New energy cable fault early warning method and early warning system based on artificial intelligence

Abstract

The application relates to the technical field of cable fault early warning, and specifically discloses a new energy cable fault early warning method and system based on artificial intelligence, which periodically collects state parameters of new energy cables, constructs a space-time fusion feature matrix after preprocessing the collected data; a fault early warning model is constructed based on a deep learning model fused with an improved capsule network and an attention mechanism, the fusion feature matrix is input into the fault early warning model, a cable real-time fault risk value is obtained by combining a preset cable dynamic aging loss model; a preset basic early warning threshold is corrected in real time according to a coupling value of cable current operation compliance and environmental stress, and a dynamic early warning threshold is obtained; the real-time fault risk value is compared with the dynamic early warning threshold, early warning information of a corresponding grade is output according to a comparison result, corresponding fault processing instructions are triggered, and the early warning earlyness, accuracy, self-adaptive capacity and operation and maintenance efficiency are effectively improved.

Description

Technical Field

[0001] This invention relates to the field of cable fault early warning technology, and in particular to a fault early warning method and system for new energy cables based on artificial intelligence. Background Technology

[0002] As a key component connecting new energy equipment to the power grid, the operating status of new energy cables directly affects the efficiency of new energy power generation and the safety and stability of the power grid. However, because new energy cables are usually laid in complex and variable environments, such as underground, underwater, or exposed to severe weather conditions, they are easily affected by various factors during operation, leading to frequent cable failures.

[0003] Existing technology acquires regional cable storage nodes based on cable monitoring data, and then acquires monitoring aggregation nodes; calculates the regional cable anomaly coefficient of the monitoring aggregation nodes and the new energy cable anomaly coefficient of each new energy cable, establishes a cable anomaly ranking, sequentially acquires the remaining monitoring data of the nodes after removing the monitoring data of individual cables, calculates the remaining anomaly coefficient of the nodes, performs difference calculation of the anomaly coefficients, and after the new energy cable is determined, sequentially extracts the data of the remaining new energy cables and calculates the remaining anomaly coefficient of the nodes until the regional cable anomaly determination result is that the cable area is normal; thereby realizing real-time monitoring, fault early warning and anomaly location of new energy cables, so as to improve the operating efficiency and safety of new energy cables.

[0004] However, the above technologies focus on macroscopic positioning through the ranking of anomaly coefficients. The purely data-driven model lacks physical mechanism support, and the interpretability of the early warning results is poor. Summary of the Invention

[0005] The purpose of this invention is to provide a method and system for early warning of faults in new energy cables based on artificial intelligence, aiming to solve the technical problems of existing pure data-driven models lacking physical mechanism support and having poor interpretability of early warning results.

[0006] To achieve the above objectives, this invention employs an artificial intelligence-based fault early warning method for new energy cables, comprising the following steps:

[0007] S1, periodically collect multi-dimensional heterogeneous data of new energy cables, preprocess the collected multi-dimensional heterogeneous data, and construct a spatiotemporal fusion feature matrix;

[0008] S2, input the spatiotemporal fusion feature matrix obtained in step S1 into the fault early warning model, extract features through the improved capsule network embedded in the fault early warning model, and fuse the output of the cable dynamic aging loss model to generate the real-time fault risk value of the cable.

[0009] S3. Generate a working condition correction coefficient based on the coupling value between the current operating load of the cable and the environmental stress; use the working condition correction coefficient to correct the preset fixed fault risk value in real time to obtain a dynamic early warning threshold applicable to the current working condition.

[0010] S4. Compare the real-time fault risk value output in step S2 with the dynamic early warning threshold obtained in step S3, and output the corresponding level of early warning information according to the comparison result; if the real-time fault risk value exceeds the dynamic early warning threshold, trigger the corresponding fault handling instruction.

[0011] In step S1,

[0012] The periodic collection of multi-dimensional heterogeneous data on new energy cables includes:

[0013] The monitoring areas for new energy cables are divided, and unique information is marked on the cables and their connection points in each area.

[0014] Sensors deployed in various monitoring areas and marking points are used to synchronously collect electrical parameters: current, voltage, partial discharge intensity; environmental parameters: ambient temperature, ambient temperature, vibration acceleration; and physical state parameters: cable surface temperature and cable strain.

[0015] The preprocessing of the collected multi-dimensional heterogeneous data includes:

[0016] A threshold denoising method based on wavelet transform is used to clean the multi-dimensional heterogeneous data.

[0017] Then, the time series moving average method is used to fill in the random missing values ​​of the multidimensional heterogeneous data;

[0018] Finally, the maximum-minimum normalization method based on the upper and lower limits of parameter safe operation is used to transform the parameters of the multidimensional heterogeneous data to the [0,1] interval.

