicates the first The first time scale Energy value of each frequency band The values ​​correspond to three time scales: microseconds, ten microseconds, and one hundred microseconds. The values ​​correspond to the low, medium, and high frequency bands.

[0071] In some embodiments, a self-organizing feature mapping network is used to cluster and summarize the frequency domain feature vectors corresponding to all discharge pulse event segments. The input layer of the self-organizing feature mapping network is set to nine nodes corresponding to nine dimensions of the frequency domain feature vectors, and the competition layer is set to a three-row, three-column two-dimensional grid structure. After training, typical discharge activity pattern categories with common characteristics are identified. The identified typical discharge activity pattern categories include three types. In some embodiments, the frequency, intensity, and spatial distribution of each type of typical discharge activity pattern within a preset monitoring period are statistically analyzed. The preset monitoring period is set to 24 hours. The first discharge activity pattern occurs 350 times with an average intensity of 15 mV, the second discharge activity pattern occurs 80 times with an average intensity of 45 mV, and the third discharge activity pattern occurs 20 times with an average intensity of 80 mV. The spatial distribution information is encoded in the form of a two-dimensional chromatogram, and a discharge activity distribution map is output. The horizontal axis of the discharge activity distribution map represents time, the vertical axis represents the sensor number, and the chromatogram represents the discharge pattern category.

[0072] In practical implementation, a time-space synchronization index is established between the cable surface temperature change pattern and the discharge activity distribution map. Time synchronization is based on the master clock of the monitoring system, aligning the timestamps of the two types of data with a millisecond accuracy. Spatial synchronization is established based on the mapping relationship between the physical installation positions of the temperature sensor and the discharge sensor. An association matrix is ​​constructed based on the time-space synchronization index. The rows of the association matrix correspond to the time-space grid of the temperature change pattern, and the columns correspond to the time-space grid of the discharge activity distribution map. The matrix element values ​​are the Pearson correlation coefficients between the corresponding grids. Using a cross-domain feature extractor guided by the association matrix, a subset of temperature features related to discharge activity is extracted from the cable surface temperature change pattern. The extraction rule is to select the temperature change pattern features corresponding to the rows in the association matrix with an absolute correlation coefficient greater than 0.7. Simultaneously, a subset of discharge features related to temperature field changes is extracted from the discharge activity distribution map. The extraction rule is to select the discharge activity distribution map features corresponding to the columns in the association matrix with an absolute correlation coefficient greater than 0.7. The temperature feature subset and the discharge feature subset are input into the feature fusion engine. The feature fusion engine performs iterative cross-updates and deep integration of the two subsets by simulating the physical interaction process. In each iteration, each element of the temperature feature subset is updated according to its association weight with the discharge feature subset, and each element of the discharge feature subset is updated according to its association weight with the temperature feature subset. After five iterations, the feature fusion engine outputs a fixed-dimensional fusion feature vector, which finally forms the comprehensive cable operating condition feature that characterizes the overall operating status of the cable.

[0073] It is understandable that establishing a time-space synchronization index relies on all sensors having a unified and high-precision time source. It is also understandable that the dimension of the correlation matrix is ​​determined by the number of temperature sampling points, the number of discharge sensors, and the number of common time segments. Optionally, the separation threshold for discharge pulse event segments can be configured to be 3 to 8 times the standard deviation of the background noise. Setting the separation threshold to 3 times allows for the capture of more weak discharge pulses, while setting it to 8 times effectively suppresses interfering pulses. Optionally, the structure of the self-organizing feature map network competition layer can be configured as a two-dimensional grid or a one-dimensional chain structure. A two-dimensional grid structure reveals the topological relationship of discharge modes in the two-dimensional feature space, while a one-dimensional chain structure reduces computational complexity.

[0074] In one embodiment of the present invention, the anomaly representation learning network includes a feature encoding layer, an association learning layer, and a state measurement layer. The feature encoding layer employs a multilayer perceptron structure. The first hidden layer performs a linear transformation on the input cable comprehensive operating condition features and processes them using a ReLU activation function to generate a primary abstract feature representation. The second hidden layer extracts local feature combinations through a convolutional kernel sliding window to form a mid-level abstract feature representation. The output layer integrates global features through a fully connected network, outputting a high-dimensional abstract feature vector with a dimension five times that of the original input. The association learning layer constructs a nonlinear mapping relationship between the high-dimensional abstract feature vector and the insulation material loss value. The mapping function adopts a hyperbolic tangent structure, and the backpropagation algorithm dynamically learns the sensitivity of each feature dimension to the insulation material loss value. The state measurement layer performs weighted recombination of the high-dimensional abstract feature vector based on the contribution sensitivity, with a sensitivity threshold set to 0.8. After recombination, an insulation loss associated state anomaly weight spectrum is generated, where the feature dimension corresponding to the peak of the spectrum indicates an anomaly state highly correlated with insulation loss.

