Cable online operation fault positioning, monitoring and early warning method and system based on neural network algorithm

Abstract

The invention relates to a cable online operation fault positioning, monitoring and early warning method and system based on a neural network algorithm, and the method comprises the steps: collecting the operation environment and state data of a cable through a multi-source sensor, and generating an original monitoring data set through noise reduction and time sequence alignment; aiming at the problem of periodic distortion caused by seasonal fluctuation of environment temperature and humidity, performing dynamic reference compensation to generate a reference data set without seasonal influence; aiming at a coupling effect of a dynamic load and temperature drift, calculating insulation performance evaluation data of temperature compensation through load-insulation correlation mapping; aiming at the problem of disconnection of waterproof sealing monitoring and fault positioning, a sealing degradation grade is generated based on regional high-humidity detection; aiming at the defect of a fixed threshold strategy, generating a fault coordinate and a risk probability in combination with spatial positioning and probability analysis; and finally, self-adaptive updating of the parameter library is realized through credibility verification.

Description

Technical Field

[0001] The present invention relates to the technical field of cable monitoring, and in particular to a method and system for locating, monitoring and warning of cable online operation faults based on a neural network algorithm. Background Art

[0002] With the increasing intelligence of modern power grids, the operating status of power cables, the core carriers of power transmission, is directly related to the safety and reliability of the power supply system. Multi-source sensing-based online cable monitoring technology, which collects parameters such as temperature, humidity, circulating current, and partial discharge in real time to build a digital twin model of the cable's operating status, has become a key means of preventing cable failures and ensuring stable grid operation. This technology uses a sensor network deployed along the cable to continuously capture the dynamic changes in the cable's operating environment. It also uses intelligent algorithms to assess insulation status and provide early warning of fault risks, providing important support for the power system's active defense system.

[0003] However, existing cable monitoring methods still suffer from significant drawbacks in practical applications. For one thing, seasonal fluctuations in ambient temperature and humidity can cause periodic distortion in monitoring data, making it difficult for traditional linear compensation models to effectively eliminate this interference, resulting in significant deviations in insulation resistance measurements. Furthermore, the multi-source data fusion process suffers from timing inaccuracies. The sampling frequency differences between temperature and circulating current sensors can cause key characteristic signals to misalign, impairing fault location accuracy. More critically, the coupled effects of dynamic load and temperature drift can cause insulation performance assessments to deviate from their true values. Existing technologies lack an effective joint compensation mechanism, resulting in a high false alarm rate. Furthermore, waterproof seal status monitoring is disconnected from fault location, making it impossible to establish a quantitative correlation model between humidity changes and insulation degradation, leaving maintenance decisions without a scientific basis. Furthermore, fixed threshold strategies struggle to adapt to parameter degradation during cable aging, necessitating the establishment of an intelligent early warning mechanism with closed-loop optimization. Summary of the Invention

[0004] Based on this, the purpose of the present invention is to provide a cable online operation fault location, monitoring and early warning method and system based on a neural network algorithm, which can dynamically compensate for environmental interference, accurately associate multi-source characteristics, and achieve reliable fault location and self-optimization early warning.

[0005] The purpose of the present invention is achieved by the following scheme:

[0006] In one aspect, the present invention provides a method for locating, monitoring and warning of cable online faults based on a neural network algorithm, comprising the following steps:

[0007] S1: Multi-source sensors deployed along the cable collect ambient temperature and humidity data, sheath circulating current data, and partial discharge pulse signal data of the cable operating environment to generate multi-source environmental data. Noise reduction and time series alignment are performed on the multi-source environmental data to generate the original monitoring data set.