[0019] In step S2,

[0020] The step of inputting the spatiotemporal fusion feature matrix obtained in step S1 into the fault early warning model, and extracting features through the improved capsule network embedded in the fault early warning model, includes:

[0021] The spatiotemporal fusion feature matrix constructed in step S1 is input into the conventional convolutional layer of the fault warning model based on the fusion of improved capsule network and attention mechanism to perform preliminary extraction of local spatiotemporal features and obtain feature map;

[0022] The feature map is input into the main capsule layer of the deep learning model, and it is transformed into a vector capsule. Output;

[0023] Then, an attention mechanism is introduced for each vector capsule. Calculate attention score : ,in, For trainable attention weight vectors, For trainable bias terms, For the activation function, prevent negative gradient vanishing;

[0024] Using attention scores We weight the vector capsules to obtain the weighted capsules. The weighted capsule is passed to the digital capsule layer via a dynamic routing algorithm, and outputs a vector representing the overall state of the cable. .

[0025] In step S2, the output of the fused cable dynamic aging loss model generates the real-time fault risk value of the cable, including:

[0026] The cumulative aging amount is calculated using the aforementioned dynamic aging loss model for cables. ,in, This represents the cumulative physical aging amount calculated at time t. These are predefined constants related to cable materials; The activation energy represents the energy barrier of cable insulation material aging. Boltzmann's constant; Let be the absolute temperature of the cable conductor at time t; The operating voltage at time t; Rated voltage; This is the voltage aging constant; Cumulative running time;

[0027] Then physical aging amount with vector The pieces are stitched together and then merged using a fully connected layer: ,in, For the final feature vector of fusion, For vector concatenation operations, , The weights and biases of the fusion layer are used; the fused features are then processed through the output layer of the fault warning model. Mapped to real-time fault risk value .

[0028] In step S3, generating the operating condition correction coefficient based on the coupling value of the current operating load of the cable and the environmental stress includes: the formula for calculating the comprehensive stress coupling value: in, For real-time current; This is the maximum safe current for the cable; This is the electrical stress component; This refers to the real-time absolute temperature of the conductor. This is the maximum permissible temperature for insulation. The thermal stress index is greater than 1; This is the thermal stress component; It is the vibration acceleration; This represents the maximum permissible value for vibration acceleration. For cable strain; This represents the maximum allowable strain of the cable. For mechanical stress components; formula for calculating the working condition correction factor: in, This is the working condition correction factor. This refers to the stress relative to the rated operating conditions.

[0029] A fault early warning system for new energy cables based on artificial intelligence, the system comprising:

[0030] The aggregation monitoring module is used to divide the new energy cables into cable monitoring areas, acquire the cable monitoring data of the new energy cables, obtain the regional cable storage nodes based on the cable monitoring data, and then construct a spatiotemporal fusion feature matrix.

[0031] The deep learning analysis and inference module is improved by inputting the spatiotemporal fusion feature matrix obtained in step S1, extracting features through the improved capsule network embedded in the fault warning model, and fusing the output of the cable dynamic aging loss model to generate the real-time fault risk value of the cable.

[0032] The dynamic threshold adjustment module is used to calculate the dynamic early warning threshold based on the coupling value between the current operating load of the cable and the environmental stress.

[0033] The graded early warning and execution module is used to compare the real-time fault risk value with the dynamic early warning threshold, output graded early warning information and trigger fault handling instructions.

[0034] The aggregated monitoring module includes:

[0035] The monitoring area division unit is used to acquire information on new energy cables and divide the monitoring area.

[0036] The cable labeling unit is used to label information for each cable;

[0037] The cable data monitoring unit is used to collect cable monitoring data for each new energy cable;

[0038] An independent storage and preprocessing unit is used to independently store, clean, fill in, and normalize cable monitoring data;

[0039] Feature matrix construction unit, used to construct spatiotemporal fusion feature matrix.

[0040] The improved deep learning analysis and inference module includes:

[0041] Capsule networks and attention units are used to perform forward inference and attention-weighted computation of the improved capsule network;

[0042] The physical aging model unit is used to calculate the cumulative physical aging of the cable based on historical operating data.

[0043] The feature fusion and risk calculation unit is used to fuse data-driven features and physical aging quantities, and calculate real-time fault risk values.

[0044] The dynamic threshold adjustment module includes:

[0045] The integrated stress calculation unit is used to calculate the combined stress coupling value of electrical stress, thermal stress, and mechanical stress.

[0046] The correction factor calculation unit is used to calculate the working condition correction factor;

[0047] An adaptive threshold calculation unit is used to generate dynamic early warning thresholds.

[0048] The tiered early warning and execution module includes:

[0049] The risk level determination unit is used to determine the warning level based on the comparison between the risk value and the threshold.

