Systematic monitoring method for electrical safety performance of cable

Abstract

The invention discloses a systematic monitoring method for the electrical safety performance of a cable, and relates to the field of electrical safety, and the method comprises the steps: collecting sensor data to an edge calculation node, preprocessing the data, and outputting the data as real-time calculation temperature data, local discharge point data, resistance and capacitance data and vibration data; the data are sent to a hidden danger classification model and a fault positioning model in the cloud, and the models output early warning information to an operation and maintenance project for processing; the sensor data sending process is judged according to the data frequency, when it is judged that the frequency is larger than 1 kHz, a 1D-CNN electric arc detection model is loaded, when it is judged that the frequency is smaller than 1 kHz, a Light GBM feature classification model is loaded, and 24-hour sensor data smaller than 1 kHz are periodically loaded into an LSTM-Attention gradient analysis model; and related data of partial discharge generated by judgment of the 1D-CNN arc detection model is sent to an LSTM-Attention gradient analysis model for evaluating the partial discharge trend of the global cable. Classification model output is used as prior knowledge of a positioning model, and the positioning accuracy can be improved.

Description

Technical Field

[0001] The present invention relates to the field of electrical safety, and particularly to a systematic monitoring method for the electrical safety performance of cables. Background Art

[0002] At present, the electrical safety control of cables still faces severe challenges. Problems such as cable insulation damage, aging and peeling, mechanical damage, uneven construction quality, and the prevalence of counterfeit and shoddy products are relatively prone to occur, and related electrical fire accidents occur frequently and at a high rate. In addition, with the continuous development of the urbanization process, affected by objective factors such as dense cable laying, the influx of non-standard cables into the market secretly, and insufficient maintenance of old lines, the electrical safety risks have increased further. However, the application of new technologies has not been popularized, resulting in difficulties in timely discovery of potential hazards. In addition, the ambiguity of the responsible entity (such as no clear responsible party for the maintenance of the line after the residential electricity meter) further exacerbates the management blind area, and a systematic solution is urgently needed.

[0003] At present, several preventive measures for the electrical safety of cables cannot form a systematic monitoring and positioning method, and how to systematically judge the specific type and location is an urgent problem to be solved. Summary of the Invention

[0004] The present invention provides a systematic monitoring method for the electrical safety performance of cables to solve the problems of the prior art.

[0005] In a first aspect, the present invention provides a systematic monitoring method for the electrical safety performance of cables, including: Collect sensor data to an edge computing node to preprocess the data and output real-time calculated temperature data, partial discharge data, resistance and capacitance data, and vibration data; Send the data to a hidden danger classification model and a fault location model in the cloud, and the model outputs a warning message for operation and maintenance engineering processing; Among them, the process of sending sensor data is judged according to the data frequency. When the judgment is greater than 1 kHz, the 1D-CNN arc detection model is loaded. When the judgment is less than 1 kHz, the LightGBM feature classification model is loaded, and the 24-hour sensor data less than 1 kHz is periodically loaded into the LSTM-Attention gradual change analysis model; Among them, the relevant data of partial discharge judged by the 1D-CNN arc detection model is sent to LSTM-Attention for training, and the trained LSTM-Attention is used to evaluate the partial discharge trend of the global cable.

[0006] Further, the fault location model includes: a cable network diagram structure model. Sensors are set as sensor nodes, and each node is set with node features such as temperature, current, humidity, and partial discharge. Cable segments are set between sensor nodes. Each cable segment and sensor node are numbered, and cable segment features such as cable length, cross-sectional area, laying depth, and topological distance are set. An adjacency matrix of the cable segments and sensor nodes is constructed based on Gaussian kernel weighting for message passing. The message passing mechanism is to splice the cable segment features and the features of adjacent sensor nodes to form a more message function. The message function is updated in real time based on the maximum value and total value of the failure rate based on node positions. Message functions are passed between sensor nodes at different levels. The numbers of the same sensor node and cable segment under different branches are totaled as one level. Multiple levels distinguish the trunk-branch relationship and form different connection layers from large to small. The message function is passed and updated between the global fully connected layers. Each connection layer is set with a failure probability, and the failure probability is determined by the output data of the hidden danger classification model.

[0007] Further, the failure probability is the maximum value of the failure probabilities of multiple sensor nodes under the same connection layer, and the numbered position data of the sensor nodes at the level under the connection layer is located based on the failure probability.

[0008] Further, the sensor data for training the hidden danger classification model also includes failure data generated by finite element simulation based on the COMSOL Multiphysics software. The generation steps are as follows: Step 1: Geometrically construct a cable hierarchical structure: conductor, insulation layer, shielding layer, sheath, and joint components; Step 2: Set the materials and corresponding properties of the cable hierarchical structure; Step 3: Construct the coupling equations of the electrical field, thermal field, and structural mechanics field of the cable hierarchical structure; Step 4: Import the failure scenario to obtain the simulation implementations of insulation aging, joint loosening, overload operation, and moisture intrusion; Step 5: Set boundary conditions for the electrical field, thermal field, and structural mechanics field respectively. Apply the rated current / voltage at the conductor end of the electrical field, convective heat dissipation on the outer surface of the cable in the thermal field, forced air cooling at the joint, and set the fixed cable end constraint displacement in the structural mechanics field; Step 6: Generate supplementary failure data for training the hidden danger classification model.