[0008] S2: Analyze the seasonal fluctuation intensity of the original monitoring data set based on the neural network algorithm, call the preset monitoring strategy parameter library to perform dynamic benchmark compensation processing, and generate a dynamic environmental benchmark data set that removes seasonal effects;

[0009] S3: Perform load-insulation correlation mapping on the dynamic environment benchmark data set, calculate the insulation resistance correction value, and generate insulation performance evaluation data with temperature compensation characteristics;

[0010] S4: Perform a quantitative waterproof seal evaluation on the insulation performance evaluation data and the original monitoring data set, call the insulation resistance threshold data in the monitoring strategy parameter library to perform deviation verification, and generate waterproof degradation level data reflecting the degradation of sealing performance;

[0011] S5: Process the waterproof degradation level data and the original monitoring data set, extract the risk characteristics of the cable operation status and perform joint feature analysis to generate fault point coordinate data and risk probability value data;

[0012] S6: Perform credibility verification on the risk probability value data, and update the benchmark weight and insulation resistance threshold data of the monitoring strategy parameter library based on the accuracy deviation between the historical alarm records and the fault point coordinate data.

[0013] On the other hand, the present invention provides a cable online operation fault location monitoring and early warning system based on a neural network algorithm, which is configured with the following modules:

[0014] The data acquisition and preprocessing module is used to collect the ambient temperature data, ambient humidity data, sheath circulating current data and partial discharge pulse signal data of the cable operating environment through multi-source sensors deployed along the cable, generate multi-source environmental data, and perform noise reduction and time series alignment on the multi-source environmental data to generate the original monitoring data set;

[0015] The data seasonality analysis module is used to analyze the seasonal fluctuation intensity of the original monitoring data set based on the neural network algorithm, call the preset monitoring strategy parameter library to perform dynamic benchmark compensation processing, and generate a dynamic environmental benchmark data set that removes seasonal effects;

[0016] The load-insulation mapping module is used to perform load-insulation correlation mapping on the dynamic environment benchmark data set, calculate the insulation resistance correction value, and generate insulation performance evaluation data with temperature compensation characteristics;

[0017] The sealing and waterproof test module is used to perform quantitative waterproof sealing evaluation on the insulation performance evaluation data and the original monitoring data set, call the insulation resistance threshold data in the monitoring strategy parameter library to verify the deviation, and generate waterproof degradation level data reflecting the degradation of sealing performance;

[0018] The fault risk analysis module is used to process the waterproof degradation level data and the original monitoring data set, extract the risk characteristics of the cable operation status and perform joint feature analysis to generate fault point coordinate data and risk probability value data;

[0019] The parameter update module is used to perform credibility verification on the risk probability value data, and update the benchmark weight and insulation resistance threshold data of the monitoring strategy parameter library based on the accuracy deviation between the historical alarm records and the fault point coordinate data.

[0020] In summary, the present application provides a method for online cable fault location, monitoring and early warning based on a neural network algorithm, which can effectively overcome the core defects of existing cable monitoring technology through systematic data processing and closed-loop optimization mechanism: first, with the help of noise reduction and time alignment processing of multi-source sensor data, the spatiotemporal synchronous fusion of heterogeneous signals such as temperature and circulation can be achieved, eliminating the problem of feature dislocation caused by sampling frequency differences; through seasonal fluctuation intensity analysis and dynamic benchmark compensation processing of the original monitoring data, nonlinear correction of periodic interference of ambient temperature and humidity can be achieved, and the insulation resistance measurement deviation can be significantly suppressed; a temperature compensation mechanism based on load-insulation correlation mapping is used to decouple the coupling effect of dynamic load and temperature drift, so that the insulation performance evaluation value truly reflects the actual state of the cable; through quantitative evaluation of waterproof sealing and verification of insulation resistance threshold deviation, a quantitative correlation model of humidity change and insulation degradation can be established, and sealing status monitoring can be converted into a quantifiable maintenance basis; finally, combined with the closed-loop verification of historical alarm records of fault point coordinate accuracy deviation analysis, dynamic optimization and update of warning thresholds and compensation parameters can be achieved, forming an intelligent early warning system that adaptively evolves with the cable aging process. This full-process technical system can comprehensively improve the accuracy of fault location and the reliability of insulation status assessment, and provide monitoring support with continuous evolution capabilities for the power grid active defense system.