[0050] The instruction set execution unit is used to execute the corresponding fault handling instruction set according to different warning levels, including notification, suggestion and control instructions.

[0051] This invention discloses an artificial intelligence-based fault early warning method and system for new energy cables. It collects multi-dimensional cable data through a multi-sensor array and preprocesses it to construct a spatiotemporal fusion feature matrix, providing a high-quality data foundation for analysis. It extracts deep fault features using an improved capsule network and attention mechanism, and combines this with a dynamic aging physical model of the cable to output a comprehensive real-time fault risk value. Based on the current operating conditions of the cable, it calculates the combined electrical-thermal-mechanical stress and dynamically generates an adaptive early warning threshold, allowing the judgment criteria to flexibly change with the "fatigue level." By comparing the risk value with the dynamic threshold, it automatically triggers tiered early warnings and corresponding handling instructions, forming a management closed loop from notification and suggestions to emergency control. Through the deep integration of physical models and AI, it significantly improves the early warning capability, accuracy, adaptability, and operational efficiency, providing a reliable guarantee for the safe and stable operation of new energy cables. Attached Figure Description

[0052] To more clearly illustrate the technical solutions in the embodiments of the present 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 only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0053] Figure 1 This is a schematic diagram of the artificial intelligence-based new energy cable fault early warning method of the present invention. Detailed Implementation

[0054] Please see Figure 1 , Figure 1 This is a schematic diagram of the artificial intelligence-based new energy cable fault early warning method of the present invention.

[0055] This invention provides a fault early warning method for new energy cables based on artificial intelligence, including the following steps: S1, periodically collecting multi-dimensional heterogeneous data of new energy cables, preprocessing the collected multi-dimensional heterogeneous data, and constructing a spatiotemporal fusion feature matrix;

[0056] S2, input the spatiotemporal fusion feature matrix obtained in step S1 into the fault early warning model, extract features through the improved capsule network embedded in the fault early warning model, and fuse the output of the cable dynamic aging loss model to generate the real-time fault risk value of the cable.

[0057] S3. Generate a working condition correction coefficient based on the coupling value between the current operating load of the cable and the environmental stress; use the working condition correction coefficient to correct the preset fixed fault risk value in real time to obtain a dynamic early warning threshold applicable to the current working condition.

[0058] S4. Compare the real-time fault risk value output in step S2 with the dynamic early warning threshold obtained in step S3, and output the corresponding level of early warning information according to the comparison result; if the real-time fault risk value exceeds the dynamic early warning threshold, trigger the corresponding fault handling instruction.

[0059] In this embodiment, sensor arrays deployed in each cable monitoring area periodically and synchronously collect monitoring data of the new energy cables. The data includes electrical parameters, environmental parameters, and physical state parameters. The cable monitoring data of all new energy cables in each cable monitoring area are stored independently to obtain regional cable storage nodes.

[0060] The cable monitoring node corresponding to the regional cable storage node is the monitoring aggregation node.

[0061] The collected raw cable monitoring data then undergoes preprocessing including data cleaning, missing value imputation, and data normalization. All preprocessed cable monitoring data are then used to construct a three-dimensional spatiotemporal fusion feature matrix based on time series and spatial location. This allows the system to simultaneously capture the evolution of cable status over time and its spatial distribution differences. A model combining an improved capsule network and an attention mechanism is employed to extract deep features, capture the spatial hierarchy of fault symptoms, and focus on key signals.

[0062] Meanwhile, a dynamic aging loss model for cables is introduced to calculate the cumulative lifespan loss of cables.

[0063] Ultimately, data-driven characteristics are integrated with physical aging parameters to output real-time cable fault risk values. Simultaneously, the comprehensive stress coupling value of the cable is calculated in real-time, and a working condition correction coefficient is generated accordingly to dynamically adjust the preset fixed fault risk value. The core logic is: the more severe the working condition, the lower the warning threshold. Then, the real-time fault risk value is compared with the dynamic warning threshold to trigger a three-level warning. Each level is linked to a tiered response instruction, from enhanced monitoring to analysis, diagnosis, and preparation for maintenance, culminating in emergency control.

[0064] By collaboratively constructing an intelligent system that integrates comprehensive perception, deep fusion diagnosis, intelligent dynamic decision-making, and precise closed-loop execution at each step, it can achieve early detection, accurate judgment, adaptive early warning, and efficient handling of faults in new energy cables, providing strong technical support for the safe, stable, and efficient operation of core power assets.

[0065] Further, step S1: periodically collect multi-dimensional heterogeneous data of new energy cables, preprocess the collected multi-dimensional heterogeneous data, and construct a spatiotemporal fusion feature matrix;

[0066] The periodic collection of multi-dimensional heterogeneous data on new energy cables includes:

[0067] The monitoring areas for new energy cables are divided, and unique information is marked on the cables and their connection points in each area.