[0075] In its implementation, the anomaly representation learning network comprises a feature encoding layer, an association learning layer, and a state measurement layer. The feature encoding layer receives a 50-dimensional input vector representing the overall cable condition features. The process of converting these features into a set of high-dimensional abstract feature vectors is implemented using a feature encoder with a multilayer perceptron structure. In the first hidden layer of the feature encoder, linear transformation and nonlinear activation are performed on the input features. The linear transformation uses a weight matrix of 50 x 75 dimensions and a bias vector of 75 dimensions. The nonlinear activation function is the ReLU function, generating a primary feature vector with 75 dimensions. The abstract feature representation is passed to the second hidden layer. The second hidden layer extracts local feature combinations through a convolutional kernel sliding window operation. The convolutional kernel size is five, the sliding stride is two, and the number of output channels is fifty, forming a mid-level abstract feature representation with fifty dimensions. The mid-level abstract feature representation is input to the output layer of the feature encoder. The output layer performs global feature integration and dimensional expansion through a fully connected network. The weight matrix of the fully connected network has a dimension of fifty times two hundred and fifty, and finally outputs a set of high-dimensional abstract feature vectors with a dimension of two hundred and fifty. The dimension of the high-dimensional abstract feature vectors is much higher than the fifty dimensions of the original input.

[0076] In some embodiments, the association learning layer constructs a nonlinear mapping relationship between high-dimensional abstract feature vectors and insulation material loss values. The nonlinear mapping function adopts a composite form of hyperbolic tangent function. During the mapping process, the contribution sensitivity of each feature dimension to the insulation material loss value is dynamically learned through the error backpropagation algorithm. The contribution sensitivity is a value between 0 and 1, representing the degree of influence of changes in each feature dimension on the predicted change in insulation material loss value. In one training iteration, the contribution sensitivity is updated after inputting ten batches of sample data. In some embodiments, the learning process of contribution sensitivity is carried out simultaneously with the adjustment of network weights. The formula for calculating contribution sensitivity is:

[0077]

[0078] in: Indicates the first Sensitivity to the contribution of each feature dimension This indicates the number of samples in a batch. Indicates the first The predicted loss value for each sample. Represents the first in a high-dimensional abstract eigenvector Each dimension of feature value, This represents the partial derivative of the loss with respect to a specific feature dimension.

[0079] In practical implementation, the state measurement layer performs weighted recombination of each dimension in the high-dimensional abstract feature vector based on contribution sensitivity. The weight coefficients of the weighted groups are obtained by normalizing the contribution sensitivity using the Softmax function, generating an insulation loss-related state anomaly weight spectrum. The insulation loss-related state anomaly weight spectrum is a spectrum with the feature dimension index as the horizontal axis and the weighted recombination feature intensity as the vertical axis. The feature dimension corresponding to the peak of the spectrum indicates the anomaly state that is highly correlated with insulation loss. The contribution sensitivity threshold corresponding to the anomaly state that is highly correlated with insulation loss is set to 0.8. Feature dimensions exceeding this threshold will show obvious spectral peaks in the insulation loss-related state anomaly weight spectrum.

[0080] It is understandable that expanding the dimensionality of high-dimensional abstract feature vectors helps to separate nonlinear patterns related to insulation loss in a higher-dimensional space. It is also understandable that dynamic learning of contribution sensitivity enables the network to adaptively focus on the most relevant feature dimensions under different operating states. Optionally, the output dimension of the first hidden layer in the feature encoder can be configured to be 1.2 to 2 times the original input dimension. A 1.2x output dimension results in lower computational cost, while a 2x output dimension provides stronger feature abstraction capabilities. Optionally, the update frequency of contribution sensitivity can be configured to be once every five batches or once per batch. Updating once every five batches is more computationally efficient, while updating once per batch provides a more timely response to contribution sensitivity.