[0021] For better understanding and implementation, the present invention is described in detail below with reference to the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] Figure 1 A flowchart of a method for locating, monitoring, and warning cable online faults based on a neural network algorithm is provided in an embodiment of the present application;

[0023] Figure 2 A schematic diagram of a process for generating a dynamic environment benchmark dataset provided in an embodiment of the present application;

[0024] Figure 3 A schematic diagram of a process for generating insulation performance evaluation data with temperature compensation characteristics provided in an embodiment of the present application;

[0025] Figure 4 A schematic diagram of a process for generating fault point coordinate data and risk probability value data of cable fault points provided in an embodiment of the present application;

[0026] Figure 5 A structural diagram of a cable online operation fault location, monitoring and early warning system based on a neural network algorithm is provided in another embodiment of the present application. DETAILED DESCRIPTION

[0027] To facilitate understanding of the present invention, the present invention will be described more fully below with reference to the accompanying drawings. The drawings illustrate preferred embodiments of the present invention. However, the present invention may be implemented in many different forms and is not limited to the embodiments described herein. Rather, these embodiments are provided to provide a more thorough and comprehensive understanding of the disclosure.

[0028] Unless otherwise defined, all technical and scientific terms used herein have the same meanings as those commonly understood by those skilled in the art to which this invention pertains. The terms used in this specification are for the purpose of describing specific embodiments only and are not intended to limit the invention. The term "and / or" as used herein includes any and all combinations of one or more of the associated listed items.

[0029] In one embodiment, Figure 1 As shown, a cable online operation fault location, monitoring and early warning method based on a neural network algorithm is provided. This embodiment uses the method applied to a terminal as an example. It can be understood that the method can also be applied to a server, and can also be applied to a system including a terminal and a server, and is implemented through the interaction between the terminal and the server.

[0030] In this embodiment, the method includes the following steps:

[0031] S1: Multi-source sensors deployed along the cable collect ambient temperature data, ambient humidity data, sheath circulating current data, and partial discharge pulse signal data of the cable operating environment to generate multi-source environmental data. The multi-source environmental data is then subjected to noise reduction and time series alignment to generate the original monitoring data set.

[0032] Specifically, multi-source sensors deployed along the cable acquire ambient temperature and humidity data, as well as sheath circulating current data and partial discharge pulse signal data, based on preset sampling frequencies and cycles. Ambient temperature data is accurately measured by thermistors within the multi-source sensors, based on the temperature-dependent resistance variation. Humidity data is accurately sensed by capacitive humidity sensors, leveraging the change in dielectric constant. Sheath circulating current is acquired by converting magnetic field strength into a current signal using a Hall effect current sensor. Partial discharge pulse signals are collected by ultra-high frequency sensors, capturing changes in electromagnetic waves.

[0033] Specifically, the system can use a Kalman filter algorithm to reduce noise in multi-source environmental data. This algorithm dynamically estimates the data state based on a state-space model and adjusts the filter gain in real time, effectively suppressing high-frequency noise and improving the signal-to-noise ratio. To address timing misalignment caused by differences in the sampling frequencies of different sensors, the system can use a time series interpolation algorithm to align the timing. Using the highest sampling frequency as a benchmark, it interpolates low-frequency data to ensure that all data accurately corresponds to the same time series. For example, if the ambient temperature sensor samples once per minute, while the partial discharge pulse signal sensor samples multiple times per second, the system will use the time series interpolation algorithm to appropriately interpolate between temperature data points, ensuring that the temperature data and pulse signal data accurately match in time.

[0034] The system integrates the noise-reducing and time-aligned data into the original monitoring dataset, providing an accurate, synchronized, and pure data foundation for subsequent analysis. During data collection, the system monitors the sensor's operating status in real time. If any sensor data anomalies or missing data are detected, an alarm will automatically trigger and attempt to re-collect or switch to a backup sensor to ensure data continuity and integrity. Furthermore, the system performs a preliminary integrity check on the collected raw data, checking whether the data is within a reasonable range. For example, this checks whether temperature data exceeds the extreme temperature range of cable operation and whether humidity data is within the acceptable range for the local environment, further ensuring the quality of the original monitoring dataset.

