Cable state sensing method and apparatus, electronic device, and readable storage medium

Abstract

This invention provides a cable condition sensing method, device, electronic device, and readable storage medium. The method includes: acquiring multi-dimensional operational data during cable operation, including at least temperature data, current data, and insulation resistance; predicting cable anomaly detection results based on a pre-trained classification model and the multi-dimensional operational data; the classification model is obtained through semi-supervised training using a labeled dataset and an unlabeled target domain dataset formed by clustering historical cable operational data and its category labels; and generating cable condition sensing results based on the cable anomaly detection results, which characterize whether the cable has an anomaly, the type of anomaly, and its severity. This method effectively overcomes the problems of low accuracy in early defect identification caused by single data acquisition, strong reliance on manual annotation, and scarcity of anomaly samples in traditional cable monitoring, achieving accurate perception and discrimination of cable anomaly types and severity.

Description

Technical Field

[0001] This invention relates to the field of power equipment condition monitoring and fault diagnosis technology, and in particular to a cable condition sensing method, device, electronic device, and readable storage medium. Background Technology

[0002] With the acceleration of urbanization and the continuous upgrading of the power system, high-voltage cables, as core equipment of the power grid, directly determine the stability and security of power supply, making cable condition monitoring technology increasingly important.

[0003] Current cable condition monitoring technologies have significant shortcomings: 1) Data acquisition relies on a single parameter, making it difficult to comprehensively reflect the cable condition; some solutions require the deployment of multiple sensors at each joint, which is complex to install, prone to errors, and has high maintenance costs; 2) Wireless signals are easily blocked in underground environments, leading to unstable or even interrupted data transmission. Coupled with harsh conditions such as long-term water accumulation and high humidity, this seriously affects the reliability and online rate of the equipment; 3) Data processing lacks the ability to fuse multi-source information, making it difficult to uncover the correlation between parameters; at the same time, due to the scarcity of abnormal samples, the model is prone to bias towards the normal state, resulting in a low early defect identification rate.

[0004] In addition, existing cable condition monitoring technologies mostly only provide alarms and cannot accurately identify the type, severity, and development trend of abnormalities, making it difficult to support the need for accurate condition assessment and prediction. Summary of the Invention

[0005] This invention provides a cable condition sensing method, device, electronic device, and readable storage medium to solve the problem that existing high-voltage cable condition monitoring technologies cannot accurately sense and predict early cable defects due to single data acquisition methods and low anomaly identification accuracy.

[0006] On one hand, the present invention provides a cable condition sensing method, comprising: acquiring multi-dimensional operational data during cable operation, wherein the multi-dimensional operational data includes at least temperature data, current data, and insulation resistance; predicting cable anomaly detection results based on the multi-dimensional operational data using a pre-trained classification model; wherein the classification model is obtained through semi-supervised training using a labeled dataset and an unlabeled target domain dataset formed by clustering historical cable operational data and category labels; and generating cable condition sensing results based on the cable anomaly detection results, wherein the cable condition sensing results characterize whether the cable has an anomaly, the type of anomaly, and the severity of the anomaly.

[0007] Furthermore, the classification model is obtained through semi-supervised training using a labeled dataset and an unlabeled target domain dataset formed by the historical cable operation data and the category labels generated by its clustering. Specifically, this includes: constructing a structured data graph based on the historical cable operation data; clustering the historical operation data records in the structured data graph and assigning category labels to each historical operation data record according to the clustering results to form the labeled dataset; performing category distribution balancing processing on the labeled dataset to alleviate the imbalance in the number of samples between normal and abnormal states; and using the labeled dataset and the unlabeled target domain dataset after category distribution balancing as training data to train the classification model through semi-supervised learning.

[0008] Furthermore, the step of using the labeled dataset after class distribution balancing and the unlabeled target domain dataset together as training data to train the classification model through semi-supervised learning includes: clustering the unlabeled target domain dataset to obtain multiple data clusters; for each sample in the unlabeled target domain dataset, calculating its average similarity with other samples in the same cluster as intra-cluster similarity, and simultaneously calculating its average similarity with samples in the nearest dissimilar cluster as inter-cluster similarity; determining the clustering evaluation value of the sample based on the intra-cluster similarity and the inter-cluster similarity; encoding the sample using a variational autoencoder to generate a predicted label, and correcting the predicted label according to the clustering evaluation value to obtain a state label; during the semi-supervised training process, using the state label as a pseudo-label, participating in model training together with the labeled dataset to optimize the classification model's ability to discriminate target domain data.

[0009] Furthermore, the acquisition of multi-dimensional operational data during cable operation includes: collecting temperature data through fiber optic temperature sensors deployed at key locations on the cable joint and the cable body; collecting current data through a Hall current sensor integrated in the cable online monitoring device; and acquiring the insulation resistance through an online insulation monitoring device based on micro-current detection technology, wherein the micro-current detection technology includes extracting leakage current through a coupling capacitor and converting it into an insulation resistance value.

[0010] Further, the acquisition of multi-dimensional operational data during cable operation includes: performing feature filtering on the multi-dimensional operational data, removing redundant features whose correlation with cable state discrimination is lower than a preset threshold, and obtaining a filtered feature set; performing time-frequency domain decomposition on the filtered feature set using wavelet transform to obtain processed multi-dimensional operational data; correspondingly, the prediction of cable anomaly detection results based on the multi-dimensional operational data using a pre-trained classification model includes: inputting the processed multi-dimensional operational data into the pre-trained classification model to obtain the output cable anomaly detection results.