[0081] In one embodiment of the present invention, guided by the anomaly weight spectrum of insulation loss associated state, a reverse heat conduction algorithm is used to trace the starting point and diffusion path of temperature anomaly evolution in the temperature field composed of cable sheath temperature scanning data, forming an anomaly trajectory of the temperature field. Simultaneously, guided by the anomaly weight spectrum of insulation loss associated state, time inversion technology is used to reversely locate the initial moment and evolution process of discharge mode distortion in the transient waveform sequence of discharge phenomena, forming an anomaly trajectory of discharge activity. Spatiotemporal parameterization equations for the temperature field anomaly trajectory and the discharge activity anomaly trajectory are defined. A coupled objective function is constructed with the goal of minimizing the spatiotemporal distance between the two trajectories; the function expression includes Euclidean distance and time difference weighted terms. The gradient descent algorithm is used to solve the coupled objective function to obtain the set of coupled parameters that makes the two trajectories closest. The intersection point of the trajectories is calculated based on the set of coupled parameters, serving as the physical origin location of the anomaly state. Temperature gradient change feature sequences near the intersection point are extracted from the temperature field anomaly trajectory, and discharge pulse intensity and repetition rate feature sequences near the intersection point are extracted from the discharge activity anomaly trajectory. The two feature sequences are synchronously compared to form a set of coupled feature pairs. The set of coupling features is input into a pre-trained defect classifier. The coupling patterns and defect category mappings stored internally by the classifier include categories such as partial discharge-overheating combined defects. The output is the determination result of the intrinsic defect category.

[0082] In specific implementation, the insulation loss associated state anomaly weight spectrum is used as a guide. The third, seventh, and twelfth feature dimensions corresponding to the spectral peaks in the insulation loss associated state anomaly weight spectrum are identified as highly correlated dimensions. In the temperature field formed by the cable sheath temperature scanning data, the reverse heat conduction algorithm is used to trace the starting point and diffusion path of the temperature anomaly evolution. The reverse heat conduction algorithm is based on the gradient descent method for iterative solution to form the temperature field anomaly trajectory. The temperature field anomaly trajectory is recorded as a series of three-dimensional coordinate points within five minutes tracing back from the time of the anomaly. Simultaneously, the insulation loss associated state anomaly weight spectrum is used as a guide. In the discharge activity sequence contained in the transient waveform of the discharge phenomenon, the dynamic time warping algorithm is used to reversely locate the initial moment and evolution process of the discharge mode distortion to form the discharge activity anomaly trajectory. The discharge activity anomaly trajectory is recorded as a series of discharge pulse timestamps and amplitude sequences within one hundred milliseconds tracing back from the time of the abnormal discharge pulse.

[0083] In some embodiments, a spatiotemporal coupling analysis model is established between the abnormal temperature field trajectory and the abnormal discharge activity trajectory. A spatiotemporal parameterized equation is defined to describe the abnormal temperature field trajectory. The equation is expressed using normalized parameters. As variables, output spatial coordinates and timestamp Define a spatiotemporal parameterized equation to describe the anomalous trajectory of discharge activity, with the equation using normalized parameters. As variables, output sensor position coordinates and timestamp A coupled objective function is constructed with the goal of minimizing the normalized spatiotemporal distance between the two trajectories. The expression of the coupled objective function is as follows:

[0084]

[0085] in: Represents the normalized spatiotemporal distance. and Let these be the dimensionless weighting coefficients of the spatial and temporal terms, respectively, and satisfy the following conditions: , Indicates the abnormal trajectory of the temperature field in the parameters The three-dimensional spatial coordinates below Indicates abnormal discharge activity trajectory in parameters The three-dimensional spatial coordinates below Indicates the maximum spatial range of the cable monitoring area. Indicates the abnormal trajectory of the temperature field in the parameters The timestamp below, Indicates abnormal discharge activity trajectory in parameters The timestamp below, This indicates the maximum time span of the analysis time window. Let represent the Euclidean norm. The coupled objective function is solved using the gradient descent numerical optimization algorithm. After initializing the parameters, the optimal value is obtained after two hundred iterations. Minimize the set of coupling parameters The coordinates of the intersection point of the abnormal temperature field trajectory and the abnormal discharge activity trajectory were calculated based on the set of coupling parameters. and time The intersection point is the physical coordinate point output by the spatiotemporal coupling analysis model.