[0035] S2: Based on the neural network algorithm, the seasonal fluctuation intensity analysis and processing of the original monitoring data set are performed, and the preset monitoring strategy parameter library is called to perform dynamic benchmark compensation processing to generate a dynamic environmental benchmark data set that removes seasonal effects.

[0036] Specifically, the system feeds raw monitoring data into a preprocessed neural network model. The model's input layer receives multidimensional data, including ambient temperature, humidity, sheath circulation values, and partial discharge pulse signals. The hidden layer uses the ReLU activation function to perform nonlinear combination and feature extraction on the input data, automatically learning the hidden seasonal fluctuations in the data. The system uses years of collected historical data to train the neural network, enabling it to accurately capture seasonal fluctuations in data, such as the significant changes in various parameters in high temperature and humidity environments in summer.

[0037] During the training process, the system minimizes the error between the predicted value and the actual value by adjusting the network weights and biases until the model converges. In this embodiment, the system uses the trained neural network to analyze the seasonal fluctuation intensity of the current original monitoring data and quantify its fluctuation amplitude and periodic characteristics. Based on the analysis results, the system automatically calls the preset monitoring strategy parameter library. This parameter library is based on cable operation experience and historical data, and stores dynamic benchmark compensation parameters for different seasons, different regions, and different load conditions. These parameters exist in the form of mathematical models, including compensation coefficients, correction functions, etc. Based on the current seasonal fluctuation characteristics, the system selects the most matching dynamic benchmark compensation parameters from the parameter library, substitutes the original monitoring data into the compensation model, and eliminates the impact of seasonal fluctuations through mathematical operations.

[0038] For example, for temperature data with significant annual cyclical fluctuations, the system can use a periodic function fitting method based on Fourier transform to construct a dynamic baseline temperature curve. The system calculates the difference between the original temperature data and the baseline curve, uses this difference as compensated data, and generates a dynamic environmental baseline data set that removes seasonal effects. During this process, the system can also perform a secondary verification of the compensated data to ensure the rationality of the compensation results. If it is found that the compensated data still has significant seasonal fluctuation characteristics, the system can re-evaluate the selected compensation parameters and make corresponding adjustments. In addition, the system can regularly update the monitoring strategy parameter library and incorporate new operating data and environmental information into the parameter library to ensure the timeliness and accuracy of the parameter library, enabling the system to better adapt to changes in the cable operating environment.

[0039] S3: Perform load-insulation correlation mapping on the dynamic environment benchmark data set, calculate the insulation resistance correction value, and generate insulation performance evaluation data with temperature compensation characteristics.

[0040] Specifically, based on the physical properties and operating principles of the cable, the system constructs a mathematical calculation model that comprehensively considers the complex relationships between multiple variables, including the cable's conductor material, insulation thickness, dielectric constant and thermal conductivity, as well as load current and operating time. Specifically, the system uses a system of partial differential equations and numerical simulation methods to simulate the changes in the cable's insulation performance under different load and temperature conditions, thereby determining the precise functional relationship between each factor and the insulation resistance. The system uses this mathematical calculation model to calculate the original insulation resistance value of the cable based on data such as the sheath circulating current value and ambient temperature in the dynamic environmental benchmark dataset.

[0041] The system also utilizes a temperature compensation algorithm. This algorithm, based on the thermal characteristic curve of the insulation material, uses a polynomial fitting method to accurately describe the relationship between the resistivity of the insulation material and temperature. The system collects a large amount of insulation material resistivity data at different temperatures and, using data fitting methods such as least squares, derives a polynomial relationship between temperature and resistivity. In actual calculations, the system substitutes the real-time temperature data into this polynomial relationship to calculate the temperature correction coefficient. This coefficient is then multiplied by the original insulation resistance value to produce insulation performance evaluation data with temperature compensation characteristics.