[0009] Further, the sensor data is collected to the edge computing node for preprocessing the data and outputting real-time calculated temperature data, partial discharge data, resistance and capacitance data, and vibration data. Among them, the data preprocessing for the LSTM-Attention gradient analysis model specifically includes: Normalize the temperature data using a sliding window, where the processing window is a multiple of 24 hours, with the multiple being at least 1 time. Calculate the correlation coefficient between the current data and the temperature. Extract the pulse count / amplitude mean of the partial discharge data.

[0010] Furthermore, the LSTM-Attention gradual analysis model accesses the preprocessed data and encodes the temporal dependence. After adding the Attention mechanism to focus on the critical period, it cooperates with the fault location model to classify and output the potential fault hazards and corresponding location.

[0011] Furthermore, it also includes that the LSTM-Attention gradual analysis model is pre-trained by accessing the fault data generated by finite element simulation, and then accesses the real sensor data and the fault data generated by finite element simulation mixed in proportion as the training set for training. After correcting the training results of the LSTM-Attention gradual analysis model, reduce the proportion of the fault data generated by finite element simulation for training and correction until the proportion of the fault data generated by finite element simulation drops to 0. When the training meets the iterative convergence condition, the training of the LSTM-Attention gradual analysis model is completed.

[0012] Furthermore, the process of the 1D-CNN arc detection model for potential hazard discrimination is as follows: The data types accessed by the 1D-CNN arc detection model for preprocessing include partial discharge data, vibration signals, and temperature mutations. The preprocessing method for partial discharge data is band-pass filtering and pulse alignment after sampling at 200 MHz, with an input dimension of 2000 points / pulse. The preprocessing method for vibration signals is to convert the STFT time-frequency diagram into a 1D vector, with an input dimension of 512 points / frequency axis. The processing method for temperature mutation signals is differential processing, with an input dimension of 120 points / 10-minute window. Substitute the processed data into the 1D-CNN arc detection model for multi-scale convolution feature extraction, and execute the feature fusion strategy for the data types involving multi-sensor data, specifically including fusing the features of PD pulses and vibration signals, and cooperating with the fault location model to output the arc, vibration position, and duration of the cable joint.

[0013] Furthermore, the LightGBM feature classification model is used to load sensor data below 1 kHz for monitoring the aging / moisture abnormal state of the cable, and label the corresponding data with the aging / moisture abnormal state label, specifically including: First, align the timestamps of different sensor data and normalize them to the same interval to eliminate the dimension. After preliminary annotation using unsupervised methods, construct the original sensor data types and corresponding basic statistical features. The basic statistical features include mean, variance, maximum value, and minimum value. Construct advanced statistical features including: time window features: statistics of different time windows (such as 1 hour, 6 hours, 24 hours); correlation features: such as the covariance between temperature and humidity, and the correlation coefficient between current and temperature; trend features: such as the slope of linear regression (reflecting the temperature rise trend); periodic features: such as the mean value at the same moment within a day (eliminating the influence of day and night); physical model features: constructed according to the physical characteristics of the cable, such as: insulation resistance degradation rate: (current resistance - initial resistance) / operating time, capacitance change rate: (current capacitance - initial capacitance) / operating time; Set the importance level of the features, and then train the model based on the three state labels of normal state, aging state, and humid state, and label the data.

[0014] Furthermore, it also includes using a GNN aggregation model to embed the sensor node vectors, loading the processed high-frequency data into a 1D-CNN arc detection model for analysis, and loading the processed low-frequency data into a LightGBM feature classification model for analysis.

[0015] A systematic monitoring method for the electrical safety performance of cables provided by the present invention uses the output of the classification model as the prior knowledge of the positioning model, which can improve the positioning accuracy; The present invention calculates basic features such as temperature gradient and current mutation in real time, and can trigger emergency power-off within 5 ms; The present invention uses finite element simulation to generate fault data, which can accurately control the fault type, location, and degree, expand the recognition breadth of the model, and reduce the false alarm rate. Description of the Drawings

[0016] The drawings described herein are used to provide a further understanding of the embodiments of the present invention, form a part of the present invention, and do not limit the embodiments of the present invention. In the drawings: Figure 1 It is the GNN aggregation data classification flowchart provided by Embodiment 1 of the present invention.

[0017] Figure 2 It is the cascade-fusion flowchart involving the LightGBM feature classification model, 1-CNN arc detection model, and LSTM-Attention gradient analysis model provided by Embodiment 1 of the present invention. Detailed Embodiment

[0018] Here, exemplary embodiments will be described in detail, and examples thereof are shown in the accompanying drawings. When the following description refers to the accompanying drawings, unless otherwise indicated, the same numbers in different drawings represent the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with the present invention.