[0011] Furthermore, the step of generating cable status perception results based on the cable anomaly detection results includes: combining the anomaly type, severity, and historical cable operation data to identify the development pattern of similar anomalies; introducing current environmental parameters to dynamically correct the development pattern to obtain the corrected development pattern; predicting the state evolution trend of the cable within a future preset time window based on the corrected development pattern, and generating fault warning information.

[0012] Secondly, the present invention also provides a cable condition sensing device, comprising: a multi-dimensional operation data acquisition module for acquiring multi-dimensional operation data during cable operation, wherein the multi-dimensional operation data includes at least temperature data, current data, and insulation resistance; a cable anomaly detection result prediction module for predicting cable anomaly detection results based on a pre-trained classification model and the multi-dimensional operation data; wherein the classification model is obtained by semi-supervised training using a labeled dataset and an unlabeled target domain dataset formed by clustering historical cable operation data and category labels; and a cable condition sensing result generation module for generating cable condition sensing results based on the cable anomaly detection results, wherein the cable condition sensing results characterize whether the cable has an anomaly, the type of anomaly, and the severity.

[0013] Thirdly, the present invention also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the cable status sensing method as described above.

[0014] Fourthly, the present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the cable status sensing method as described above.

[0015] Fifthly, the present invention also provides a computer program product, including a computer program that, when executed by a processor, implements the cable status sensing method as described above.

[0016] The cable condition sensing method provided by this invention acquires multi-dimensional operational data during cable operation, including at least temperature, current, and insulation resistance data. Based on a pre-trained classification model, it predicts cable anomaly detection results according to the multi-dimensional operational data. The classification model is obtained through semi-supervised training using labeled datasets and unlabeled target domain datasets formed by clustering historical cable operational data and their generated category labels. Based on the cable anomaly detection results, a cable condition sensing result is generated, characterizing whether an anomaly exists, the type of anomaly, and its severity. This method effectively overcomes the problems of low early defect identification accuracy in traditional cable monitoring caused by single data acquisition, strong reliance on manual annotation, and scarcity of anomaly samples, through multi-dimensional operational data fusion and a semi-supervised learning mechanism based on cluster self-labeling. It achieves accurate perception and discrimination of cable anomaly types and severity. Attached Figure Description

[0017] To more clearly illustrate the technical solutions in this invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0018] Figure 1 This is a schematic flowchart of the cable status sensing method provided in an embodiment of the present invention.

[0019] Figure 2 This is a schematic diagram of the cable status sensing device provided in an embodiment of the present invention.

[0020] Figure 3 This is a schematic diagram of the physical structure of the electronic device provided in the embodiment of the present invention. Detailed Implementation

[0021] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention.

[0022] It should be noted that with the acceleration of urbanization and the continuous upgrading of the power system, the operating status of high-voltage cables, as core equipment of the power grid, directly determines the stability and security of power supply, and the importance of cable condition monitoring technology is becoming increasingly prominent.

[0023] Current mainstream cable condition monitoring methods have several limitations. In the data acquisition stage, traditional monitoring often relies on single-parameter acquisition, such as monitoring only cable temperature or partial discharge signals, which is insufficient to comprehensively reflect the cable's operating status. Furthermore, some sensors are complex to install; for example, some solutions require 6-12 sensors per cable joint, necessitating individual sensor ID verification during installation. This not only increases construction difficulty and the risk of human error but also significantly raises subsequent maintenance costs. Simultaneously, the monitoring environment significantly interferes with data transmission and equipment operation. In underground environments such as cable trenches and joint wells, wireless signals are easily blocked, leading to unstable data reception and potentially reducing the online rate to zero. Moreover, prolonged exposure to water accumulation and high humidity can shorten equipment lifespan and affect the long-term reliability of the monitoring system.

[0024] At the data processing and analysis level, traditional methods lack the ability to deeply integrate multi-dimensional data, making it difficult to uncover potential correlations between data points. Furthermore, the monitoring models constructed are often limited by the imbalanced distribution of data samples, with far more data from normal operation than from anomalies. This leads to overfitting of the model to the normal state, resulting in low accuracy in identifying early latent defects in cables. Moreover, traditional monitoring often remains at the level of anomaly alarms, failing to accurately determine the type, severity, and development trend of anomalies, thus failing to meet the needs of modern power grids for precise perception and prediction of cable conditions.

[0025] In view of this, the present invention proposes a novel cable status sensing method, specifically, Figure 1 A flowchart illustrating the cable status sensing method provided in an embodiment of the present invention is shown.

[0026] like Figure 1 As shown, the method includes: S110, acquiring multi-dimensional operating data during cable operation, wherein the multi-dimensional operating data includes at least temperature data, current data, and insulation resistance; S120, predicting cable anomaly detection results based on a pre-trained classification model and the multi-dimensional operating data; wherein the classification model is obtained through semi-supervised training using a labeled dataset and an unlabeled target domain dataset formed by clustering category labels of historical cable operating data; S130, generating cable state perception results based on the cable anomaly detection results, wherein the cable state perception results characterize whether the cable has an anomaly, the anomaly category, and the severity.