[0086] In the specific implementation, the coupling mode between the abnormal temperature field trajectory and the abnormal discharge activity trajectory at the physical origin location is analyzed. The temperature gradient change feature sequence near the intersection point is extracted from the abnormal temperature field trajectory. The extraction time window is 30 seconds before and after the intersection point. The temperature gradient change feature sequence includes the gradient amplitude and direction of ten sampling points. The discharge pulse intensity and repetition rate feature sequence near the intersection point is extracted from the abnormal discharge activity trajectory. The extraction time window is 100 milliseconds before and after the intersection point. The discharge pulse intensity and repetition rate feature sequence includes the pulse peak value and pulse time interval. The temperature gradient change feature sequence and the discharge pulse intensity and repetition rate feature sequence are synchronously compared. The two types of feature sequences are resampled and aligned according to a unified time axis to form a set of coupled feature pairs. See Table 1 for partial data of the set of coupled feature pairs.

[0087] Table 1: Coupling Feature Pairs Data Table

[0088]

[0089] The defect classifier, trained by inputting a set of coupled feature pairs, is a classification model built on support vector machines. The classifier internally stores a mapping relationship between coupling patterns and defect categories, including three defect categories: overheating caused by partial discharge, partial discharge induced by overheating, and discharge associated with mechanical damage. The classifier outputs the determination result of the intrinsic defect category based on this mapping relationship. The weight coefficients in the normalized spatiotemporal distance formula can be understood as... and The value of determines the relative importance of spatial proximity and temporal synchronization in coupling judgment. It can be understood that the construction of the coupling feature set depends on accurate timestamp synchronization and data interpolation. Optionally, the minimization problem of the coupling objective function can be solved using either the quasi-Newton method or the particle swarm optimization algorithm. The quasi-Newton method converges quickly when the parameter space is smooth, while the particle swarm optimization algorithm is more suitable for non-convex optimization problems.

[0090] See Figure 4This is a chart analyzing the coupling characteristics of cable condition monitoring, a professional industrial data visualization chart. This type of chart is commonly used for analyzing abnormal conditions in power distribution cables: the temperature gradient is positively correlated with the discharge pulse peak value and negatively correlated with the pulse interval, a typical characteristic of "partial discharge-overheating combined defect," which can help locate the origin of cable anomalies. It intuitively presents the linkage between "temperature gradient and discharge parameters," and through the coupling characteristics of "temperature increase + pulse peak value increase + pulse interval shortening," it quickly identifies the abnormal evolution trend of the cable, avoiding misjudgment based on a single indicator. The quantitative trend of the data can serve as a direct basis for real-time optimization schemes such as "voltage adjustment and current limiting," improving the accuracy of cable operation control.

[0091] In one embodiment of the invention, the physical origin of the abnormal state is analyzed, and the specific affected cable line section is determined through geographic information system mapping. Combining the inherent defect category, the operation and maintenance knowledge base is queried to obtain a set of targeted standard intervention measures. The current load data, environmental parameters, and historical operation records of the specific cable section are integrated to conduct an adaptability and feasibility assessment of the measures in the standard intervention measure set. The assessment indicators include the implementation delay time and resource cost. Based on the assessment results, a multi-objective decision-making method is used to generate a composite operation instruction set. The instruction set includes voltage adjustment, current limiting, scheduling switching, and alarm threshold resetting, forming a real-time optimization scheme for cable operation control parameters.

[0092] In practice, the physical origin of the abnormal state is analyzed. This physical origin is output as three-dimensional geographic coordinates (116.5 degrees East longitude, 39.9 degrees North latitude, depth -1.2 meters) by the source tracing and positioning module. The specific affected cable section is determined through geographic information system mapping. This specific cable section is identified as "between joints 15 and 16 in section A of the cable trench," and its length is fifteen meters. Based on the inherent defect category, which is "combined partial discharge and overheating defect," the operation and maintenance knowledge base is queried. This knowledge base contains a defect-response mapping table constructed from historical operation and maintenance records and standard procedures. A set of standard intervention measures for the "combined partial discharge and overheating defect" is obtained, including four measures: reducing load current, improving ventilation and heat dissipation, arranging power outage maintenance, and suppressing partial discharge.