[0068] Sensors deployed in various monitoring areas and marking points are used to synchronously collect electrical parameters: current, voltage, partial discharge intensity; environmental parameters: ambient temperature, ambient temperature, vibration acceleration; and physical state parameters: cable surface temperature and cable strain.

[0069] The preprocessing of the collected multi-dimensional heterogeneous data includes:

[0070] A threshold denoising method based on wavelet transform is used to clean the multi-dimensional heterogeneous data.

[0071] Then, the time series moving average method is used to fill in the random missing values ​​of the multidimensional heterogeneous data;

[0072] Finally, the maximum-minimum normalization method based on the upper and lower limits of parameter safe operation is used to transform the parameters of the multidimensional heterogeneous data to the [0,1] interval.

[0073] In this embodiment, it is deployed along the new energy cable line. Sensor arrays at key monitoring nodes, in a periodic manner Simultaneously acquire multi-dimensional heterogeneous data of the cable, including electrical parameters: current, voltage, partial discharge intensity; environmental parameters: ambient temperature, ambient temperature, vibration acceleration; and physical state parameters: cable surface temperature, cable strain. At any given time... From the first The original parameter vector collected by each node is denoted as: Each symbol represents a time. ,Location Raw data collected: For current, For voltage, For partial discharge intensity, For ambient temperature, For temperature, For vibration acceleration, For cable surface temperature, The data is used to measure cable strain. Then, the collected heterogeneous data is preprocessed, using wavelet transform thresholding to remove noise and interpolation to fill in missing values. Since the parameters have different dimensions, a maximum-minimum normalization method based on the upper and lower limits of safe operation is used. Wherein, any original parameter ∈{ , , ,…}, yes In position ,time The value, It is a parameter The historical minimum observed value, It is a parameter The historical maximum observed value, yes The normalized value. After normalizing each parameter, we obtain the first normalized value. Normalized parameter vector at each position .

[0074] The preprocessed data is used to construct a three-dimensional spatiotemporal fusion feature matrix: in, The time dimension refers to the number of sampling points within a time window. The spatial dimension refers to the number of sensor nodes or monitoring areas deployed along the cable. For feature dimension, i.e., vector The length of the spatiotemporal fusion feature matrix. It also encapsulates the evolution of cable states in the time dimension and the distribution differences in the spatial dimension, providing a comprehensive and structured input for subsequent deep learning models.

[0075] Further, step S2: input the spatiotemporal fusion feature matrix obtained in step S1 into the fault early warning model, extract features through the improved capsule network embedded in the fault early warning model, and fuse the output of the cable dynamic aging loss model to generate the real-time fault risk value of the cable.

[0076] The step of inputting the spatiotemporal fusion feature matrix obtained in step S1 into the fault early warning model, and extracting features through the improved capsule network embedded in the fault early warning model, includes:

[0077] The spatiotemporal fusion feature matrix constructed in step S1 is input into the conventional convolutional layer of the fault warning model based on the fusion of improved capsule network and attention mechanism to perform preliminary extraction of local spatiotemporal features and obtain feature map;

[0078] The feature map is input into the main capsule layer of the deep learning model, and it is transformed into a vector capsule. Output;

[0079] Then, an attention mechanism is introduced for each vector capsule. Calculate attention score : in, For trainable attention weight vectors, Here, is the trainable bias term, and is the activation function, to prevent negative gradient vanishing.

[0080] Using attention scores We weight the vector capsules to obtain the weighted capsules. The weighted capsule is passed to the digital capsule layer via a dynamic routing algorithm, and outputs a vector representing the overall state of the cable. .

[0081] The output of the fusion cable dynamic aging loss model generates real-time fault risk values ​​for the cable, including:

[0082] The cumulative aging amount is calculated using the aforementioned dynamic aging loss model for cables. ,in, This represents the cumulative physical aging amount calculated at time t. These are predefined constants related to cable materials; The activation energy represents the energy barrier of cable insulation material aging. Boltzmann's constant; Let be the absolute temperature of the cable conductor at time t; The operating voltage at time t; Rated voltage; This is the voltage aging constant; The cumulative running time; then the physical aging amount. with vector The pieces are stitched together and then merged using a fully connected layer:

[0083] ,in, For the final feature vector of fusion, For vector concatenation operations, , The weights and biases of the fusion layer are used; the fused features are then processed through the output layer of the fault warning model. Mapped to real-time fault risk value .

[0084] In this embodiment, the fused features are obtained through the output layer of the fault early warning model. Mapped to the final failure risk value : in, The Sigmoid activation function ensures the output It is a scalar value between 0 and 1; and the fault risk value By comprehensively detecting the instantaneous abnormal state of the cable and its long-term cumulative aging, accurate early warning can be achieved.

