A cable partial discharge intelligent positioning and insulation degradation early warning method and system

Abstract

This invention relates to the field of cable testing technology, and more specifically, to a method and system for intelligent location of partial discharge and early warning of insulation degradation in cables. By deploying a distributed fiber optic ultrasonic and ultra-high frequency sensor array, partial discharge signals are simultaneously acquired. Deep learning algorithms are used for interference stripping and feature extraction. A dual-channel neural network is employed to collaboratively analyze the acoustic-electric fusion signal to obtain preliminary location data. Subsequently, the location information is imported into a high-precision digital twin model of the cable gallery. Accurate corrections for multipath effects and environmental attenuation are performed through acoustic-electromagnetic wave propagation simulation, achieving three-dimensional sub-meter-level precise location of the fault point. A partial discharge event fingerprint database and an insulation state evolution model are constructed to dynamically assess and predict the degradation stage and remaining electrical life of the cable insulation, forming an intuitive insulation "health cloud map." Finally, differentiated operation and maintenance strategies, from online monitoring to planned maintenance, are automatically generated based on the risk level, forming a complete predictive maintenance closed loop.

Description

Technical Field

[0001] This invention relates to the field of cable testing, and more specifically, to a method and system for intelligent location of partial discharge and early warning of insulation degradation in cables. Background Technology

[0002] With the increasing cable coverage of urban power grids and the large-scale construction of integrated utility tunnels, the safe operation of high-voltage cables faces new challenges. The enclosed spaces, complex pipeline network, and harsh electromagnetic environment of urban underground utility tunnels render traditional cable fault monitoring methods severely inadequate in this context.

[0003] Currently, monitoring of partial discharge in cables mainly relies on ultra-high frequency (UHF) methods, ultrasonic methods, or a simple combination of both. However, in the specific scenario of utility tunnels, existing technologies face the following bottlenecks: First, strong background electromagnetic interference and complex acoustic noise generated by lighting, frequency converters, communication signals, and other cable lines within the tunnel result in extremely low signal-to-noise ratios for partial discharge signals, making effective feature extraction difficult. Second, the multipath reflection, refraction, and attenuation effects generated by signals in the metal structure and multi-medium interfaces of the tunnel lead to a significant decrease in the accuracy of positioning algorithms based on single physical quantities (such as electromagnetic waves or sound waves). Third, existing methods focus primarily on the immediate detection and rough location of partial discharge events, lacking the ability to analyze and predict the correlation between the historical evolution patterns of partial discharge signals and the long-term degradation trends of cable insulation, thus failing to achieve the transition from "fault alarm" to "condition warning." Finally, the monitoring system is disconnected from the integrated management platform of the utility tunnel (such as BIM systems, environmental control systems, and robotic inspection systems), preventing early warning information from driving accurate and efficient collaborative operation and maintenance decisions. Summary of the Invention

[0004] To address the shortcomings of the existing technologies, the present invention aims to provide a method and system for intelligent location of partial discharge and early warning of insulation degradation in cables. This system addresses the technical problems of signal extraction difficulties, low location accuracy, lack of state prediction capabilities, and poor system linkage in cable partial discharge monitoring under complex pipe gallery environments.

[0005] The above-mentioned technical objective of this invention is achieved through the following technical solution: a method for intelligent location of partial discharge and early warning of insulation degradation in cables, comprising the following steps: Step S1: Multi-physical quantity fusion signal acquisition and preprocessing. Using a distributed fiber optic ultrasonic sensor array and a UHF electromagnetic sensor array deployed at key nodes of the cable tunnel, acoustic and electromagnetic signals excited by partial discharge of the cable are simultaneously acquired. The original signals are synchronously aligned and adaptively denoised, and a blind source separation algorithm is used to suppress background interference from the tunnel, extracting a clean partial discharge pulse sequence.

[0006] Step S2: Signal feature extraction and initial localization based on deep learning. The preprocessed acoustic-electric fusion signal is input into a dual-channel spatiotemporal convolutional neural network to learn the time delay characteristics of the ultrasonic signal and the direction of arrival characteristics of the UHF signal, respectively. Through the feature fusion layer, the differences in the physical propagation characteristics of the two signals are combined to generate a candidate set of spatiotemporal coordinates containing the logical location of the suspected partial discharge source.