[0027] The following will provide a detailed description of steps S110-S130 and related steps.

[0028] S110, acquire multi-dimensional operating data during cable operation, the multi-dimensional operating data including at least temperature data, current data and insulation resistance.

[0029] In step S110, the multi-dimensional operational data refers to real-time monitoring data of multiple physical dimensions synchronously collected from the operating high-voltage cable system. The multi-dimensional operational data includes at least temperature data, current data, and insulation resistance.

[0030] Temperature data reflects the thermal state of the cable conductor or joint and can be obtained by a distributed fiber Bragg grating temperature sensor; current data characterizes the cable's load level and current waveform characteristics and can be acquired by a Hall current sensor; insulation resistance is a key electrical parameter for measuring the main insulation performance of the cable and is obtained by an online insulation monitoring device based on microcurrent detection technology. Specifically, it uses coupling capacitance to extract the cable's leakage current to ground and converts it into insulation resistance value according to Ohm's law.

[0031] Temperature data, current data, and insulation resistance data are used to characterize the cable's operating status from three dimensions: heat, electricity, and insulation, respectively, avoiding the loss of status information due to monitoring a single parameter.

[0032] Based on the multi-dimensional operation data obtained in step S110 during cable operation, step S120 is further executed.

[0033] S120, Based on a pre-trained classification model, predict the cable anomaly detection result according to the multi-dimensional operation data; the classification model is obtained by semi-supervised training of a labeled dataset and an unlabeled target domain dataset formed by the category labels generated by the historical operation data of the cable and its clustering.

[0034] In step S120, the classification model is a machine learning model used to determine the operating status of the cable. Its input is multi-dimensional operating data, and its output is whether the cable is abnormal, the preliminary judgment result of the abnormality, and the corresponding confidence value.

[0035] It is worth mentioning that the classification model was not trained using manually labeled samples, but rather through semi-supervised learning. Specifically, the training data used in the semi-supervised training consisted of two parts: one part was a labeled dataset formed by historical cable operation data and the category labels generated by its clustering, and the other part was an unlabeled target domain dataset.

[0036] The cable historical operation data consists of cable operation records accumulated over a period of time without manual annotation. Cluster-generated category labels are cluster identifiers assigned to each data record after unsupervised clustering of the historical operation data, representing a typical operating state pattern, such as normal steady state, slow insulation degradation, or localized overheating. The labeled dataset consists of paired cable historical operation data records with their corresponding cluster-generated category labels. The unlabeled target domain dataset refers to the collection of multi-dimensional operation data generated by the currently monitored cable lines that has not yet been labeled.

[0037] By inputting both labeled and unlabeled target domain datasets into a semi-supervised learning framework, a classification model is trained, thereby enabling effective identification of cable anomalies without relying on manual annotation.

[0038] Based on the pre-trained classification model and the prediction of cable anomaly detection results according to multi-dimensional running data in step S120, step S130 is further executed.

[0039] S130, Generate cable status perception results based on the cable anomaly detection results, wherein the cable status perception results characterize whether the cable has an anomaly, the type of anomaly, and the severity of the anomaly.

[0040] In step S130, the cable status perception result is a structured description of the current health status of the cable, which not only indicates whether there is an abnormality in the cable, but also further clarifies the specific category and severity of the abnormality.

[0041] The anomaly categories include, but are not limited to, typical fault modes such as poor joint contact, insulation aging, overload operation, and partial discharge; the severity is divided into three levels—low, medium, and high—based on the intensity, duration, and development trend of the anomaly characteristics, to guide maintenance priorities.

[0042] For example, when the classification model outputs "an anomaly exists" and the anomaly category is "decreased insulation resistance", and the confidence level is higher than the preset confidence threshold, the cable status perception result will be generated as: "The cable has an anomaly, the anomaly category is insulation degradation, and the severity is medium".

[0043] Subsequently, the cable status perception results are directly pushed to the operation and maintenance management platform to provide accurate basis for maintenance decisions.

[0044] In this embodiment, multi-dimensional operational data during cable operation is acquired, including at least temperature, current, and insulation resistance data. Based on a pre-trained classification model, cable anomaly detection results are predicted using this multi-dimensional operational data. The classification model is obtained through semi-supervised training using labeled datasets and unlabeled target domain datasets formed from historical cable operational data and clustered category labels. Cable status perception results are generated based on the cable anomaly detection results, characterizing the presence, type, and severity of anomalies. This method, through multi-dimensional operational data fusion and a semi-supervised learning mechanism based on clustering self-labeling, effectively overcomes the problems of low early defect identification accuracy in traditional cable monitoring caused by single data acquisition, strong reliance on manual labeling, and scarcity of anomaly samples. It achieves accurate perception and discrimination of cable anomaly types and severity.

[0045] Based on the above embodiments, the following will further describe in detail the process of acquiring multi-dimensional running data in step S110.

[0046] Acquire multi-dimensional operational data during cable operation, including: collecting temperature data through fiber optic temperature sensors deployed at key locations on the cable joints and body; collecting current data through Hall current sensors integrated into the cable online monitoring device; and obtaining insulation resistance through an online insulation monitoring device based on micro-current detection technology, which includes extracting leakage current through coupling capacitors and converting it into insulation resistance values.