[0019] First, the terms related to the present invention are explained: Classification of the main hidden dangers related to the electrical safety of cables: 1. Exposed live parts: Hidden danger: The conductor is directly exposed due to damage, aging and cracking of the cable insulation layer, improper joint treatment (such as falling off of insulating tape, damage to the joint box).

[0020] Risk: Personnel directly come into contact with the exposed conductor, resulting in an electric shock accident.

[0021] Common scenarios: Old lines, construction damage, animal gnawing, mechanical external damage.

[0022] 2. Insulation failure: Hidden danger: Aging: Long-term use, high temperature, ultraviolet radiation lead to a decline in the performance of insulating materials (becoming brittle, cracking, reducing resistance). Inferior materials: Using cables or insulating materials that do not meet the standards. Environmental erosion: Moisture, corrosive gases / liquids, oil erosion of the insulation layer. Overvoltage: Lightning strike or switching overvoltage breaks down the insulation.

[0023] Risk: Insulation failure makes the originally non-live metal sheath, armor layer or adjacent conductor live, or causes a short circuit between conductors to generate an arc, increasing the risk of indirect electric shock or arc burn.

[0024] 3. Mechanical damage: Hidden danger: The outer sheath or insulation layer of the cable is damaged due to construction digging, heavy object extrusion, sharp object scratching, excessive bending, vehicle rolling, animal gnawing, etc.

[0025] Risk: It directly causes the conductor to be exposed or the insulation performance to decline, leading to electric shock or short circuit.

[0026] 4. Damp or waterlogged environment: Hidden danger: The cable trench, pipeline, joint box get water, or the cable is in a damp environment for a long time.

[0027] Risk: Moisture reduces the insulation resistance, may cause an increase in leakage current, and even forms a conductive path along the damp surface, increasing the risk of indirect electric shock (especially for metal sheath or armored cables). A live cable in water is extremely dangerous.

[0028] 5. Overheating: Hidden dangers: Overload: The current exceeds the safe current-carrying capacity of the cable. Poor heat dissipation: Heat accumulation occurs due to dense cable laying, too thin conduit, too thick burial depth, being covered by thermal insulation materials, etc. Poor contact: Loose connections at joints and terminals, resulting in increased resistance and heat generation.

[0029] Risks: High temperature accelerates the aging and failure of insulation, and may even ignite surrounding combustibles, causing a fire. After insulation failure, the risk of electric shock increases.

[0030] 6. Poor or missing grounding or equipotential bonding: Hidden dangers: The metal cable sheath, armor layer, and cable tray are not reliably grounded or the grounding resistance is too large. The exposed conductive parts of electrical equipment are not reliably connected to the protective conductor (PE). Local equipotential bonding is not done in special places (such as bathrooms, swimming pools).

[0031] Risks: When the equipment insulation fails and the shell becomes energized, the fault current cannot be effectively introduced into the ground, and the touch voltage may reach a dangerous level, leading to serious indirect electric shock accidents.

[0032] 7. Failure or absence of protection devices: Hidden dangers: Residual current operated protective device (RCD / leakage protector) is not installed or installed unreasonably. The circuit breaker and fuse are improperly selected or ineffective, and cannot cut off the power supply in time in case of short circuit or overload.

[0033] Risks: In case of leakage, short circuit, or overload, the fault circuit cannot be cut off in time, and the dangerous state persists.

[0034] 8. Improper design, installation, and maintenance: Hidden dangers: Incorrect cable selection (voltage level, current-carrying capacity, environmental adaptability do not match). Improper laying method (such as too small bending radius, excessive stretching, insufficient distance from heat source). Poor joint manufacturing process. Lack of regular inspection, testing, and maintenance.

[0035] Risks: Laying long-term safety hazards and increasing the probability of various faults.

[0036] 9. Insufficient safety distance: Hidden dangers: The distance between the cable and the area accessible to personnel is insufficient, or the electrical clearance and creepage distance from non-insulated live parts are not enough (especially in high-voltage cases).

[0037] Risks: Personnel accidentally touch or touch the live part through tools or metal objects, or air breakdown or surface flashover occurs in high humidity and polluted environments.

[0038] The present invention classifies the above 9 types into insulation aging, joint overheating, moisture intrusion, and mechanical damage according to the characteristic type, constructs characteristic indicators corresponding to the characteristic classification, and forms a three-in-one protection of upgrading the original system threshold to early prediction of potential hazards - precise classification - spatial positioning. For the rest of the safety distance, design, installation, and maintenance aspects, the installation construction and maintenance specifications are used to correct the improper behaviors of the staff and prevent electric shock accidents from occurring.

[0039] The specific application scenarios of the present invention are cable electrical safety protection scenarios in industrial, public, and household multi-scenarios.

[0040] A systematic monitoring method for the electrical safety performance of cables provided by the present invention aims to solve the above technical problems of the prior art.

[0041] The technical solution of the present invention and how the technical solution of the present invention solves the above technical problems will be described in detail below with specific embodiments. The embodiments of the present invention will be described below in conjunction with the accompanying drawings.