[0007] Step S3: High-precision coordinate correction based on the digital twin of the utility tunnel. A high-precision BIM digital twin model is constructed, including the 3D structure of the utility tunnel, pipeline layout, and material properties. The coordinate candidate set generated in Step S2 is imported into the model, and a simulation algorithm for the propagation of sound waves and electromagnetic waves in a multi-medium environment is loaded to calculate the residual between the theoretical propagation path and the actual receiving path of the signal. An iterative optimization algorithm is used to compensate for propagation loss and correct for multipath effects on the candidate coordinates, outputting the accurate 3D physical coordinates of the partial discharge source in the digital twin model.

[0008] Step S4: Insulation State Evolution Modeling and Risk Prediction. A hidden Markov model of the cable insulation state is established, with hidden states defined as "healthy," "early defect," "developing defect," and "near failure." The historical and real-time partial discharge events located in Step S3, along with their corresponding signal spectra, discharge quantities, frequency of occurrence, and physical coordinates, constitute a "partial discharge event fingerprint," which serves as the model's observation sequence. Using a transfer learning framework, the model can learn the evolutionary patterns of insulation defects from the accumulated "event fingerprint" database, predict the probability of the current cable segment being in various degradation states within a preset timeframe, and estimate its remaining electrical lifetime.

[0009] Step S5: Visualized Early Warning and Linked Decision-Making with the Utility Tunnel System. The insulation risk probability predicted in Step S4 is mapped to the digital twin model of the utility tunnel to generate a dynamically updated "health cloud map" of cable insulation. Based on the risk level and location information, a differentiated operation and maintenance strategy, ranging from "enhanced online monitoring" to "planned maintenance," is automatically generated through a dynamic priority algorithm. This triggers linkage instructions with the utility tunnel environmental control system and the inspection robot system, enabling intelligent decision-making and proactive defense.

[0010] Further, in step S1, the blind source separation algorithm is used to suppress background interference in the pipe gallery, which specifically includes: constructing a deep convolutional embedding module to map the high-dimensional time-frequency map of the mixed signal to a low-dimensional feature space; using a loss function with clustering constraints in the feature space, so that the network can automatically separate the features into subspaces belonging to different source signals such as cable partial discharge, background electromagnetic noise, and mechanical vibration noise without relying on prior labels, thereby reconstructing a pure partial discharge signal.

[0011] Furthermore, in step S2, the dual-channel spatiotemporal convolutional neural network uses one-dimensional dilated convolution to capture signal delay in the ultrasonic channel and two-dimensional convolution to process spectral-spatial features in the ultra-high frequency channel; the feature fusion layer uses a cross-attention mechanism to enable the ultrasonic feature map to focus on the potential region indicated by the ultra-high frequency feature map, and vice versa, thereby enhancing the ability to perceive partial discharge sources in weak signals and complex propagation environments.

[0012] Furthermore, in step S3, the acoustic and electromagnetic wave propagation simulation algorithm is a time-domain finite difference model generated based on the material library and geometric structure of the pipe gallery BIM model; the iterative optimization algorithm is a hybrid algorithm combining particle swarm optimization and the Levenberg-Marquardt method, used to quickly converge and obtain the optimal source coordinates that minimize the residual between the theoretical signal and the actual received signal.

[0013] Furthermore, in step S4, the transfer learning framework is used to first pre-train a general insulation state prediction model on a historical amplified dataset containing multiple cable types and multiple defect patterns; then, the last few layers of the model are fine-tuned using specific monitoring data of the current target pipe gallery to enable it to quickly adapt to the signal characteristics and evolution rhythm of the new environment, thereby achieving knowledge transfer and adaptation.

[0014] In addition, the present invention provides another system for implementing the above-described method, comprising: A multi-physical quantity fusion acquisition module is used for synchronous acquisition, alignment, and preprocessing of acoustic and electromagnetic wave signals; The intelligent initial positioning module, which embeds the dual-channel spatiotemporal convolutional neural network, is used to output preliminary logical coordinates; The digital twin correction module is used to load the BIM model, run propagation simulation and iterative optimization, and output accurate physical coordinates; The insulation state prediction module is used to run the hidden Markov model and transfer learning framework and output the risk prediction results. The visualization-based collaborative decision-making module is used to generate health cloud maps, operation and maintenance strategies, and trigger system linkage.

[0015] This application also provides a storage medium storing a computer program, wherein the computer program is configured to execute the method described in any of the preceding claims when running.