[0047] It is easy to understand that multi-dimensional operating data is a collection of multiple physical parameters used to comprehensively characterize the operating status of the cable. In this embodiment, it includes at least temperature data, current data, and insulation resistance.

[0048] Temperature data is collected using fiber Bragg grating temperature sensors deployed at cable joints and key locations on the cable body. Based on the physical characteristic of fiber Bragg gratings where wavelength changes with temperature, these sensors enable high-precision, distributed temperature measurement without introducing electromagnetic interference. In practical deployments, the fiber Bragg grating temperature sensors are arranged along cable joints, intermediate connection points, and areas prone to heat generation to detect localized temperature anomalies.

[0049] During cable operation, faults such as overload, poor contact, or insulation aging can generate abnormal heat at the fault point, causing a significant temperature increase. For example, when a cable joint experiences increased contact resistance due to poor crimping, its temperature will be significantly higher than in normal sections. Therefore, real-time monitoring of temperature data can help detect potential thermal faults in a timely manner, preventing serious accidents such as cable burnout or insulation breakdown caused by localized overheating.

[0050] Current data is acquired through a Hall current sensor integrated into the cable online monitoring device. Based on the Hall effect principle, the Hall current sensor can measure AC or DC current non-contactly, offering advantages such as fast response, good isolation, and strong anti-interference capability. The Hall current sensor is typically installed at the cable terminal or inside the monitoring cabinet, directly surrounding the cable conductor to obtain the cable's load current value in real time. Current is a direct reflection of the electrical energy transmitted through a cable, and its magnitude reflects the cable's load level.

[0051] When the current exceeds the cable's rated current-carrying capacity, it not only accelerates the aging process of the insulation material but may also cause overheating due to the Joule heating effect. Simultaneously, abnormal current fluctuations (such as sudden increases, decreases, or harmonic distortion) often indicate electrical anomalies such as short circuits or grounding faults in the power supply system or the cable itself. Therefore, current data is a crucial basis for assessing the operational safety of cables and the stability of the system.

[0052] Insulation resistance is obtained through an online insulation monitoring device based on micro-current detection technology. Micro-current detection technology refers to a method that uses a high-sensitivity circuit to detect the weak leakage current flowing through the main insulation layer of the cable and calculates the insulation resistance value accordingly. Specifically, the online insulation monitoring device uses a coupling capacitor as a signal extraction unit. The coupling capacitor is connected between the cable's metal shielding layer and ground to couple and extract the ground leakage current signal in the nanoampere to microampere range. Subsequently, after signal conditioning and analog-to-digital conversion, the ground leakage current signal is used to calculate the insulation resistance value according to Ohm's law.

[0053] Insulation resistance is a core indicator for evaluating the main insulation performance of cables. When the cable insulation layer is affected by factors such as chemical corrosion, mechanical damage, moisture intrusion, or long-term electrothermal aging, its insulation performance deteriorates, manifested as a continuous decrease in insulation resistance value. By continuously monitoring the trend of insulation resistance changes, the integrity of the insulation condition can be effectively assessed, potential insulation defects can be detected in advance, and safety accidents such as leakage, phase-to-phase short circuits, or grounding breakdown caused by insulation failure can be prevented.

[0054] This embodiment comprehensively reflects the operating status of the cable by acquiring multi-dimensional operating data during the cable's operation, from three interrelated and complementary physical dimensions: heat, electricity, and insulation.

[0055] Based on the above embodiments, the training and optimization process of the classification model in step S120 will be described in detail below.

[0056] The classification model is obtained through semi-supervised training using a labeled dataset and an unlabeled target domain dataset, both formed from historical cable operation data and category labels generated by clustering. Specifically, this includes: constructing a structured data map based on historical cable operation data; clustering historical operation data records in the structured data map and assigning category labels to each historical operation data record based on the clustering results to form a labeled dataset; performing category distribution balancing on the labeled dataset to alleviate the imbalance in the number of samples between normal and abnormal states; and using the category-balanced labeled dataset and the unlabeled target domain dataset together as training data to train the classification model through semi-supervised learning.

[0057] The labeled dataset (after class distribution balancing) and the unlabeled target domain dataset are used together as training data to train a classification model using semi-supervised learning. This includes: clustering the unlabeled target domain dataset to obtain multiple data clusters; for each sample in the unlabeled target domain dataset, calculating its average similarity with other samples in the same cluster as intra-cluster similarity, and simultaneously calculating its average similarity with samples in the nearest dissimilar clusters as inter-cluster similarity; determining the clustering evaluation value of the sample based on intra-cluster and inter-cluster similarities; encoding the sample using a variational autoencoder to generate predicted labels, and correcting the predicted labels based on the clustering evaluation value to obtain state labels; during semi-supervised training, the state labels are used as pseudo-labels and participate in model training along with the labeled dataset to optimize the classification model's ability to discriminate target domain data. The process is straightforward: first, collect historical cable operation data, and then construct a structured data graph based on this data. A structured data graph is a data structure formed by organizing previously scattered and disordered historical operation records according to time sequence, line topology, equipment attributes, and multi-dimensional operation characteristics. It is used to clearly present the inherent connections and contextual relationships between different data records.