[0042] To ensure the accuracy of the calculation results, the system verifies the intermediate results of each step. For example, when calculating the original value of the insulation resistance, the system can check whether the input sheath circulation value and ambient temperature data are within a reasonable range and whether they conform to the physical characteristics of the cable. If abnormal data is found, the system automatically starts the data cleaning program to remove the erroneous data or use interpolation methods to supplement the missing data. At the same time, the system can also perform a rationality check on the insulation performance evaluation data after temperature compensation, compare and analyze it with historical data, and determine whether the data has sudden changes or abnormal fluctuations. If the data is abnormal, the system can further analyze the cause, which may be that the model parameters need to be adjusted, or there are errors in the input data. Through the verification and processing mechanism, the system ensures the accuracy and objectivity of the insulation performance evaluation data, providing high-quality data support for subsequent fault location and risk assessment.

[0043] S4: Perform waterproof sealing quantitative evaluation processing on the insulation performance evaluation data and the original monitoring data set, call the insulation resistance threshold data of the monitoring strategy parameter library to perform deviation verification, and generate waterproof degradation level data reflecting the degradation of sealing performance.

[0044] Specifically, the system constructs a quantitative assessment model for waterproof sealing based on the cable's geometric structure model, the sealing material's physical properties such as permeability and diffusion coefficient, and the changing patterns of ambient humidity. This model uses finite element analysis to divide the cable's sealing layer into multiple tiny cells, simulating the penetration path and accumulation of water molecules in each cell. By establishing a set of differential equations for water molecule penetration and accounting for the physical processes of water molecules in the sealing material, such as diffusion, adsorption, and desorption, the system can accurately calculate the cable's waterproof sealing performance indicators.

[0045] The system also accesses insulation resistance threshold data from the monitoring strategy parameter library. These thresholds are determined through statistical analysis based on cable design standards, operational experience, and historical fault data. They cover the insulation resistance range required for safe operation of cables in different operating stages and environmental conditions. The system compares and analyzes the insulation performance assessment data against the insulation resistance thresholds and calculates their deviation values. By constructing a deviation verification model and combining statistical hypothesis testing methods such as the t-test and F-test, the system determines whether the current insulation performance deviates significantly from the normal range.

[0046] For example, the system calculates the difference between the insulation performance evaluation data and the threshold, and determines whether the difference is statistically significant based on the distribution characteristics of the sample data. If the difference is significant, it means that there may be a problem with the waterproof sealing performance of the cable. Based on the analysis results, the system automatically generates waterproof degradation level data that reflects the degradation of sealing performance. The data is presented in the form of quantitative levels, and the waterproof sealing status of the cable is divided into multiple levels by setting different level thresholds. For example, level 1 indicates good sealing performance, level 2 indicates a slight decrease in sealing performance, level 3 indicates a significant decrease in sealing performance, and level 4 indicates a serious failure of sealing performance, etc. The system automatically performs operations such as model construction, data comparison, statistical analysis, and level division throughout the process.

[0047] To improve the reliability of assessment results, the system regularly calibrates and verifies the waterproof and sealant quantitative assessment model. By introducing sample data from cables with known sealing performance, the system compares the model's assessment results with actual conditions, adjusts model parameters, and optimizes the assessment algorithm. The system also dynamically updates the insulation resistance thresholds in the monitoring strategy parameter library, adjusting the threshold ranges appropriately based on cable aging and changes in the operating environment to ensure that the thresholds accurately reflect the safe operating status of the cable. These measures ensure the scientific and accurate nature of the waterproof and sealant quantitative assessment results, providing a strong basis for subsequent maintenance decisions.

[0048] S5: Process the waterproof degradation level data and the original monitoring data set, extract the risk characteristics of the cable operation status and perform joint feature analysis to generate fault point coordinate data and risk probability value data.