[0016] This application also provides an electronic device including a memory and a processor, wherein the memory stores a computer program and the processor is configured to run the computer program to perform the method described in any of the preceding claims.

[0017] In summary, the present invention has the following beneficial effects: 1. Strong anti-interference capability and accurate signal extraction: Through the fusion sensing of acoustic and electrical multi-physical quantities and advanced blind source separation technology, it can effectively remove the complex background noise of the pipe gallery, greatly improve the signal-to-noise ratio and recognizability of the partial discharge signal, and lay the foundation for subsequent accurate analysis. 2. High positioning accuracy: The initial positioning result based on deep learning is innovatively placed in a high-fidelity digital twin environment of the utility tunnel for wave propagation simulation correction, which fundamentally overcomes the positioning error caused by multipath effect and uneven attenuation, and achieves sub-meter level accurate positioning from "logical position" to "physical coordinates". 3. Possesses state prediction capabilities and enables early warning: It breaks through the limitations of traditional methods that only alarm but do not predict. By establishing a hidden Markov model of insulation state and integrating "partial discharge event fingerprint", it can dynamically predict the health degradation stage and remaining life of cable insulation, transforming the operation and maintenance mode from "post-event maintenance" to "predictive maintenance". 4. High level of system intelligence and linkage: The generated insulation risk cloud map and decision-making instructions can be directly and seamlessly connected with the BIM management platform, environmental control system and inspection robot of the smart utility tunnel to form a closed loop of "perception-diagnosis-prediction-decision-execution", which greatly improves the intelligence level of urban lifeline infrastructure operation and maintenance and emergency response efficiency. Attached Figure Description

[0018] Figure 1 This is a schematic diagram of the core three-stage framework of an embodiment of the present invention. Detailed Implementation

[0019] To make the technical problems to be solved, the technical solutions, and the beneficial effects of the present invention clearer, the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the present invention and are not intended to limit the present invention.

[0020] It should be noted that when a component is referred to as being "fixed to" or "set on" another component, it can be directly or indirectly attached to that other component. When a component is referred to as being "connected to" another component, it can be directly or indirectly connected to that other component.

[0021] It should be understood that the terms "length", "width", "upper", "lower", "front", "rear", "left", "right", "vertical", "horizontal", "top", "bottom", "inner", "outer", etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are only for the convenience of describing the present invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on the present invention.

[0022] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature. In the description of this invention, "a plurality of" means two or more, unless otherwise explicitly specified.

[0023] This embodiment uses a 5-kilometer-long integrated utility tunnel containing a high-voltage power compartment as the implementation scenario. The system hardware deployment follows an architecture of "sensor layer - transmission layer - platform layer," see [link to relevant documentation]. Figure 1 .

[0024] The sensing layer is deployed with: Distributed fiber optic ultrasonic sensing array: Utilizing corrosion-resistant and electromagnetic interference-resistant special optical fibers as the sensing medium. The optical fibers are tightly laid on the outer sheath of a 110kV cross-linked polyethylene cable in a helical winding manner, with a winding pitch of 30 cm, ensuring omnidirectional contact with the cable surface. A single continuous sensing optical cable is deployed along the entire cable.

[0025] UHF sensor array: Broadband UHF sensors are installed at cable joints, terminations, and every 200 meters. The sensors are encapsulated in waterproof and explosion-proof housings and connected via radio frequency coaxial cables. This embodiment deploys a total of 30 UHF sensor nodes.

[0026] Synchronization and Data Acquisition Unit: One main acquisition station is deployed at each end of the utility tunnel between the equipment rooms. All UHF sensor signals are fed into the acquisition station via coaxial cables. The acquisition station has a built-in high-precision GPS / BeiDou dual-mode clock synchronization module, providing nanosecond-level synchronization signals for all sensor channels. Fiber optic ultrasonic signals, after being demodulated by a demodulator, are also connected to the acquisition station for timestamp alignment.

[0027] The data acquisition station uploads massive amounts of raw data in real time to the edge computing server cluster located in the pipeline corridor monitoring center through the industrial ring network of the pipeline corridor.

[0028] The monitoring center deploys core analysis servers to run digital twin models, AI algorithms, and early warning decision-making programs.