[0058] Subsequently, the historical operation data records in the structured data graph are clustered. In this embodiment, the K-means clustering algorithm is used to perform unsupervised partitioning of the historical operation data records: a portion of data samples are randomly selected as initial seed points, the distance between each data sample and each initial seed point is calculated, and each sample is assigned to the cluster to which the nearest seed point belongs; then the center position of each cluster is updated, and the aforementioned allocation and update process is repeated until a preset stopping condition is met, such as the change in cluster center is less than a threshold or the maximum number of iterations is reached, thereby completing the clustering of all historical operation data records.

[0059] After clustering is completed, a category label is assigned to each historical operating data record based on the clustering results. This category label is the identifier of its cluster and represents a typical cable operating state mode, such as normal steady state, slow insulation degradation, or localized overheating. The resulting set of historical operating data records and their corresponding category labels constitutes the labeled dataset.

[0060] Because cables spend far more time in normal conditions than in abnormal conditions during actual operation, the number of normal state samples in the labeled dataset significantly exceeds the number of abnormal state samples, resulting in an imbalanced class distribution. Therefore, class distribution balancing is performed on the labeled dataset. This balancing can be achieved through methods such as oversampling, undersampling, or synthesizing minority class samples to adjust the proportion of samples in each class, resulting in a more reasonably distributed labeled dataset. This prevents the model from overfitting to normal states during training due to data bias, thereby improving the ability to identify abnormal states.

[0061] Next, the labeled dataset after class distribution balancing and the unlabeled target domain dataset are used together as training data to train a classification model through semi-supervised learning. The unlabeled target domain dataset refers to a collection of historical operational data from the cable lines to be monitored, which has not yet undergone any manual or automatic labeling.

[0062] In the semi-supervised training process, the unlabeled target domain dataset is further processed to generate high-quality supervision signals. Specifically, the unlabeled target domain dataset is clustered to obtain multiple data clusters, each representing a class of cable operating states with similar characteristics. For each sample in the unlabeled target domain dataset, its average similarity with other samples in the same cluster is calculated as the intra-cluster similarity, and its average similarity with samples in the nearest different cluster is calculated as the inter-cluster similarity. Intra-cluster similarity and inter-cluster similarity can be determined based on metrics such as Euclidean distance and cosine similarity.

[0063] Based on intra-cluster similarity and inter-cluster similarity, a cluster evaluation value is determined for each sample. The cluster evaluation value is used to measure the reasonableness of the corresponding sample in its respective cluster. The higher the intra-cluster similarity and the lower the inter-cluster similarity, the higher the cluster evaluation value, indicating that the cluster assignment of the sample is more reliable.

[0064] Subsequently, a variational autoencoder (VAE) is used to encode each sample to generate predicted labels. A VAE is a deep generative model that learns low-dimensional latent feature representations from input data and generates category predictions based on these representations. In this embodiment, the VAE is trained to output predicted labels corresponding to historical clustering categories. To further improve label accuracy, the predicted labels are corrected based on clustering evaluation values. Specifically, if a sample's clustering evaluation value is lower than a preset threshold, indicating that its original predicted label may be unreliable, it is corrected by incorporating label information from its neighboring high-confidence samples, ultimately obtaining the state label.

[0065] Finally, during semi-supervised training, state labels are used as pseudo-labels and participate in model training alongside the labeled dataset. Specifically, the labeled dataset provides strong supervisory signals for calculating cross-entropy loss, while the unlabeled target domain data and its pseudo-labels are used to calculate consistency loss or soft-label supervision loss. By jointly optimizing the two types of losses, the classification model not only learns typical state patterns in historical cable operation data but also adaptively aligns with the actual distribution of the target domain data, thereby significantly enhancing its ability to discriminate cable anomalies in real-world operating scenarios.

[0066] This embodiment achieves efficient training of a high-precision classification model using historical cable operation data and unlabeled target domain data without any manual annotation intervention by constructing a structured data graph, automatically generating category labels through clustering, balancing category distribution, and a pseudo-label correction mechanism based on clustering evaluation. This effectively overcomes the problems of poor model generalization ability caused by high annotation costs, data imbalance, and domain mismatch in traditional methods.

[0067] Based on the above embodiments, the preprocessing process of multi-dimensional running data in step S110 will be described in detail below.

[0068] The process involves acquiring multi-dimensional operational data during cable operation, followed by: feature filtering of the multi-dimensional operational data, removing redundant features whose correlation with cable status is lower than a preset threshold, and obtaining a filtered feature set; and using wavelet transform to perform time-frequency domain decomposition on the filtered feature set to obtain processed multi-dimensional operational data.

[0069] Accordingly, based on the pre-trained classification model, the cable anomaly detection results are predicted according to the multi-dimensional operational data, including: inputting the processed multi-dimensional operational data into the pre-trained classification model to obtain the output cable anomaly detection results.

[0070] It's easy to understand that after acquiring multi-dimensional operational data during cable operation, the first step is to perform feature filtering on this data. Since some features have a weak correlation with cable condition assessment and may introduce noise or increase computational burden, redundant features need to be removed.

[0071] Specifically, the correlation index between each feature and the cable condition category (such as normal, overheated, insulation deterioration, etc.) is calculated. The correlation index can be obtained using methods such as mutual information, Pearson correlation coefficient, or feature importance scoring based on a tree model. Features with a correlation index below a preset threshold are identified as redundant features and removed, thus obtaining the filtered feature set. The preset threshold can be determined based on historical data statistical analysis or cross-validation to ensure that the retained features make a significant contribution to condition determination.