[0085] Further, in step S3: Based on the coupling value between the current operating load of the cable and the environmental stress, a working condition correction coefficient is generated; through the working condition correction coefficient, the preset fixed fault risk value is corrected in real time to obtain a dynamic early warning threshold applicable to the current working condition.

[0086] The process of generating the operating condition correction coefficient based on the coupling value between the current operating load of the cable and environmental stress includes:

[0087] Formula for calculating the combined stress coupling value: in, For real-time current; This is the maximum safe current for the cable; This is the electrical stress component; This refers to the real-time absolute temperature of the conductor. This is the maximum permissible temperature for insulation. The thermal stress index is greater than 1; This is the thermal stress component; It is the vibration acceleration; This represents the maximum permissible value for vibration acceleration. For cable strain; This represents the maximum allowable strain of the cable. For mechanical stress components;

[0088] Formula for calculating the working condition correction factor: in, This is the working condition correction factor. This refers to the stress relative to the rated operating condition.

[0089] In this embodiment, the comprehensive stress coupling value is utilized. The preset fixed fault risk value is dynamically corrected to obtain the operating condition correction coefficient. The dynamic early warning threshold is generated, and the calculation formula is as follows: in, The threshold is a dynamic early warning threshold; For fixed failure risk value ( =0.7, =0.85); The adjustment coefficient (0 < ≤1), used to control the sensitivity of the threshold as stress changes;

[0090] because It represents the maximum value of each stress component relative to the load, with a value range around 0-1. When any stress component approaches its safety limit, the dynamic warning threshold will drop to its minimum. When all stress components are low, the system enters its most sensitive early warning state. When all stress components are low, the threshold is close to the base threshold, avoiding false alarms.

[0091] Further, in step S4: the real-time fault risk value output in step S2 is compared with the dynamic early warning threshold obtained in step S3, and the corresponding level of early warning information is output according to the comparison result; if the real-time fault risk value exceeds the dynamic early warning threshold, the corresponding fault handling instruction is triggered.

[0092] In this embodiment, the system presets a fixed fault risk value ( =0.7, =0.85), based on the real-time fault warning value Generate a Level 3 warning:

[0093] Level 1 warning: When 0.7 ≤ If the value is less than 0.85, the system determines that the cable is in an early abnormal state and triggers the following instructions: a yellow flashing indicator is displayed on the monitoring screen, and a warning message is pushed to the mobile devices of relevant maintenance personnel; the sampling frequency of all monitoring points of the cable is automatically increased to 1 time / minute.

[0094] Level 2 warning: When 0.85≤ If the value is less than 0.95, the system determines that the cable is in a confirmed abnormal state and triggers the following instructions: send a text message and a phone call to the maintenance team leader, requesting preparation for maintenance; parse the capsule network output and display diagnostic conclusions such as "suspected insulation aging" on the interface; generate an instruction to "reduce the load" and wait for the maintenance personnel to confirm and execute it.

[0095] Level 3 Warning: When If the value is ≥0.95, the system determines that the cable is in an emergency state and triggers the following commands: activate the audible and visual alarm in the control room and automatically call the maintenance supervisor; after a preset short delay or manual one-click confirmation, automatically send a trip command to the circuit breaker to cut off the power supply to the faulty cable; based on multi-sensor data fusion, output the precise location information of the fault point to the emergency repair terminal.

[0096] Through the aforementioned hierarchical early warning and command linkage, the system achieves automated and intelligent closed-loop management from early detection and early warning to final handling, greatly improving the safety and reliability of the new energy cable system.

[0097] Furthermore, an artificial intelligence-based fault early warning system for new energy cables includes:

[0098] The aggregation monitoring module is used to divide the new energy cable into cable monitoring areas, acquire the cable monitoring data of the new energy cable, obtain the regional cable storage nodes based on the cable monitoring data, and then construct a spatiotemporal fusion feature matrix; the improved deep learning analysis and inference module is used to analyze the spatiotemporal fusion feature matrix based on a deep learning model that integrates an improved capsule network and an attention mechanism and a cable dynamic aging loss model, and output the real-time fault risk value of the cable.

[0099] The dynamic threshold adjustment module is used to calculate the dynamic early warning threshold based on the coupling value between the current operating load of the cable and the environmental stress.

[0100] The graded early warning and execution module is used to compare the real-time fault risk value with the dynamic early warning threshold, output graded early warning information and trigger fault handling instructions.