[0042] Embodiment 1: I. Deploy sensors and edge computing gateways: The potential hazard classification model is equipped with physical sensor types including high-precision Hall sensors and CT coils for monitoring current and voltage, which are deployed at the beginning and end of each section of the cable to monitor load fluctuations and harmonic distortions. Distributed optical fibers (DTS) and wireless temperature measurement tags for monitoring temperature are deployed at intervals of ≤5 meters at the monitored joints and cable bends. Capacitive humidity sensors for monitoring humidity are deployed inside cable trenches / wells and inside joint boxes (when the humidity > 80%, the insulation resistance drops suddenly). High-frequency HFCT / UHF sensors for monitoring partial discharge are used to detect discharge pulses ≥5 pC (early signs of insulation deterioration). MEMS accelerometers for monitoring vibration data are deployed above the directly buried section of the cable and the section laid on the bridge. An edge computing gateway is also deployed for edge computing and data processing.

[0043] II. Simulate and generate fault data with labels; Before collecting sensor data to the edge computing node and preprocessing the data to output real-time calculated temperature data, partial discharge data, resistance-capacitance data, and vibration data, since the real cable fault samples of the present invention are relatively few and do not meet the data requirements for systematically constructing the potential hazard classification model of the present invention, the present invention constructs an accurate mathematical model of the cable in a virtual environment. By applying different fault conditions (such as insulation breakage, overload, moisture ingress, etc.), the response of the cable under the action of multiple physical fields such as electricity, heat, and mechanical force is simulated, and the fault type, location, and degree can be accurately controlled, so as to batch generate "fault data" with labels. The principle and steps of generating fault data based on finite element simulation using COMSOL Multiphysics software are as follows: Step 1: Geometrically construct the cable hierarchical structure: conductor, insulation layer, shielding layer, sheath, and joint components; Step 2: Set the materials and corresponding properties of the cable hierarchical structure, specifically including setting the conductivity and thermal expansion coefficient of the copper conductor, the thermal conductivity, dielectric constant, and loss tangent value of the XLPE insulation layer; the non-linear heat conduction characteristics of the silicone rubber joint; Step 3: Construct the coupled equations of the electrical field, thermal field, and structural mechanics field; For the electrical field, solve the current continuity equation; for the thermal field, solve the heat conduction equation; for the structural mechanics field, solve the thermal expansion deformation equation; Step 4: Import the fault scenarios to obtain the simulation implementations of four different fault types. The simulation implementation of insulation aging is to reduce the conductivity of the insulation layer / increase the tanδ value, or add air gap defects; the simulation implementation of joint loosening is to increase the crimping contact resistance; the simulation implementation of overload operation is to increase the conductor current density to 150% - 200% of the design value, and the simulation implementation of moisture intrusion is to add moisture at the insulation layer defect.

[0044] Step 5: Set the boundary conditions corresponding to the electrical field, thermal field, and structural mechanics field respectively. Apply the rated current / voltage at the conductor end of the electrical field, convective heat dissipation on the outer surface of the cable in the thermal field, forced air cooling at the joint, and set the fixed displacement constraints at both ends of the cable in the structural mechanics field; Step 6: Generate an available machine learning dataset, which is used to train the hidden danger classification model; It also includes mixing the real data and the obtained simulation data in a certain proportion to train the hidden danger classification model.

[0045] Based on Steps 1 - 6, generate the partial discharge simulation data instances as follows: Generate data based on the above partial discharge (PD) simulation; Model in COMSOL: Set a bubble defect with a diameter of 0.1 mm in the insulation layer; Apply a 10 kV AC voltage; Output results: Discharge pulse waveform (amplitude, phase, frequency); Spatial electromagnetic field distribution; # Generate pulse sequence samples pulses = [] for phase in [30°, 90°, 270°]: # Discharge phase within the power frequency cycle pulse = simulate_pd(voltage = 10e3, defect_size = 0.1e - 3, phase) pulses.append(pulse) III. Collecting real-time sensor data: Collect real-time sensor data to the edge computing node for preprocessing the data, specifically including: The edge computing gateway processes and obtains the following feature types: Feature types corresponding to overheating of the corresponding joint: temperature rise rate (ΔT / Δt), three-phase temperature difference imbalance, hot spot diffusion gradient of the thermal image; Feature types corresponding to moisture intrusion: humidity-insulation resistance correlation coefficient, capacitance increment ratio; Feature types corresponding to insulation aging: harmonic components of leakage current, tangent value of dielectric loss angle (tanδ), partial discharge amplitude / frequency; Feature types corresponding to mechanical damage: vibration spectrum energy entropy, abnormal shock wave detection data.

[0046] IV. The cloud model receives data and outputs warning information: Output the preprocessed data to the cloud platform analysis engine for processing. The cloud platform analysis engine is loaded with a hidden danger classification model and a fault location model, and outputs warning information to the warning dashboard. According to the warning dashboard, issue an operation and maintenance work order system to process the warning information, specifically including: Among them, the hidden danger classification models constructed in the present invention respectively correspond to different data application scenarios. Among them, the LightGBM feature classification model is applicable to medium-scale data. The medium-scale data is specifically of the order of 10^4 sample quantities. The advantage of adopting the LightGBM feature classification model is that the selected features have high visualization degree and strong interpretability.