[0072] Subsequently, wavelet transform was used to perform time-frequency domain decomposition on the filtered feature set to obtain processed multi-dimensional operational data. Wavelet transform is a signal analysis method that can decompose the original time series signal into sub-band components of different scales and frequencies, thereby simultaneously capturing the signal's local abrupt changes and overall trends. In this embodiment, for key features such as temperature and current, wavelet transform was used to extract their high-frequency detail components (reflecting transient anomalies, such as microsecond-level fluctuations caused by short-circuit impacts and partial discharges) and low-frequency approximate components (reflecting slow degradation trends, such as long-term decreases in insulation resistance). Through time-frequency domain decomposition, the anomalous features that were originally blurred or submerged in noise in the time domain were significantly enhanced, greatly improving the recognizability and effectively solving the problem of model recognition difficulties caused by the lack of obvious features in the original data.

[0073] Accordingly, when predicting cable anomaly detection results based on a pre-trained classification model, the processed multi-dimensional operational data is input into the trained classification model. The classification model consists of three parts: a pre-trained feature extraction network, an attention mechanism module, and a classifier. The pre-trained feature extraction network is a deep neural network (such as a convolutional neural network or a Transformer encoder) pre-trained on large-scale historical cable operational data. It has the ability to automatically extract deep, implicit features from the input data. With the help of the feature extraction network, high-order abstract representations can be quickly extracted from the processed multi-dimensional operational data without having to learn basic feature patterns from scratch.

[0074] Furthermore, the classification model introduces an attention mechanism to dynamically weight the extracted features. The attention mechanism is an adaptive feature selection mechanism that automatically assigns higher weights to feature channels or time segments highly correlated with abnormal states based on the content of the current input data, while suppressing the contribution of secondary or irrelevant features. For example, when a sustained decrease in the low-frequency component of insulation resistance is detected, the attention mechanism strengthens the influence of this feature in the final decision; conversely, when the current signal is stable and undisturbed, its weight is reduced. By introducing the attention mechanism, the sensitivity to abnormally related features is significantly improved, and the anti-interference capability is enhanced.

[0075] Finally, the classifier classifies the cable anomaly based on the attention-weighted feature representation, outputting the category of the cable anomaly and the corresponding confidence score. The anomaly categories include, but are not limited to, typical conditions such as joint overheating, insulation aging, overload operation, and grounding faults. The confidence score reflects the model's certainty about the classification result, typically ranging from 0 to 1; a higher value indicates a more reliable judgment.

[0076] Operations and maintenance personnel can classify alarm results based on confidence scores to avoid unnecessary on-site verification caused by false alarms with low confidence scores.

[0077] This embodiment significantly improves the accuracy and robustness of cable anomaly detection by eliminating redundant information through feature filtering, enhancing key anomaly features through wavelet transform, extracting deep representations through a pre-trained network, and focusing discriminative information through an attention mechanism.

[0078] Based on the above embodiments, the post-processing of the cable status sensing results in step S130 will be described in detail below.

[0079] Based on the cable anomaly detection results, a cable status perception result is generated, which then includes: combining the anomaly type, severity, and historical cable operation data to identify the development pattern of similar anomalies; introducing current environmental parameters to dynamically correct the development pattern to obtain the corrected development pattern; predicting the cable's status evolution trend within a preset time window based on the corrected development pattern, and generating fault warning information.

[0080] The process is straightforward: first, cable status perception results are generated based on cable anomaly detection results. These results clearly indicate whether an anomaly exists in the cable, the specific type of anomaly, and its severity. Anomaly types include, but are not limited to, typical patterns such as joint overheating, insulation degradation, overload operation, or grounding faults. Severity is categorized into low, medium, and high levels based on the intensity, duration, and confidence level of the anomaly characteristics. For example, when a continuous decrease in insulation resistance is detected with a confidence level higher than 0.9, the system generates the status perception result: "Anomaly exists; anomaly type: insulation degradation; severity: high." These cable status perception results provide intuitive and actionable maintenance information, enabling maintenance personnel to quickly pinpoint the core issue and avoid the blind troubleshooting that often results from simply indicating "anomaly exists."

[0081] Furthermore, after generating the cable status perception results, trend prediction and early warning functions are executed. Specifically, by combining the anomaly category, severity, and historical cable operation data, the development patterns of similar anomalies are identified. Through statistical analysis or pattern mining of historical cable operation data, typical development paths for specific anomaly categories can be summarized. For example, if historical cable operation data shows that approximately 70% of cases of "abnormal joint temperature" in a certain type of cross-linked polyethylene cable result in insulation breakdown within 30 days, then the risk probability within this time window (30 days) can serve as a reference for predicting similar anomalies in the future. Development patterns are a quantified or rule-based expression of such common evolutionary patterns.

[0082] However, relying solely on historical cable operating data may overlook the differences in current operating conditions. Therefore, this embodiment introduces current environmental parameters to dynamically correct the development pattern, resulting in a corrected development pattern. These current environmental parameters include, but are not limited to, external or operating condition factors such as ambient temperature, humidity, load rate, and soil thermal conductivity.