[0029] The system achieves data exchange and command linkage with the existing BIM management platform, environmental monitoring system (EMS), and inspection robot scheduling system of the utility tunnel through a standard API interface. Step 1: Synchronize triggering and data collection. When partial discharge occurs in the cable, the system operates according to the following timing sequence: UHF channel triggers first: If any UHF sensor detects that the electromagnetic pulse amplitude exceeds the threshold (e.g., 50mV), it immediately sends a hardware trigger signal to the entire network.

[0030] Synchronous capture across all channels: Upon receiving the trigger signal, all data acquisition units synchronously retrieve complete waveform data within a 100-microsecond time window before and after the trigger moment. Although the fiber optic ultrasound channel responds slightly slower, its effective signal remains within the capture window due to the slower signal propagation speed.

[0031] Data packet encapsulation: Each event's data packet contains: a globally unique event ID, a precise UTC timestamp, and raw waveform data from all sensor channels.

[0032] Step 2: Adaptive noise reduction and blind source separation The core approach employs an improved deep convolutional embedded blind source separation network (DCEC-BSS). The processing flow is as follows: Input: Convert the 100-microsecond waveforms of 30 UHF channels and 1 fiber optic ultrasonic signal (which has been decomposed into multi-channel strain signals) into a time-frequency graph (e.g., through short-time Fourier transform) to form a set of multi-dimensional images.

[0033] Network Processing: The DCEC-BSS network first extracts deep features through convolutional layers, and then performs unsupervised clustering in the feature space. The network is trained to automatically cluster features into different "sources": Cluster 1 (Target Source): Pulse clusters with typical partial discharge characteristics (such as steep rising edges and clustering at specific frequencies).

[0034] Cluster 2 (electromagnetic interference sources): continuous or periodic interference such as 50Hz power frequency harmonics and inverter switching noise.

[0035] Cluster 3 (Mechanical noise source): Acoustic noise from the vibration of water pumps and fans in the pipe gallery that is propagated through the structure.

[0036] Output: The network outputs a clean time-frequency waveform containing only the target partial discharge signal, which is then inversely transformed into a time-domain waveform for subsequent analysis. This processing improves the signal-to-noise ratio by an average of over 35dB.

[0037] This embodiment uses a dual-channel neural network called "SEFusionNet" for initial localization.

[0038] Ultrasonic Channel: The input is the separated multi-channel ultrasonic signal. A one-dimensional dilated convolutional layer is used, with the dilation rate gradually increasing (1, 2, 4, 8) to expand the receptive field and accurately capture the time difference of the signal arriving at different sensors.

[0039] UHF channel: The input is the separated multi-sensor UHF signal spectrum-space matrix. A two-dimensional convolutional layer is used to extract the joint features of the signal in the spectrum dimension and the sensor spatial distribution dimension.

[0040] Cross-attention fusion module: This module allows the two channels to "ask each other questions".

[0041] The ultrasound machine asks the UHF machine: "Based on my time delay information, the fault point is roughly in this direction. Where are your spectral characteristics most prominent?" The UHF feature map will then amplify the response in that region accordingly.

[0042] UHF asks ultrasound: "My spectrum shows an abnormal discharge in this area. Does your signal delay support this?" The ultrasound feature map will verify and correct the delay calculation.

[0043] Output: The network ultimately outputs a "probability distribution heatmap" covering the entire length of the cable tunnel. The peak areas of the heatmap are marked as preliminary logical location results, such as "high-pressure chamber, K2+150 to K2+180 interval, probability 87%".

[0044] This step converts the "logical interval" into "physical coordinates (x, y, z)".

[0045] Digital twin environment construction: Import the Revit BIM model of the utility tunnel. The model contains geometric and material information of all compartment structures, precise locations of cable trays, other pipelines, ventilation ducts, and metal supports.

[0046] Propagation Simulation Engine: In the twin model, using the initial location output by SEFusionNet as the suspected source, initiate parallel simulation: Acoustic simulation: Based on the finite element method, calculate the path of ultrasonic pulses from the suspected source to each fiber optic sensing point, the change in sound velocity (affected by temperature and medium), and the reflection on the metal surface.

[0047] Electromagnetic wave simulation: Based on the finite-difference time-domain method, calculate the multipath propagation, attenuation and coupling effects of UHF electromagnetic waves in complex pipe gallery environments.

[0048] Iterative optimization positioning: The theoretical signal waveform obtained from the simulation is compared with the actual acquired signal waveform, and the residual is calculated.