[0093] In some embodiments, current load data, environmental parameters, and historical operating records of a specific cable line section are integrated. Current load data includes a real-time current of 800 amps and a real-time voltage of 10 kV. Environmental parameters include an ambient temperature of 35 degrees Celsius and an ambient humidity of 70%. Historical operating records include three similar abnormal alarms occurring in the section within the past thirty days. Adaptability and feasibility assessments are performed on each measure in the standard intervention set. Adaptability assessment is based on the degree of matching between the measure and current operating conditions, while feasibility assessment is based on the resource and time costs required to implement the measure. The assessment uses a percentage-based scoring system. Measures to reduce load current receive an adaptability score of 85 and a feasibility score of 90; measures to improve ventilation and heat dissipation receive an adaptability score of 70 and a feasibility score of 65; measures to schedule power outages for maintenance receive an adaptability score of 95 and a feasibility score of 50; and measures to suppress partial discharge receive an adaptability score of 80 and a feasibility score of 75. In some embodiments, a comprehensive evaluation value is calculated for each measure during the evaluation process. The formula for calculating the comprehensive evaluation value is:

[0094]

[0095] in: Indicates the first The overall evaluation value of the measures and The weighting coefficients representing adaptability and feasibility are as follows: , Indicates the first The adaptability score of the measures Indicates the first Feasibility score for each measure. Weighting coefficient. Set to 0.6, weighting coefficient With the value set to 0.4, the comprehensive evaluation score for measures to reduce load current is 87 points, the comprehensive evaluation score for measures to enhance ventilation and heat dissipation is 68 points, the comprehensive evaluation score for measures to arrange power outages for maintenance is 77 points, and the comprehensive evaluation score for measures to suppress partial discharge is 78 points.

[0096] In practical implementation, based on the evaluation results, a multi-objective decision-making method is adopted to generate a composite operation instruction set including voltage adjustment, current limiting, dispatch switching, and alarm threshold resetting. The multi-objective decision-making method employs the analytic hierarchy process (AHP), with decision objectives including improved safety, guaranteed power supply continuity, and cost control. The specific content of the generated composite operation instruction set is as follows: the voltage adjustment instruction reduces the operating voltage from 10 kV to 9.5 kV; the current limiting instruction limits the maximum allowable current from 1000 amps to 900 amps; the dispatch switching instruction transfers 30% of the load to standby cable line B; and the alarm threshold resetting instruction adjusts the overheat alarm threshold from 70 degrees Celsius to 65 degrees Celsius and the partial discharge alarm threshold from 20 picocubes to 15 picocubes. This forms a real-time optimization scheme for cable operation control parameters, which is delivered to the cable management execution terminal in a structured data message format. It is understandable that the defect-response mapping table in the operation and maintenance knowledge base needs to be updated regularly to reflect the latest operation and maintenance experience. Optionally, the instructions in the composite operation instruction set can be executed independently or in combination in a preset order. When executed independently, they offer high flexibility, while when executed in combination in a preset order, they ensure logical consistency between operations.

[0097] See Figure 5 This is an evaluation and scoring chart for cable fault intervention measures, a professional analytical tool used in power distribution operation and maintenance decision-making. This chart is used to quantitatively screen the optimal intervention measures for cable faults. Through a dual-dimensional evaluation of "adaptability + feasibility," it quickly identifies highly adaptable and easily implemented priority measures such as "reducing load current," providing a decision-making basis for optimizing cable operation control parameters. The multi-dimensional scoring of "adaptability + feasibility + comprehensive evaluation" intuitively distinguishes the superiority and inferiority levels of measures, quickly identifying "reducing load current" as the current optimal measure, avoiding the subjectivity and inefficiency of manual experience-based judgment. It clarifies the contradiction that "arranging power outage maintenance," while highly adaptable, has low feasibility, avoiding blindly investing in high-cost, low-implementation measures and helping operation and maintenance teams optimize resource allocation.

[0098] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0099] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A multi-state low-power sensing system for power distribution cables, characterized in that, The system includes: The information acquisition module collects three types of raw monitoring information that reflect the real-time operating status of the cable. These three types of raw monitoring information include cable sheath temperature scanning data, transient waveforms of discharge phenomena, and insulation material loss values. The operating condition generation module performs three-dimensional thermal field gradient calculation on the cable sheath temperature scanning data, outputs the cable surface temperature change pattern, and performs pulse event deconstruction analysis on the transient waveform of the discharge phenomenon, outputs the discharge activity distribution map. Combining the cable surface temperature change pattern and the discharge activity distribution map, the module generates comprehensive cable operating condition characteristics. The weight learning module inputs the comprehensive working condition characteristics of the cable and the insulation material loss value into the trained anomaly representation learning network, and outputs the insulation loss associated state anomaly weight spectrum generated by the state metric layer inside the anomaly representation learning network. The source tracing and positioning module, based on the abnormal weight spectrum of the insulation loss associated state, performs dual-source abnormal trajectory back-tracing analysis on the cable sheath temperature scanning data and the transient waveform of the discharge phenomenon to locate the physical origin of the abnormal state and identify its inherent defect category. The parameter tuning module generates a real-time tuning scheme for the cable operation control parameters based on the physical origin location and inherent defect category of the abnormal state, and delivers the real-time tuning scheme to the cable management execution terminal.