[0101] In this embodiment, the system first collects raw, heterogeneous data from the cable monitoring area through the aggregation monitoring module, and performs preprocessing such as cleaning and normalization to construct a regular feature matrix containing spatiotemporal information, preparing it for AI analysis. The improved deep learning analysis and inference module receives the feature matrix and, on the one hand, uses an improved capsule network and attention mechanism to extract deep fault features from the data; on the other hand, it calculates the cumulative loss of the cable due to long-term service through a cable dynamic aging loss model. Finally, the two are fused to output a precise, quantified real-time fault risk value. The dynamic threshold adjustment module works in parallel, calculating the dynamic early warning threshold in real time based on the current operating load and environmental stress of the cable, so that the early warning standard can dynamically change with the "fatigue level" of the cable. Finally, the real-time fault risk value obtained by the hierarchical early warning and execution module is compared with the dynamic early warning threshold. The early warning level is automatically determined through preset two-level thresholds, and corresponding, progressively escalating disposal instructions are triggered, forming a management closed loop from notification and early warning to emergency control. This enables the system to construct a complete intelligent closed loop of "perception-diagnosis-decision-execution". The fusion of multi-dimensional spatiotemporal data and the improvement of AI models can capture subtle early fault characteristics that cannot be identified by the human eye and traditional models, realizing early and accurate fault warning. Furthermore, by introducing a dynamic aging loss model of cables, data-driven analysis and physical mechanisms are deeply integrated, so that the risk assessment results are not only based on instantaneous data, but also take into account the historical health status of the equipment, which greatly improves the scientificity, credibility and interpretability of the warning.

[0102] Furthermore, the aggregated monitoring module includes:

[0103] The monitoring area division unit is used to acquire information on new energy cables and divide the monitoring area.

[0104] The cable labeling unit is used to label information for each cable;

[0105] The cable data monitoring unit is used to collect cable monitoring data for each new energy cable;

[0106] The independent storage and preprocessing unit is used to independently store, clean, fill, and normalize cable monitoring data; the feature matrix construction unit is used to construct a spatiotemporal fusion feature matrix.

[0107] In this embodiment, the monitoring area division unit first divides the entire cable network into several logical cable monitoring areas based on the laying topology and key nodes (such as joints and terminals) of the new energy cables. The cable labeling unit then uniquely identifies and labels each cable segment and key component within each area, establishing a cable identity file. The cable data monitoring unit drives sensor arrays deployed in each area and labeling point to periodically and synchronously collect electrical parameters (current, voltage, partial discharge), environmental parameters (temperature, humidity, vibration), and physical state parameters (surface temperature, strain). The independent storage and preprocessing unit independently stores the raw monitoring data collected from each area in the corresponding area cable storage node, performs data cleaning and missing value imputation, and finally uses the max-min normalization method to unify various parameters to the [0,1] interval. The feature matrix construction unit ultimately processes all the preprocessed data according to time series. Spatial location and feature dimensions By performing structured recombination, a three-dimensional spatiotemporal fusion feature matrix was constructed, laying a reliable data foundation for subsequent in-depth analysis.

[0108] Furthermore, the improved deep learning analysis and inference module includes:

[0109] Capsule networks and attention units are used to perform forward inference and attention-weighted computation of the improved capsule network;

[0110] The physical aging model unit is used to calculate the cumulative physical aging of the cable based on historical operating data.

[0111] The feature fusion and risk calculation unit is used to fuse data-driven features and physical aging quantities, and calculate real-time fault risk values.

[0112] In this embodiment, the capsule network and attention unit receive a spatiotemporal fusion feature matrix. The improved capsule network inside first extracts primary features through convolutional layers, and then generates vector capsules in the main capsule layer to encapsulate the instantiation parameters of the features. Then, the attention mechanism is activated, the attention score of each capsule is calculated, and the capsules are weighted so that the model focuses on the features most relevant to the fault. The physical aging model unit runs in parallel. Based on the cable's historical load, temperature, and other data, and using electro-thermal aging physical models such as the Arrhenius equation, it calculates the cumulative physical aging of the cable and quantifies its lifespan loss caused by long-term use. The feature fusion and risk calculation unit concatenates and fuses the attention-weighted data-driven feature vector with the calculated physical aging amount, and finally outputs a real-time cable fault risk value between 0 and 1 through a fully connected layer and a Sigmoid function. Thus, through the improved deep learning analysis and reasoning module, the model can accurately capture the spatial hierarchical relationship of fault features and focus on key signs, significantly improving the model's feature extraction capability and interpretability. Furthermore, the model enables fault risk assessment to consider both instantaneous anomalies and historical cumulative aging, greatly enhancing the accuracy and early warning of early warnings.

[0113] Furthermore, the dynamic threshold adjustment module includes:

[0114] The integrated stress calculation unit is used to calculate the combined stress coupling value of electrical stress, thermal stress, and mechanical stress.

[0115] The correction factor calculation unit is used to calculate the working condition correction factor;

[0116] An adaptive threshold calculation unit is used to generate dynamic early warning thresholds.