[0049] Specifically, the system uses feature extraction algorithms to extract time-frequency domain features of partial discharge pulse signals to obtain key characteristic parameters such as signal intensity, frequency, and duration; extracts fluctuation features of sheath circulating current values ​​to determine their fluctuation amplitude, change frequency and other characteristics; and combines ambient temperature, humidity and insulation performance evaluation data to calculate statistical indicators such as correlation coefficients and covariance between parameters to extract comprehensive features reflecting the cable operating status.

[0050] After feature extraction, the system normalizes the extracted feature data to eliminate dimensional differences and construct a feature vector matrix. Subsequently, the system applies a joint feature analysis model, based on machine learning algorithms such as support vector machines and random forests. This model learns the distribution patterns of cable features under normal and faulty conditions through training, determining the weight coefficients and correlations between different features. Based on the input feature vectors, the model calculates the probability of a cable fault. In combination with the cable's geometric distribution model and feature propagation characteristics, it uses mathematical methods such as triangulation and maximum likelihood estimation to comprehensively determine the coordinates of the fault point within the cable.

[0051] S6: Perform credibility verification on the risk probability value data, and update the benchmark weight and insulation resistance threshold data of the monitoring strategy parameter library based on the accuracy deviation between the historical alarm records and the fault point coordinate data.

[0052] Specifically, the system retrieves a large number of historical alarm records from a historical database. These records contain detailed information such as the time, location, and type of past faults, as well as the corresponding monitoring data characteristics. Simultaneously, the system obtains the coordinate data of the current fault point. By comparing the coordinate deviation between the historical records and the current data, the system calculates statistical indicators such as the root mean square error and standard deviation of the deviation to assess the accuracy of coordinate positioning. Preferably, the system can apply statistical process control methods to construct an accuracy deviation control chart, analyze whether the deviation is within the control limits, and thus determine whether the coordinate accuracy is stable and reliable.

[0053] Preferably, the system can calculate a credibility index based on a Bayesian credibility assessment model, using historical fault data and current risk probability values ​​as input variables. The Bayesian credibility assessment model updates the posterior probability distribution based on the prior fault probability and likelihood function, and determines the credibility of the risk probability value by calculating the confidence interval and credibility coefficient of the posterior probability. Based on the credibility verification results, the system can update the baseline weights and insulation resistance threshold data of the monitoring strategy parameter library. The update process can use an online learning algorithm, using the verification results as feedback signals to adjust the weight coefficients and threshold ranges in the parameter library.

[0054] For example, for threshold parameters with low credibility, the system can use gradient descent to optimize the threshold. By constructing a loss function, it calculates the error gradient between the predicted risk and the actual risk, and gradually adjusts the threshold to minimize the error. For baseline weights, the system can use a particle swarm optimization algorithm to search for the optimal weight combination in the weight space to maximize the system's overall monitoring performance indicators. Throughout the entire process, the system automatically performs operations such as data retrieval, deviation calculation, model evaluation, and parameter updates, ensuring that the monitoring strategy parameter library can promptly adapt to parameter changes during cable aging, optimizing monitoring strategies, improving the accuracy and reliability of the monitoring system, reducing false alarm rates, and providing continuous and effective monitoring and early warning services for the safe operation of cables.

[0055] In summary, the present application provides a method for online cable fault location, monitoring and early warning based on a neural network algorithm, which can effectively overcome the core defects of existing cable monitoring technology through systematic data processing and closed-loop optimization mechanism: first, with the help of noise reduction and time alignment processing of multi-source sensor data, the spatiotemporal synchronous fusion of heterogeneous signals such as temperature and circulation can be achieved, eliminating the problem of feature dislocation caused by sampling frequency differences; through seasonal fluctuation intensity analysis and dynamic benchmark compensation processing of the original monitoring data, nonlinear correction of periodic interference of ambient temperature and humidity can be achieved, and the insulation resistance measurement deviation can be significantly suppressed; a temperature compensation mechanism based on load-insulation correlation mapping is used to decouple the coupling effect of dynamic load and temperature drift, so that the insulation performance evaluation value truly reflects the actual state of the cable; through quantitative evaluation of waterproof sealing and verification of insulation resistance threshold deviation, a quantitative correlation model of humidity change and insulation degradation can be established, and sealing status monitoring can be converted into a quantifiable maintenance basis; finally, combined with the closed-loop verification of historical alarm records of fault point coordinate accuracy deviation analysis, dynamic optimization and update of warning thresholds and compensation parameters can be achieved, forming an intelligent early warning system that adaptively evolves with the cable aging process. This full-process technical system can comprehensively improve the accuracy of fault location and the reliability of insulation status assessment, and provide monitoring support with continuous evolution capabilities for the power grid active defense system.