[0047] The 1D-CNN arc detection model is used to detect partial discharge pulse waveforms and vibration spectrum data, corresponding to partial discharge data generated by insulation aging and moisture intrusion and vibration spectrum data generated by mechanical damage respectively. The 1D-CNN arc detection model can automatically extract spatio-temporal features in the data; The LSTM-Attention gradual change analysis model is used to monitor temperature and current data under long time series, and can capture gradual change type hidden dangers based on the long-term memory dependence relationship; The specific analysis of each model in the cloud is as follows: In the LSTM-Attention gradual change analysis model of the present invention, regarding the key event focus of Attention: assign weights to the LSTM outputs of each time step, and set a scoring function. In the model, abnormal inflection points will be automatically amplified, such as the weights of temperature mutation + humidity sudden increase, and the data in the stable period will be weakened, which is convenient for intuitive display and has high readability. Map the Attention weights of each monitoring point to the cable network graph structure model summary, and output the located sensor nodes and abnormal time periods.

[0048] Regarding the 1D-CNN arc detection model, the input data is preprocessed as follows: The preprocessing method for partial discharge data is band-pass filtering pulse alignment after 200MHz sampling, and the input dimension is 2000 points / pulse; the preprocessing method for vibration signals is to convert the STFT time-frequency diagram into a 1D vector, and the input dimension is 512 points / frequency axis; the processing method for temperature mutation signals is differential processing T t -T t-1 , and the input dimension is 120 points / 10-minute window; As Figure 2 shown, it involves the cascade-fusion of the LightGBM feature classification model, the 1-CNN arc detection model, and the LSTM-Attention gradual change analysis model. The discrimination in terms of data frequency is implemented in the cloud as follows: When the sampling frequency is greater than 1kHz, the original waveform data is sent into the 1D-CNN arc detection model; When the sampling frequency is less than 1kHz, the feature vector is extracted and then sent into the LightGBM feature classification model; A 24-hour data buffer is established to trigger the LSTM-Attention gradual change analysis periodically; The special processing for high-frequency data is as follows: After the 1D-CNN model recognizes the partial discharge feature: Trigger a real-time alarm; Upload the discharge feature data to the cloud LSTM training queue; The partial discharge data is used to train the global trend model to improve the system's prediction ability for slow discharge development; The process of line-of-sight positioning of the fault location model based on the graph neural network is as follows: 1. Cable network modeling: Abstract the sensors as nodes, and the features include: temperature, current, humidity, partial discharge amount; Abstract the cable segments as edges, and the features include: length, cross-sectional area, laying depth, topological distance; Divide the levels according to the physical topology: The main line is the high level (Level 0); The branch lines are degraded in turn (Level 1-N); Each node records its own level and parent node; 2. Hierarchical message passing: Message generation: The node fuses its own features with the features of the connected cable to form a message packet; Intra-level transmission: Nodes at the same level exchange current status information; Cross-level transmission: The child node uploads the aggregated information to the parent node; The parent node broadcasts the global state downward; Fault probability calculation: Update the node state vector by passing in each round; Finally, output the fault probability values of each node; 3. Location decision-making process: Hierarchical fault probability = the maximum value of the probabilities of all nodes in this hierarchy; After determining the highest probability hierarchy, within this hierarchy: Filter out the faulty nodes exceeding the threshold; Sort by probability value to generate candidate positions; The output includes: fault hierarchy, node ID list, location coordinates, probability distribution.