[0049] The Levenberg-Marquardt algorithm is used to automatically adjust the coordinates of the suspected source in three-dimensional space (fine-tuning x, y, z) to minimize the residual between the simulated waveform and the real waveform.

[0050] After several iterations (usually 3-5 times), the algorithm converges when the residual is less than the threshold, outputting the optimal 3D coordinates. For example, the final positioning result is: (X: 1254.32m, Y: -3.15m, Z: altitude -8.72m), with an accuracy of ±0.5 meters.

[0051] Constructing a partial discharge event fingerprint: Each successfully located partial discharge event generates a "fingerprint" record, which is stored in the database. For example:{ "event_id": "PD20231027_142355_001", "location": {"x": 1254.32, "y": -3.15, "z": -8.72}, "timestamp": "2023-10-27T14:23:55.123Z", “features”: { “amplitude_pC”: 250, / / Apparent discharge level "dominant_freq_MHz": 312, / / Dominant frequency “repetition_rate_Hz”: 50, / / Repetition rate “phase_angle_deg”: 45, / / Power frequency phase "waveform_similarity": 0.82 / / Similarity to typical defect waveforms } } Step 2: Hidden Markov Model Training and Prediction State definition: Insulation state is divided into four categories: S0 healthy, S1 early defect (such as micropores inside the insulation), S2 developing defect (such as electrical treeing), and S3 near failure (severe damage to the main insulation).

[0052] Model training: The Hidden Markov Model (HMM) is trained using historical data. For example, the sequence of "event fingerprints" of a cable that eventually fails is used to learn the state transition probabilities (such as the probability of transitioning from S1 to S2).

[0053] Real-time prediction: The system runs a prediction every 24 hours. The "event fingerprint" sequence of the current cable over the past week is input into the trained HMM.

[0054] The model will output the most likely insulation state at present, for example: P(S0)=10%, P(S1)=20%, P(S2)=65%, P(S3)=5%, and determine that it is currently in the "developing defect" stage.

[0055] The Viterbi algorithm is used to predict the most likely state evolution path over the next 30 days and calculate the remaining electrical lifetime index, for example, 78, which means that the relative remaining lifetime is 78%.

[0056] On the utility tunnel BIM operation and maintenance platform, the cable model is rendered with colors (green → yellow → orange → red) based on the predicted insulation state probability. When a user clicks on any cable segment, a details panel pops up, displaying historical partial discharge events, current state probability, predicted trend curve, and remaining life index.

[0057] Intelligent decision-making and collaboration: The system has a built-in rules engine that automatically triggers workflows based on risk levels. Risk Level I (Monitoring): Status is S1. The system automatically increases the monitoring frequency of this section and provides a prompt on the BIM platform.

[0058] Risk Level II (Warning): Status S2. The system automatically generates a warning work order and pushes it to the mobile terminal of maintenance personnel. At the same time, it links with the environmental control system to increase the ventilation volume in this section to reduce ambient humidity and suppress discharge.

[0059] Risk Level III (Alarm): Status is S3 or the remaining life index drops sharply. The system triggers the highest-level alarm and automatically dispatches an inspection robot along the optimal path to the precise coordinate point for on-site verification using high-definition cameras and infrared thermal imagers. Simultaneously, it generates a maintenance plan package for maintenance personnel, including location information, risk analysis, handling suggestions, and a spare parts list.

[0060] During the three-month trial operation of this embodiment, the system successfully issued early warnings for two early cable joint insulation defects, and the positioning accuracy was verified to be less than 0.8 meters during excavation. This reduces the time for handling potential faults from the traditional "repair within several hours after a fault" to "handling within a two-hour window of planned power outages," greatly improving the safety of power supply to the utility tunnel and the economic efficiency of operation and maintenance.

[0061] Another embodiment of the present invention provides a system for implementing the above-described method, comprising: A multi-physical quantity fusion acquisition module is used for synchronous acquisition, alignment, and preprocessing of acoustic and electromagnetic wave signals; The intelligent initial positioning module, which embeds the dual-channel spatiotemporal convolutional neural network, is used to output preliminary logical coordinates; The digital twin correction module is used to load the BIM model, run propagation simulation and iterative optimization, and output accurate physical coordinates; The insulation state prediction module is used to run the hidden Markov model and transfer learning framework and output the risk prediction results. The visualization-based collaborative decision-making module is used to generate health cloud maps, operation and maintenance strategies, and trigger system linkage.