[0056] In one embodiment, the present invention provides a method for locating, monitoring and warning of cable online faults based on a neural network algorithm, step S1 of which specifically includes the following steps:

[0057] S11: Multi-source sensors deployed along the cable collect ambient temperature data, ambient humidity data, sheath circulating current value data of the cable operating environment, and partial discharge pulse signal data to generate multi-source environmental data.

[0058] Specifically, the system generates multi-source environmental data by rationally deploying a multi-source sensor network along the cable. This system collects, in real time, ambient temperature and humidity in the cable's operating environment, as well as key indicators of the cable's operating status—sheath circulating current values ​​and partial discharge pulse signals. These sensors are located at different locations and depths along the cable to ensure the comprehensiveness and representativeness of the collected data. The system performs preliminary processing on the collected ambient temperature and humidity data, marking them with precise timestamps. This timestamp information is crucial for subsequent data synchronization. To address the timing deviation caused by differences in the sampling frequencies of multiple sensors, the system employs an adaptive window function compensation algorithm. The adaptive window function dynamically adjusts the window size based on the data's temporal characteristics, effectively compensating for the time difference between data sampled at different frequencies. This allows the data collected by different sensors to be aligned on the time axis, generating time-aligned temporal environmental parameters. This ensures the consistency and coherence of the data in the temporal dimension, laying a solid foundation for subsequent analysis and processing.

[0059] S12: Process the ambient temperature data and ambient humidity data of the multi-source environmental data, mark the timestamps and perform adaptive window function compensation to eliminate the timing deviation caused by the sampling frequency differences of multiple sensors, and generate time-aligned time series environmental parameters.

[0060] Specifically, the system processes the ambient temperature and humidity data from multiple sources, timestamping and recording the acquisition time of each data point to ensure the accuracy of the time information. Based on the collected data time series, the system adaptively adjusts the width and shape of the window function to accommodate the sampling frequency differences between different sensors. For example, for sensors with lower sampling frequencies, the system automatically increases the width of the window function to include enough data points within the time window, effectively eliminating the timing deviation caused by the sampling frequency differences between multiple sensors, and ensuring that the compensated data is precisely aligned in the time dimension.

[0061] During the adaptive window function compensation process, the system analyzes the timing characteristics of the data in real time, dynamically adjusts the window function parameters to achieve the optimal compensation effect, and generates time-aligned timing environment parameters. These parameters remain consistent over time, providing a precise time basis for subsequent multidimensional data fusion processing. The system performs a secondary verification of the compensated timing environment parameters to ensure that the data's time alignment accuracy meets the preset error range. During this process, if timing deviations are still found in the compensated data, the system re-evaluates the window function parameter settings and makes corresponding adjustments until the data's time alignment meets the requirements. In addition, the system records the parameter settings and results of each compensation operation for tracing and optimization in subsequent data analysis.

[0062] S13: Perform sliding window filtering on the sheath circulating current value data and partial discharge pulse signal data of the cable operation status in the multi-source environmental data, eliminate high-frequency interference through wavelet threshold noise reduction, and generate purification status data.

[0063] Specifically, the system applies sliding window filtering to cable sheath circulating current data and partial discharge pulse signal data from multi-source environmental data. By setting an appropriate window length and step size, the system slides the window across the data time series, smoothing the data within the window to reduce random fluctuations and noise interference. Sliding window filtering effectively reduces high-frequency noise components in the data while preserving the main trends and characteristics of the data.