2. The multi-state low-power sensing system for power distribution cables according to claim 1, characterized in that, The process of performing three-dimensional thermal field gradient calculation on the cable sheath temperature scanning data and outputting the cable surface temperature change pattern includes the following steps: Establish a fine grid of temperature sampling points in the physical surface coordinate system of the cable; A spatial neighborhood heat flow extrapolation operation is performed on the temperature readings of each point within the temperature sampling point grid to calculate the local heat flow vector of each point; the local heat flow vectors corresponding to all temperature sampling points are aggregated to form an original heat flow field map reflecting the overall heat flow direction of the cable; By introducing a cable structure geometric model and a material thermal property parameter library, a field strength correction operation based on physical structure is performed on the original thermal flow field spectrum to generate a structure-calibrated thermal flow field spectrum. A sliding window feature analysis in the time dimension is performed on the structure-calibrated thermal flow field spectrum to capture the temporal evolution of the thermal flow pattern and output the temperature change pattern of the cable surface.

3. The multi-state low-power sensing system for power distribution cables according to claim 1, characterized in that, The process of performing pulse event deconstruction analysis on the transient waveform of the discharge phenomenon and outputting a discharge activity distribution map includes the following specific steps: Multiple independent discharge pulse event segments are separated from the transient waveform of the discharge phenomenon; for each discharge pulse event segment, waveform morphological decomposition is performed at multiple time scales to extract the set of morphological components of the discharge pulse event at different scales. Perform spectral energy distribution analysis on each morphological component set to map the temporal morphological features to the frequency domain energy distribution features, forming the frequency domain feature vector of the discharge pulse event; A self-organizing feature mapping network is used to cluster and summarize the frequency domain feature vectors corresponding to all discharge pulse event segments to identify typical discharge activity pattern categories with common characteristics. The frequency, intensity, and spatial distribution of various typical discharge activity patterns within a preset monitoring period are statistically analyzed, and the discharge activity distribution map is encoded and output in the form of a graph.

4. The multi-state low-power sensing system for power distribution cables according to claim 1, characterized in that, The comprehensive operating condition characteristics of the cable are generated by combining the cable surface temperature change pattern and the discharge activity distribution map, which is achieved through the following interactive process: Establish a time-space synchronization index between cable surface temperature change patterns and discharge activity distribution maps; Based on the aforementioned time-space synchronization index, a correlation matrix is ​​constructed to characterize the interaction between temperature field changes and discharge activity; Using the cross-domain feature extractor guided by the correlation matrix, a subset of temperature features related to discharge activity is extracted from the temperature change pattern of the cable surface, and a subset of discharge features related to temperature field changes is extracted from the discharge activity distribution map. The temperature feature subset and the discharge feature subset are input into the feature fusion engine. The feature fusion engine performs iterative cross-updates and deep integration of the two subsets by simulating a physical interaction process, and finally forms a comprehensive cable condition feature that characterizes the overall operating status of the cable.

5. The multi-state low-power sensing system for power distribution cables according to claim 1, characterized in that, The anomaly representation learning network includes: a feature encoding layer, an association learning layer, and a state measurement layer; The feature encoding layer converts the comprehensive cable operating condition features into a set of high-dimensional abstract feature vectors; The association learning layer constructs a nonlinear mapping relationship between the high-dimensional abstract feature vector and the insulation material loss value, and dynamically learns the sensitivity of each feature dimension to the insulation material loss value during the mapping process. The state metric layer reorganizes each dimension of the high-dimensional abstract feature vector according to the contribution sensitivity to generate the insulation loss associated state anomaly weight spectrum, wherein the feature dimension corresponding to the spectral peak of the insulation loss associated state anomaly weight spectrum indicates the anomaly state that is highly associated with insulation loss.