[0049] Among them, the cascaded pipeline hybrid model of the LightGBM feature classification model and the 1D-CNN arc detection model is used to cooperate to label the data; The GNN model is also loaded, which is used to process multi-sensor nodes and aggregate and split the data to the LightGBM feature classification model and the 1D-CNN arc detection model; Among them, the data processing process of the 1D-CNN arc detection model for multiple feature dimensions is as follows: Execute multi-scale convolution feature extraction; Execute the feature fusion strategy: When two feature dimensions are involved: Process the PD pulse feature pd_branch and the vibration signal feature vib_branch separately, and then fuse the two types of features using the formula: merged = Concatenate()([Flatten()(pd_branch), vib_branch]) According to the time series dimension, after feature classification, locate the data and find the abnormal probability at each time point; When three feature dimensions are involved, the procedure is as follows: import torch import torch.nn as nn class CableFaultCNN(nn.Module): def __init__(self, in_channels=3): # 3 kinds of sensors super().__init__() # Branch 1: Large kernel captures long-term features self.branch1 = nn.Sequential( nn.Conv1d(in_channels, 32, kernel_size=51, stride=3,padding=25), nn.BatchNorm1d(32), nn.ReLU(), nn.MaxPool1d(3, stride=2) ) # Branch 2: Medium kernel captures local features self.branch2 = nn.Sequential( nn.Conv1d(in_channels, 64, kernel_size=25, padding=12), nn.BatchNorm1d(64), nn.ReLU(), nn.MaxPool1d(3, stride=2) ) # Branch 3: Small kernel captures details self.branch3 = nn.Sequential( nn.Conv1d(in_channels, 64, kernel_size=9, padding=4), nn.BatchNorm1d(64), nn.ReLU() ) # Feature fusion self.fusion = nn.Sequential( nn.Conv1d(32+64+64, 128, kernel_size=3, padding=1), nn.ReLU(), nn.AdaptiveMaxPool1d(1) ) self.classifier = nn.Linear(128, 4) # 4 types of potential hazards def forward(self, x): b1 = self.branch1(x) b2 = self.branch2(x) b3 = self.branch3(x) # Unify the time dimension b1 = nn.functional.interpolate(b1, size=b3.size(2), mode='linear') b2 = nn.functional.interpolate(b2, size=b3.size(2), mode='linear') fused = torch.cat([b1, b2, b3],dim=1) fused = self.fusion(fused).squeeze(2) return self.classifier(fused) The process of handling labels for the cascaded pipeline hybrid model of the LightGBM feature classification model and the 1D-CNN arc detection model is as follows: Classify the original sensor sampling data according to the data frequency. The corresponding LightGBM feature classification model is used to load sensor data less than 1 kHz for monitoring the aging / moisture abnormal state of the cable, and the corresponding data is labeled with the aging / moisture abnormal state, specifically including: First, align the timestamps of different sensor data and normalize them to the same interval to eliminate the dimension. After preliminary annotation using an unsupervised method, construct the original sensor data type and corresponding basic statistical features. The basic statistical features include mean, variance, maximum value, and minimum value; construct advanced statistical features including: time window features; correlation features; trend features; periodic features; physical model features, set the importance of the features, and then train the model according to the three state labels of normal state, aging state, and moisture state, and label the data.

[0050] The processing process of GNN aggregating data for the data aggregation of three types of sensor nodes is as follows: As Figure 1 shown in the GNN aggregated data classification flowchart, use the GNN aggregation model to embed the sensor node vectors, load the processed high-frequency data into the 1D-CNN arc detection model for analysis, and load the processed low-frequency data into the LightGBM feature classification model for analysis; In the project, the input data: Low frequency: joint temperature (1 Hz), ambient humidity (0.2 Hz) High frequency: ultrasonic signal (200 kHz) Processing flow: LightGBM branch: Features: 10 - minute temperature rise slope = 0.8 °C / min, humidity change rate > 5% / min Output: Risk of moisture intrusion (probability 87%) 1D - CNN branch: # Ultrasonic signal feature extraction cnn_feat = model_cnn(ultrasonic_wave) # Output: [0.02, 0.91, 0.03] → Micro - discharge Among them, cnn_feat is the CNN Features, representing the feature vector extracted by CNN. In the present invention, it means the feature tensor output by the 1D - CNN model, characterizing the fault mode in the ultrasonic signal. model_cnn is the CNN Model, which is the CNN model in the present invention, a trained 1D convolutional neural network model (for processing ultrasonic waveforms). The ultrasonic_wave in the parentheses is the Ultrasonic Wave Signal, representing the ultrasonic signal. In the present invention, it means the original ultrasonic waveform data collected from the sensor (represented in the time domain or frequency domain); Output: Partial discharge (probability 91%) Decision fusion: Pfault = 1−(1− PGBM )×(1− PCNN ) = 1−0.13×0.09 = 98.8%; Pfault is the probability of at least one fault existing, PCNN is the partial discharge probability output by the 1D - CNN model; PGBM is the moisture intrusion probability output by the LightGBM model; Conclusion: There is a composite fault of moisture + discharge in the joint.

[0051] The following is the performance comparison experiment result table: 2. Special processing of high - frequency data; After the 1D - CNN model identifies the partial discharge feature: Trigger real - time alarm; Upload the discharge feature data to the cloud LSTM training queue; The partial discharge data is used to train the global trend model to improve the system's prediction ability for slow discharge development; II. Cooperative work of hidden danger classification models 1. Multi - model division of labor mechanism 1D-CNN model: focuses on processing high-frequency arc characteristics and identifying microsecond-level transient faults; LightGBM model: Analyzes low-frequency characteristics (temperature rise trend, leakage current harmonics, etc.) to identify gradual problems such as insulation aging; LSTM-Attention model: Periodic analysis of 24-hour data windows; Catching slow-developing hazards (such as insulation degradation caused by moisture intrusion); Receive partial discharge data identified by CNN and continuously optimize discharge trend prediction; 2. Model result fusion Each model outputs a hidden danger type label with confidence level; Dynamic weighted fusion strategy: Arc faults give higher weight to CNN results; Gradual faults focus on LSTM analysis results; The final output includes: hidden danger type, severity level, and development trend forecast.