[0062] This invention also provides a storage medium storing a computer program, wherein the computer program is configured to execute the steps in any of the above method embodiments when it is run.

[0063] This invention also provides an electronic device, including a memory and a processor, wherein the memory stores a computer program, and the processor is configured to run the computer program to perform the steps in any of the above method embodiments.

[0064] Specifically, the aforementioned electronic device may further include a transmission device and an input / output device, wherein the transmission device is connected to the aforementioned processor, and the input / output device is connected to the aforementioned processor.

[0065] The above embodiments are merely explanations of the present invention and are not intended to limit the present invention. After reading this specification, those skilled in the art can make modifications to these embodiments without contributing any inventive step, but as long as they are within the scope of the claims of the present invention, they are protected by patent law.

Claims

1. A method for intelligent location of partial discharge and early warning of insulation degradation in cables, characterized in that, include: By using a distributed fiber optic ultrasonic sensing array and an ultra-high frequency electromagnetic sensing array, the acoustic and electromagnetic signals of partial discharge in the cable are simultaneously acquired, and signal alignment, noise reduction and interference separation preprocessing are performed to extract a clean partial discharge pulse sequence. The preprocessed acoustic-electric fusion signal is input into a dual-channel spatiotemporal convolutional neural network to extract and fuse features, generating a preliminary set of candidate logical coordinates for the partial discharge source. Based on the high-precision BIM digital twin model of the utility tunnel, environmental propagation effect compensation is performed on the preliminary logical coordinates through acoustic-electromagnetic wave propagation simulation and iterative optimization algorithm, and accurate three-dimensional physical coordinates are output. A hidden Markov model with cable insulation degradation stages as hidden states is constructed. The "event fingerprint" formed by historical and real-time partial discharge events and their characteristics is used as the observation sequence. Through transfer learning, the evolution probability of each insulation state and the remaining electrical lifetime within a preset time period are predicted. The insulation risk prediction results are mapped to a digital twin model to generate a health cloud map. Based on the dynamic priority algorithm, differentiated operation and maintenance strategies are generated, and linkage instructions with the utility tunnel environmental control and inspection system are triggered.

2. The method for intelligent location of partial discharge and early warning of insulation degradation in cables according to claim 1, characterized in that, The interference separation preprocessing specifically includes: using a blind source separation algorithm based on deep convolutional embedding and clustering constraints to separate the mixed signal in the feature space and reconstruct the source signal from the cable partial discharge.

3. The method for intelligent location of partial discharge and early warning of insulation degradation in cables according to claim 1, characterized in that, The dual-channel spatiotemporal convolutional neural network includes a one-dimensional dilated convolutional channel for processing ultrasonic signals, a two-dimensional convolutional channel for processing ultra-high frequency signals, and a feature fusion layer employing a cross-attention mechanism.

4. The method for intelligent location of partial discharge and early warning of insulation degradation in cables according to claim 1, characterized in that, The coordinate correction based on the high-precision BIM digital twin model of the utility tunnel specifically includes: establishing a time-domain finite-difference propagation model in the digital twin environment, and using a hybrid iterative algorithm of particle swarm optimization and the Levenberg-Marquardt method to minimize the path residual between the theoretical signal and the actual received signal.

5. The method for intelligent location of partial discharge and early warning of insulation degradation in cables according to claim 1, characterized in that, The method of predicting insulation state evolution through transfer learning specifically includes: pre-training a general prediction model using a large historical dataset, and then fine-tuning the model using specific data from the target utility tunnel to achieve cross-scenario transfer and adaptation of knowledge.

6. A system, characterized in that, Including the method based on any one of claims 1-5: A multi-physical quantity fusion acquisition module is used for synchronous acquisition, alignment, and preprocessing of acoustic and electromagnetic wave signals; The intelligent initial positioning module, which embeds a dual-channel spatiotemporal convolutional neural network, is used to output preliminary logical coordinates; The digital twin correction module is used to load the BIM model, run propagation simulation and iterative optimization, and output accurate physical coordinates; The insulation state prediction module is used to run hidden Markov models and transfer learning frameworks and output risk prediction results. The visualization-based collaborative decision-making module is used to generate health cloud maps, operation and maintenance strategies, and trigger system linkage.

7. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the program is executed by the processor, it implements the method as described in any one of claims 1-5.

8. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the method as described in any one of claims 1-5.

Images