[0083] For example, in high-temperature and high-humidity environments, the aging rate of cable insulation materials accelerates significantly, and the same degree of decrease in insulation resistance may evolve into a breakdown fault in a shorter time; while under dry and low-temperature conditions, the degradation process is relatively slow. By using current environmental parameters as correction factors, the time scale, deterioration rate, or risk threshold in the development pattern can be adaptively adjusted, making the corrected development pattern more consistent with the current actual scenario and improving the accuracy and practicality of predictions.

[0084] Finally, based on the corrected development pattern, the cable's state evolution trend is predicted within a preset time window, and fault warning information is generated. The preset time window can be set according to maintenance needs, such as 7 days, 15 days, or 30 days. The predicted state evolution trend includes whether the anomaly continues to worsen, whether it will evolve into a more serious fault (such as from "insulation deterioration" to "insulation breakdown"), and the expected time range for its occurrence.

[0085] Fault warning information is output in the form of structured messages, including risk level, suggested handling time limit, and recommended maintenance measures. For example: "High-risk warning: The current insulation degradation is expected to develop into a breakdown fault within 10-14 days. It is recommended to arrange a power outage for maintenance within 72 hours." Fault warning information supports priority sorting of multiple line anomalies, automatically prioritizing anomalies with "high severity + near predicted fault time" to allocate human and material resources first, thereby optimizing operation and maintenance efficiency and avoiding delays in handling high-risk hazards due to resource misallocation.

[0086] This embodiment constructs a trend prediction mechanism with dynamic adaptability, which significantly improves the foresight and safety of cable operation and maintenance, and effectively reduces the losses caused by sudden failures.

[0087] Corresponding to the cable state sensing methods described in the above embodiments, the present invention also proposes a cable state sensing device, specifically, Figure 2 A schematic diagram of the cable status sensing device provided in an embodiment of the present invention is shown.

[0088] like Figure 2As shown, the device includes: a multi-dimensional operation data acquisition module 210, used to acquire multi-dimensional operation data during cable operation, the multi-dimensional operation data including at least temperature data, current data, and insulation resistance; a cable anomaly detection result prediction module 220, used to predict cable anomaly detection results based on a pre-trained classification model and the multi-dimensional operation data; the classification model is obtained through semi-supervised training using a labeled dataset and an unlabeled target domain dataset formed by the category labels generated from the cable's historical operation data and its clustering; and a cable state perception result generation module 230, used to generate cable state perception results based on the cable anomaly detection results, the cable state perception results representing whether the cable has an anomaly, the anomaly category, and the severity.

[0089] In this embodiment, a multi-dimensional operational data acquisition module 210 acquires multi-dimensional operational data during cable operation, including at least temperature data, current data, and insulation resistance. A cable anomaly detection result prediction module 220 predicts cable anomaly detection results based on a pre-trained classification model and the multi-dimensional operational data. The classification model is obtained through semi-supervised training using labeled datasets and unlabeled target domain datasets formed by clustering historical cable operational data and their generated category labels. A cable state perception result generation module 230 generates cable state perception results based on the cable anomaly detection results. These results characterize whether an anomaly exists in the cable, the type of anomaly, and its severity. This device effectively overcomes the problems of low early defect identification accuracy in traditional cable monitoring caused by single data acquisition, strong reliance on manual annotation, and scarcity of anomaly samples, through multi-dimensional operational data fusion and a semi-supervised learning mechanism based on cluster self-labeling. It achieves accurate perception and discrimination of cable anomaly types and severity.

[0090] It should be noted that the cable status sensing device provided in this embodiment can be referred to in correspondence with the cable status sensing methods described in the above embodiments, and will not be repeated here.

[0091] Figure 3 An example is a schematic diagram of the physical structure of an electronic device, such as... Figure 3As shown, the electronic device may include a processor 310, a communications interface 320, a memory 330, and a communication bus 340, wherein the processor 310, communications interface 320, and memory 330 communicate with each other via the communication bus 340. The processor 310 can call logical instructions in the memory 330 to execute a cable state sensing method. This method includes: acquiring multi-dimensional operational data during cable operation, the multi-dimensional operational data including at least temperature data, current data, and insulation resistance; predicting cable anomaly detection results based on a pre-trained classification model according to the multi-dimensional operational data; the classification model is obtained through semi-supervised training using a labeled dataset and an unlabeled target domain dataset formed by clustering historical cable operational data and its category labels; and generating a cable state sensing result based on the cable anomaly detection result, the cable state sensing result representing whether the cable has an anomaly, the anomaly category, and its severity.

[0092] Furthermore, the logical instructions in the aforementioned memory 330 can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, essentially, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0093] On the other hand, the present invention also provides a computer program product, which includes a computer program that can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer can execute the cable status sensing method provided by the above methods. The method includes: acquiring multi-dimensional operating data during cable operation, wherein the multi-dimensional operating data includes at least temperature data, current data, and insulation resistance; predicting cable anomaly detection results based on a pre-trained classification model according to the multi-dimensional operating data; the classification model is obtained by semi-supervised training using a labeled dataset and an unlabeled target domain dataset formed by the category labels generated from the cable's historical operating data and its clustering; and generating a cable status sensing result based on the cable anomaly detection result, wherein the cable status sensing result characterizes whether the cable has an anomaly, the type of anomaly, and the severity.