[0052] 3. Graph Neural Network Positioning Implementation 1. Cable network modeling: The sensor is abstracted as a node, and its features include: temperature, current, humidity, and partial discharge; Abstract the cable segment into an edge, with characteristics including length, cross-sectional area, laying depth, and topological distance; Divide the layers by physical topology: The trunk line is at a high level (Level 0); Branch lines are downgraded in sequence (Level 1-N); Each node records the level and parent node to which it belongs; 2. Hierarchical messaging: Message generation: The node integrates its own characteristics with the characteristics of the connecting cable to form a message packet; Intra-level transmission: nodes at the same level exchange current status information; Cross-level transfer: The child node uploads aggregated information to the parent node; The parent node broadcasts the global state downward; Failure probability calculation: Each round of transmission updates the node state vector; Finally, the failure probability value of each node is output; 3. Positioning decision-making process: Level failure probability = the maximum value of all node probabilities at that level; Once the highest probability tier is determined, within that tier: Screen for faulty nodes exceeding the threshold; Sort by probability values to generate candidate locations; Output includes: fault level, list of node IDs, location coordinates, probability distribution.

[0053] IV. System collaborative workflow 1. Real-time monitoring loop: Edge nodes continuously process sensor data streams; High-frequency arc data triggers real-time alarms (response time < 10ms); The results of hidden danger classification are used to update the characteristics of graph nodes in real time; 2. Periodic analysis loop; Start the LSTM global trend assessment every 24 hours; Update the prediction model by fusing recent partial discharge data; Output a heat map of the health of the cable network; 3. Location trigger mechanism: When the confidence level of hidden danger classification exceeds the threshold: Freeze the current graph state; Execute hierarchical message passing (3 - 5 rounds); Generate a location report and trigger a work order; Start multi-level joint location in case of major faults.

[0054] Case effect: Cable intelligent monitoring system for a commercial complex: Cable laying methods: cable trench, cable tray, shaft, buried Main hidden dangers: insulation aging, joint overheating, moisture intrusion Typical scenario: outgoing line from the power distribution room Sensor deployment plan: Laying method Sensor type Installation density Monitoring focus Cable trench Distributed optical fiber temperature measurement Continuous monitoring Local overheating point Bridge Wireless temperature measurement tag Every 5 meters Poor heat dissipation area Buried pipeline Fault current indicator Branch node Insulation deterioration The process of handling hidden dangers after the systematic deployment of the present invention: LSTM early warning: humidity > 85%RH for 7 consecutive days Graph network location: power distribution room → cable trench section of the merchant On-site treatment: found that the drain pipe was broken → insulation restored after repair Implementation comparison results: Index Commercial property system Traditional method Hidden danger identification accuracy rate 97.8% <85% Location accuracy ±0.5m >5m Response time 15s >30min Electrical fire reduction rate 76% - Maintenance cost reduction 57% - After the present invention is applied to this commercial complex, the effect is obvious, the cable life is extended by 40%, and for commercial properties, it realizes management innovation by adopting predictive maintenance to replace planned maintenance.

[0055] The module described as a separation component may or may not be physically separated. The component shown as a module may or may not be a physical module, that is, it may be located in one place or distributed across multiple network modules. Some or all of the modules can be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0056] Those skilled in the art should understand that the embodiments of the present invention can be provided as a method or a system. Therefore, the present invention can take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware aspects.

[0057] It should also be noted that the term "comprising", "including" or any other variant thereof is intended to cover non-exclusive inclusion, so that a process, method, commodity or device comprising a series of elements not only includes those elements but also includes other elements not expressly listed, or also includes elements inherent to such process, method, commodity or device. Without further limitation, an element defined by the statement "comprising one..." does not exclude the presence of additional identical elements in the process, method, commodity or device comprising the element.

[0058] The above are only the embodiments of the present invention and are not used to limit the present invention. For those skilled in the art, the present invention can have various changes and modifications. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included within the scope of the claims of the present invention.

[0059] After considering the present invention disclosed in the specification and the embodiments, those skilled in the art will readily conceive of other embodiments of the present invention. The present invention is intended to cover any variations, uses or adaptations of the present invention, which follow the general principles of the present invention and include common general knowledge or conventional technical means in the technical field not disclosed by the present invention. The specification and the embodiments are only regarded as exemplary, and the true scope and spirit of the present invention are pointed out by the claims above.

[0060] It should be understood that the present invention is not limited to the exact structures described above and shown in the drawings, and various modifications and changes can be made without departing from its scope. The scope of the present invention is only limited by the appended claims.

Claims

1. A systematic monitoring method for the electrical safety performance of cables, characterized in that Including: Collect sensor data to the edge computing node, preprocess the data, and output real-time calculated temperature data, partial discharge data, resistance-capacitance data, and vibration data; Send the data to the hidden danger classification model and fault location model in the cloud, and the model outputs early warning information for operation and maintenance engineering processing; Among them, the sensor data sending process is judged according to the data frequency. When the judgment is greater than 1 kHz, the 1D-CNN arc detection model is loaded. When the judgment is less than 1 kHz, the LightGBM feature classification model is loaded, and the 24-hour sensor data less than 1 kHz is periodically loaded into the LSTM-Attention gradient analysis model; Among them, the relevant data of partial discharge judged by the 1D-CNN arc detection model is sent to the LSTM-Attention gradient analysis model for training, and the trained LSTM-Attention gradient analysis model is used to evaluate the partial discharge trend of the global cable.