[0117] In this embodiment, the integrated stress calculation unit acquires the cable's current, conductor temperature, vibration, and strain data in real time, calculates its electrical stress, thermal stress, and mechanical stress components respectively, and fuses them to generate a comprehensive stress coupling value. It reflects the most dangerous stress level at present; the correction coefficient calculation unit is based on the comprehensive stress coupling value. Calculate the stress relative to the rated operating condition. The relative deviation is used to obtain the working condition correction coefficient. The adaptive threshold calculation unit utilizes the operating condition correction coefficient. The preset basic warning threshold is dynamically adjusted to generate a dynamic warning threshold. Thus, the dynamic threshold adjustment module achieves a fundamental shift in warning thresholds from fixed to dynamically adaptive, effectively solving the problems of alarm delay and frequent false alarms inherent in traditional methods.

[0118] Furthermore, the tiered early warning and execution module includes:

[0119] The risk level determination unit is used to determine the warning level based on the comparison between the risk value and the threshold.

[0120] The instruction set execution unit is used to execute the corresponding fault handling instruction set according to different warning levels, including notification, suggestion and control instructions.

[0121] In this embodiment, the level determination unit continuously receives real-time fault risk values ​​sent by the improved deep learning analysis and inference module, compares them with the dynamic early warning thresholds sent by the dynamic threshold adjustment module, and automatically determines and triggers the corresponding early warning level based on two preset threshold ranges. Then, the instruction set execution unit automatically executes a predefined, progressively escalating fault handling instruction set according to the determined early warning level. Specifically, a level 1 early warning executes notification-type instructions (such as pushing alarms or increasing the sampling rate); a level 2 early warning adds analysis and suggestion-type instructions (such as outputting fault types or generating load reduction suggestions); and a level 3 early warning triggers control-type instructions (such as emergency calls, automatic load shedding, or tripping). Thus, the hierarchical early warning and execution module realizes closed-loop management from risk warning to operation and maintenance actions, transforms intelligent analysis results into actionable decisions, compresses the time from anomaly occurrence to response, and optimizes resource allocation through a hierarchical response mechanism, avoiding overreaction to minor issues and underreaction to major issues.

[0122] The above description discloses only one preferred embodiment of the present invention, and should not be construed as limiting the scope of the present invention. Those skilled in the art will understand that all or part of the processes of the above embodiments can be implemented, and equivalent changes made in accordance with the claims of the present invention are still within the scope of the invention.

Claims

1. A method for early warning of faults in new energy cables based on artificial intelligence, characterized in that, Includes the following steps: S1, periodically collect multi-dimensional heterogeneous data of new energy cables, preprocess the collected multi-dimensional heterogeneous data, and construct a spatiotemporal fusion feature matrix; S2, input the spatiotemporal fusion feature matrix obtained in step S1 into the fault early warning model, extract features through the improved capsule network embedded in the fault early warning model, and fuse the output of the cable dynamic aging loss model to generate the real-time fault risk value of the cable. S3 generates a working condition correction coefficient based on the coupling value between the current operating load of the cable and the environmental stress. The preset fixed fault risk value is corrected in real time using the operating condition correction coefficient to obtain a dynamic early warning threshold applicable to the current operating condition. The process of generating the operating condition correction coefficient based on the coupling value between the current operating load of the cable and environmental stress includes: Formula for calculating the combined stress coupling value: in, For real-time current; This is the maximum safe current for the cable; This is the electrical stress component; This refers to the real-time absolute temperature of the conductor. This is the maximum permissible temperature for insulation. The thermal stress index is greater than 1; This is the thermal stress component; It is the vibration acceleration; This represents the maximum permissible value for vibration acceleration. For cable strain; This represents the maximum allowable strain of the cable. For mechanical stress components; Formula for calculating the working condition correction factor: in, This is the working condition correction factor. This is relative to the rated operating stress; The preset fixed fault risk value is: =0.7, =0.8; S4. Compare the real-time fault risk value output in step S2 with the dynamic early warning threshold obtained in step S3, and output the corresponding level of early warning information according to the comparison result; if the real-time fault risk value exceeds the dynamic early warning threshold, trigger the corresponding fault handling instruction.

2. The artificial intelligence-based fault early warning method for new energy cables as described in claim 1, characterized in that, In step S1, The periodic collection of multi-dimensional heterogeneous data on new energy cables includes: The monitoring areas for new energy cables are divided, and unique information is marked on the cables and their connection points in each area. Sensors deployed in various monitoring areas and marking points are used to synchronously collect electrical parameters: current, voltage, partial discharge intensity; environmental parameters: ambient temperature, ambient temperature, vibration acceleration; and physical state parameters: cable surface temperature and cable strain. The preprocessing of the collected multi-dimensional heterogeneous data includes: A threshold denoising method based on wavelet transform is used to clean the multi-dimensional heterogeneous data. Then, the time series moving average method is used to fill in the random missing values ​​of the multidimensional heterogeneous data; Finally, the maximum-minimum normalization method based on the upper and lower limits of parameter safe operation is used to transform the parameters of the multidimensional heterogeneous data to the [0,1] interval.