[0094] In another aspect, the present invention also provides a non-transitory computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the cable state sensing method provided by the above methods. The method includes: acquiring multi-dimensional operational data during cable operation, the multi-dimensional operational data including at least temperature data, current data, and insulation resistance; predicting cable anomaly detection results based on a pre-trained classification model according to the multi-dimensional operational data; the classification model being obtained through semi-supervised training using a labeled dataset and an unlabeled target domain dataset formed by clustering historical cable operational data and its generated category labels; and generating a cable state sensing result based on the cable anomaly detection result, the cable state sensing result characterizing whether the cable has an anomaly, the anomaly category, and its severity.

[0095] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0096] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0097] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A cable status sensing method, characterized in that, include: Acquire multi-dimensional operational data during cable operation, including at least temperature data, current data, and insulation resistance; Based on a pre-trained classification model, the cable anomaly detection results are predicted according to the multi-dimensional operational data. The classification model is obtained through semi-supervised training using a labeled dataset and an unlabeled target domain dataset formed by the category labels generated from the historical operation data of cables and their clustering. Based on the cable anomaly detection results, a cable status perception result is generated, which characterizes whether the cable has an anomaly, the type of anomaly, and the severity of the anomaly.

2. The cable status sensing method according to claim 1, characterized in that, The classification model is obtained through semi-supervised training using a labeled dataset and an unlabeled target domain dataset formed by clustering historical cable operation data and its resulting category labels. Specifically, it includes: A structured data map is constructed based on the historical operating data of the cable; The historical operational data records in the structured data map are clustered, and a category label is assigned to each historical operational data record according to the clustering results to form the labeled dataset; The labeled dataset is subjected to class distribution balancing processing to alleviate the imbalance in the number of samples between normal and abnormal states; The labeled dataset after class distribution balancing and the unlabeled target domain dataset are used together as training data, and the classification model is trained through semi-supervised learning.

3. The cable status sensing method according to claim 2, characterized in that, The step of using both the labeled dataset (after class distribution balancing) and the unlabeled target domain dataset as training data, and training the classification model through semi-supervised learning, includes: Clustering is performed on the unlabeled target domain dataset to obtain multiple data clusters; For each sample in the unlabeled target domain dataset, the average similarity between it and other samples in the same cluster is calculated as the intra-cluster similarity, and the average similarity between it and each sample in the nearest dissimilar cluster is calculated as the inter-cluster similarity. Based on the intra-cluster similarity and the inter-cluster similarity, the cluster evaluation value of the sample is determined; The samples are encoded using a variational autoencoder to generate predicted labels, and the predicted labels are corrected based on the clustering evaluation values ​​to obtain state labels. During semi-supervised training, the state labels are used as pseudo-labels and participate in model training together with the labeled dataset to optimize the classification model's ability to discriminate target domain data.

4. The cable status sensing method according to claim 1, characterized in that, The acquisition of multi-dimensional operational data during cable operation includes: The temperature data is collected by fiber optic grating temperature sensors deployed at key locations on the cable connector and the main body. The current data is acquired by a Hall current sensor integrated into the cable online monitoring device; The insulation resistance is obtained by an online insulation monitoring device based on microcurrent detection technology, which includes extracting leakage current through a coupling capacitor and converting it into an insulation resistance value.

5. The cable status sensing method according to claim 1, characterized in that, The acquisition of multi-dimensional operational data during cable operation includes: The multi-dimensional operational data is subjected to feature filtering to remove redundant features whose correlation with cable status discrimination is lower than a preset threshold, resulting in a filtered feature set. Wavelet transform is used to decompose the filtered feature set in the time-frequency domain to obtain the processed multi-dimensional operational data. Accordingly, the pre-trained classification model predicts cable anomaly detection results based on the multi-dimensional operational data, including: The processed multi-dimensional operational data is input into a pre-trained classification model to obtain the output cable anomaly detection results.

6. The cable status sensing method according to any one of claims 1-5, characterized in that, The step of generating cable status perception results based on the cable anomaly detection results then includes: By combining the aforementioned anomaly categories, severity levels, and historical cable operation data, the development patterns of similar anomalies can be identified; The development pattern is dynamically corrected by incorporating current environmental parameters, resulting in the corrected development pattern. Based on the corrected development law, the cable's state evolution trend within a preset time window is predicted, and fault warning information is generated.

7. A cable status sensing device, characterized in that, include: A multi-dimensional operation data acquisition module is used to acquire multi-dimensional operation data during cable operation, wherein the multi-dimensional operation data includes at least temperature data, current data, and insulation resistance; The cable anomaly detection result prediction module is used to predict the cable anomaly detection result based on the multi-dimensional operational data and a pre-trained classification model. The classification model is obtained through semi-supervised training using a labeled dataset and an unlabeled target domain dataset formed by the category labels generated from the historical operation data of cables and their clustering. The cable status perception result generation module is used to generate cable status perception results based on the cable anomaly detection results. The cable status perception results characterize whether the cable has an anomaly, the type of anomaly, and the severity of the anomaly.

8. An electronic device comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that, When the processor executes the computer program, it implements the cable status sensing method as described in any one of claims 1 to 6.

9. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the cable status sensing method as described in any one of claims 1 to 6.

10. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by the processor, it implements the cable status sensing method as described in any one of claims 1 to 6.

Images