2. The systematic monitoring method for the electrical safety performance of a cable according to claim 1, characterized in that The fault location model includes: a cable network graph structure model. The sensors are set as sensor nodes, and each node is set with node features such as temperature, current, humidity, and partial discharge amount. Cable segments are set between sensor nodes, and each cable segment and sensor node are numbered. The cable segment features are set as cable length, cross-sectional area, laying depth, and topological distance. An adjacency matrix of the cable segment and the sensor node is constructed based on Gaussian kernel weighting for message passing. The message passing mechanism is to splice the linked cable segment features and the features of adjacent sensor nodes to form a more message function. The message function is updated in real time based on the maximum value and total value of the failure rate based on the node position. The message function is passed between sensor nodes at different levels. The numbers of the same sensor node and cable segment under different branches are totaled as one level. Multiple levels distinguish the trunk-branch relationship and form different connection layers from large to small. The message function is passed and updated between the global fully connected layers. Each connection layer is set with a failure probability, and the failure probability is determined by the data output by the hidden danger classification model.

3. The systematic monitoring method for the electrical safety performance of a cable according to claim 2, wherein The failure probability is the maximum value of the failure probabilities of multiple sensor nodes under the same connection layer, and the number position data of the sensor nodes at the level under the connection layer is located based on the failure probability.

4. The systematic monitoring method for the electrical safety performance of a cable according to claim 1, characterized in that, Among them, For the sensor data used to train the hidden danger classification model, it also includes failure data generated by finite element simulation based on the COMSOL Multiphysics software. The generation steps are as follows: Step 1: Geometrically construct the cable hierarchical structure: conductor, insulation layer, shielding layer, sheath, and joint components; Step 2: Set the materials and corresponding properties of the cable hierarchical structure; Step 3: Construct the coupling equations of the electrical field, thermal field, and structural mechanics field of the cable hierarchical structure; Step 4: Import the failure scenario to obtain the simulation implementation of insulation aging, joint loosening, overload operation, and moisture intrusion; Step 5: Set the boundary conditions for the electrical field, thermal field, and structural mechanics field respectively. Apply the rated current / voltage at the conductor end of the electrical field, convective heat dissipation on the outer surface of the cable in the thermal field, forced air cooling at the joint, and set the fixed cable end constraint displacement in the structural mechanics field; Step 6: Generate supplementary fault data for training the hidden danger classification model.

5. The systematic monitoring method for the electrical safety performance of a cable according to claim 4, characterized in that, The LSTM-Attention gradient analysis model accesses the preprocessed data and encodes the temporal dependence. After adding the Attention mechanism to focus on the key time periods, it cooperates with the fault location model to classify and output the fault hidden dangers and corresponding location positions.

6. The systematic monitoring method for the electrical safety performance of a cable according to claim 5, characterized in that, It also includes that after the LSTM-Attention gradient analysis model is pre-trained with the fault data generated by finite element simulation, it is then accessed with the real sensor data and the fault data generated by finite element simulation mixed in proportion as the training set for training. After correcting the training results of the LSTM-Attention gradient analysis model, the proportion of the fault data generated by finite element simulation is reduced for training and correction until the proportion of the fault data generated by finite element simulation drops to 0. When the training meets the iterative convergence condition, the training of the LSTM-Attention gradient analysis model is completed.

7. A systematic monitoring method for the electrical safety performance of a cable according to claim 1, characterized in that, Among them, The process of the 1D-CNN arc detection model for distinguishing hidden dangers is as follows: The data types accessed by the 1D-CNN arc detection model for preprocessing include partial discharge data, vibration signals, and temperature mutations; Substitute the processed data into the 1D-CNN arc detection model for multi-scale convolution feature extraction, and execute the feature fusion strategy for the data types involving multi-sensor data. Specifically, it includes fusing the features of PD pulses and vibration signals, and cooperating with the fault location model to output the arc, vibration position, and duration of the cable joint.

8. The systematic monitoring method for the electrical safety performance of a cable according to claim 1, characterized in that, The LightGBM feature classification model is used to load sensor data less than 1kHz for monitoring the abnormal states of cable aging / wetness, and label the corresponding data with the abnormal state labels of aging / wetness. Specifically, it includes: First, align the timestamps of different sensor data and normalize them to the same interval to eliminate the dimension. After preliminary annotation using an unsupervised method, construct the original sensor data types and corresponding basic statistical features. The basic statistical features include mean, variance, maximum value, and minimum value; construct advanced statistical features including: time window features; correlation features; trend features; periodic features; physical model features, set the importance of the features, and then train the model according to the three state labels of normal state, aging state, and wet state, and label the data.

9. A systematic monitoring method for the electrical safety performance of a cable according to claim 7 or 8, characterized in that, It also includes using the GNN aggregation model to embed the sensor node vectors, loading the processed high-frequency data into the 1D-CNN arc detection model for analysis, and loading the processed low-frequency data into the LightGBM feature classification model for analysis.

Images