3. The artificial intelligence-based fault early warning method for new energy cables as described in claim 1, characterized in that, In step S2, The step of inputting the spatiotemporal fusion feature matrix obtained in step S1 into the fault early warning model, and extracting features through the improved capsule network embedded in the fault early warning model, includes: The spatiotemporal fusion feature matrix constructed in step S1 is input into the conventional convolutional layer of the fault warning model based on the fusion of improved capsule network and attention mechanism to perform preliminary extraction of local spatiotemporal features and obtain feature map; The feature map is input into the main capsule layer of the deep learning model, transforming it into a vector capsule. Output; Then, an attention mechanism is introduced for each vector capsule. Calculate attention score : ,in, For trainable attention weight vectors, For trainable bias terms, For the activation function, prevent negative gradient vanishing; Using attention scores Weigh the vector capsules to obtain the weighted capsules. ; The weighted capsules are passed to the digital capsule layer via a dynamic routing algorithm, and the output is a vector representing the overall state of the cable. .

4. The artificial intelligence-based fault early warning method for new energy cables as described in claim 3, characterized in that, In step S2, the output of the fused cable dynamic aging loss model generates the real-time fault risk value of the cable, including: The cumulative aging amount is calculated using the aforementioned dynamic aging loss model for cables. ,in, This represents the cumulative physical aging amount calculated at time t. These are predefined constants related to cable materials; The activation energy represents the energy barrier of cable insulation material aging. Boltzmann's constant; Let be the absolute temperature of the cable conductor at time t; The operating voltage at time t; Rated voltage; This is the voltage aging constant; Cumulative running time; Then physical aging amount with vector The pieces are stitched together and then merged using a fully connected layer: ,in, For the final feature vector of fusion, For vector concatenation operations, , The weights and biases of the fusion layer; The output layer of the fault early warning model will fuse features. Mapped to real-time fault risk value .

5. An artificial intelligence-based fault early warning system for new energy cables, applied to the artificial intelligence-based fault early warning method for new energy cables as described in any one of claims 1-4, characterized in that, The system includes: The aggregation monitoring module is used to divide the new energy cables into cable monitoring areas, acquire the cable monitoring data of the new energy cables, obtain the regional cable storage nodes based on the cable monitoring data, and then construct a spatiotemporal fusion feature matrix. The deep learning analysis and inference module is improved by inputting the spatiotemporal fusion feature matrix obtained in step S1, extracting features through the improved capsule network embedded in the fault warning model, and fusing the output of the cable dynamic aging loss model to generate the real-time fault risk value of the cable. The dynamic threshold adjustment module is used to calculate the dynamic early warning threshold based on the coupling value between the current operating load of the cable and the environmental stress. The graded early warning and execution module is used to compare the real-time fault risk value with the dynamic early warning threshold, output graded early warning information and trigger fault handling instructions.

6. The system as described in claim 5, characterized in that, The aggregated monitoring module includes: The monitoring area division unit is used to acquire information on new energy cables and divide the monitoring area. The cable labeling unit is used to label information for each cable; The cable data monitoring unit is used to collect cable monitoring data for each new energy cable; An independent storage and preprocessing unit is used to independently store, clean, fill in, and normalize cable monitoring data; Feature matrix construction unit, used to construct spatiotemporal fusion feature matrix.

7. The system as described in claim 5, characterized in that, The improved deep learning analysis and inference module includes: Capsule networks and attention units are used to perform forward inference and attention-weighted computation of the improved capsule network; The physical aging model unit is used to calculate the cumulative physical aging of the cable based on historical operating data. The feature fusion and risk calculation unit is used to fuse data-driven features and physical aging quantities, and calculate real-time fault risk values.

8. The system as described in claim 5, characterized in that, The dynamic threshold adjustment module includes: The integrated stress calculation unit is used to calculate the combined stress coupling value of electrical stress, thermal stress, and mechanical stress. The correction factor calculation unit is used to calculate the working condition correction factor; An adaptive threshold calculation unit is used to calculate dynamic warning thresholds applicable to the current operating conditions.

9. The system as described in claim 5, characterized in that, The tiered early warning and execution module includes: The risk level determination unit is used to determine the warning level based on the comparison between the risk value and the threshold. The instruction set execution unit is used to execute the corresponding fault handling instruction set according to different warning levels, including notification, suggestion and control instructions.

Images