6. The multi-state low-power sensing system for power distribution cables according to claim 5, characterized in that, The feature encoding layer converts the comprehensive operating condition features of the cable into a set of high-dimensional abstract feature vectors, which is achieved through the following processing steps: The comprehensive operating conditions of the cable are input into a feature encoder with a multilayer perceptron structure. In the first hidden layer of the feature encoder, the comprehensive working condition features of the cable are subjected to linear transformation and nonlinear activation processing to generate a primary abstract feature representation. The primary abstract feature representation is passed to the second hidden layer, and local feature combinations are extracted through the convolution kernel sliding window operation to form the intermediate abstract feature representation; The intermediate-level abstract feature representation is input into the output layer of the feature encoder. Global feature integration and dimensional expansion are performed through a fully connected network, and finally a set of high-dimensional abstract feature vectors with dimensions much higher than the original input are output.

7. The multi-state low-power sensing system for power distribution cables according to claim 1, characterized in that, The execution steps for performing dual-source anomaly trajectory backtracking analysis on the cable sheath temperature scanning data and the transient waveform of the discharge phenomenon are as follows: Guided by the insulation loss associated state anomaly weight spectrum, the starting point and diffusion path of temperature anomaly evolution are traced in reverse in the temperature field formed by the cable sheath temperature scanning data to form a temperature field anomaly trajectory. Simultaneously, guided by the insulation loss associated state anomaly weight spectrum, the initial moment and evolution process of discharge mode distortion are located in reverse in the discharge activity sequence contained in the transient waveform of the discharge phenomenon to form a discharge activity anomaly trajectory. A spatiotemporal coupling analysis model is established between the abnormal trajectory of the temperature field and the abnormal trajectory of the discharge activity. The physical coordinate point where the two trajectories intersect in space and time is solved by the spatiotemporal coupling analysis model, and the physical coordinate point is determined as the physical origin of the abnormal state. The coupling pattern between the abnormal temperature field trajectory and the abnormal discharge activity trajectory at the physical origin location is analyzed, and the coupling pattern is matched with a pre-stored defect pattern library to identify the intrinsic defect category.

8. The multi-state low-power sensing system for power distribution cables according to claim 7, characterized in that, The establishment of the spatiotemporal coupling analysis model between the abnormal temperature field trajectory and the abnormal discharge activity trajectory specifically includes: Define a spatiotemporal parameterized equation to describe the abnormal trajectory of the temperature field, and define a spatiotemporal parameterized equation to describe the abnormal trajectory of the discharge activity. A coupled objective function is constructed with the goal of minimizing the spatiotemporal distance between two trajectories. The coupled objective function is solved using a numerical optimization algorithm to obtain the set of coupling parameters that make the two trajectories closest in spatiotemporal space. Based on the set of coupling parameters, the intersection point of the abnormal temperature field trajectory and the abnormal discharge activity trajectory is calculated, and the intersection point is the physical coordinate point output by the spatiotemporal coupling analysis model.

9. The multi-state low-power sensing system for power distribution cables according to claim 7, characterized in that, The analysis of the coupling mode between the abnormal temperature field trajectory and the abnormal discharge activity trajectory at the physical origin location includes: Extract the temperature gradient change feature sequence near the intersection point from the abnormal temperature field trajectory; Extract the discharge pulse intensity and repetition rate feature sequences near the intersection point from the abnormal discharge activity trajectory; synchronously compare the temperature gradient change feature sequence with the discharge pulse intensity and repetition rate feature sequence to form a set of coupled feature pairs; The coupled feature set is input into the trained defect classifier, which outputs the determination result of the intrinsic defect category based on the internally stored mapping relationship between coupling patterns and defect categories.

10. The multi-state low-power sensing system for power distribution cables according to claim 1, characterized in that, The real-time optimization scheme for generating cable operation control parameters includes the following decision-making steps: Analyze the physical origin of the abnormal state to determine the specific affected cable line section; combine the inherent defect category with the operation and maintenance knowledge base to obtain a set of standard intervention measures for the inherent defect category; By integrating the current load data, environmental parameters, and historical operation records of the specific cable line section, an adaptability and feasibility assessment is conducted on each measure in the set of standard intervention measures; Based on the evaluation results, a multi-objective decision-making method is used to generate a composite operation instruction set that includes voltage adjustment, current limiting, scheduling switching, and alarm threshold resetting, forming a real-time optimization scheme for the cable operation control parameters.

Images