[0064] After sliding window filtering, the system further uses wavelet threshold denoising to eliminate high-frequency interference. Wavelet threshold denoising uses a wavelet transform to decompose the signal into coefficients of varying scales. It then applies threshold processing to the wavelet coefficients to remove high-frequency noise components, and finally reconstructs the signal through an inverse wavelet transform. During this process, the system automatically adjusts the threshold parameters based on the signal characteristics and noise distribution, ensuring that the denoised data retains the key features of the original signal while minimizing noise interference. After sliding window filtering and wavelet threshold denoising, the system generates purified status data that more accurately reflects the actual operating status of the cable.

[0065] S14: Perform multi-dimensional data fusion processing on the time series environmental parameters and purification status data, build a monitoring data matrix with a unified timestamp, and generate an original monitoring data set.

[0066] Specifically, the system uniformly converts the data format and dimensions of the time-series environmental parameters and purification status data to ensure that they can be integrated into the same mathematical space. The system then constructs a monitoring data matrix based on a unified timestamp, arranging the time-series environmental parameters and purification status data in chronological order to form a matrix structure containing multi-dimensional information. In this matrix, each row represents the monitoring data at a point in time, and each column represents a specific type of monitoring parameter. Through this matrix processing, the system achieves deep fusion of multi-source data, allowing different types of monitoring data to be comprehensively analyzed within the same time frame.

[0067] When constructing the monitoring data matrix, the system fills in missing values ​​and processes outliers in multi-source data to ensure the integrity of the matrix and the reliability of the data. For example, the system can use interpolation to fill in missing data points; for outliers, the system can correct or eliminate them according to preset rules. After processing, the system generates the original monitoring data set, which not only contains temperature and humidity information of the cable operating environment, but also includes sheath circulating current values ​​and partial discharge pulse signal data of the cable operating status. All data has a unified timestamp and can accurately reflect the operating status of the cable at different time points. The system performs a final quality check on the original monitoring data set to verify the integrity and consistency of the data.

[0068] In one embodiment, Figure 2 As shown, the present invention provides a method for locating, monitoring and warning of cable online faults based on a neural network algorithm, step S2 specifically comprising the following steps:

[0069] S21: The original monitoring data set is input into the pre-trained neural network model for quarterly period processing, and the seasonal characteristic values ​​of the ambient temperature data and the ambient humidity data are extracted to generate seasonal fluctuation characteristics. The seasonal fluctuation characteristics are used to indicate the temperature and humidity fluctuation amplitude and periodic change pattern of the cable operating environment during the seasonal alternation process.

[0070] Specifically, the system inputs the original monitoring data set into a pre-trained neural network model for quarterly period processing. The neural network model is trained based on a large amount of historical data and can automatically identify and learn the time patterns and seasonal characteristics in the data. During the training process, the system uses many years of historical monitoring data, including parameters such as ambient temperature and humidity, and labels them according to the quarterly division standards. By adjusting the weights and biases of the network, the model can accurately divide the input data into different quarterly periods. The input layer of the model receives the time series data in the original monitoring data set, the hidden layer extracts features and models time dependencies of the data through activation functions, and the output layer generates quarterly labels corresponding to each data point. The system extracts the ambient temperature data and ambient humidity data within each quarter based on the quarterly labels output by the neural network model. The system further performs statistical analysis on the extracted data, calculates statistical indicators such as the mean, variance, standard deviation, etc. for each quarter, and extracts these statistical indicators as seasonal characteristic values. Preferably, the variance σ 2 The calculation formula for the standard deviation σ is as follows:

[0071]

[0072] Among them, σ 2 is the variance, μ is the sample mean, x iis the number of samples. The system integrates these seasonal eigenvalues ​​through mathematical transformation and feature extraction algorithm to generate a seasonal fluctuation eigenvector. The eigenvector is presented in the form of a vector, with one vector element corresponding to each quarter, which contains all the seasonal eigenvalues ​​of the quarter, and is used to indicate the temperature and humidity fluctuation amplitude and periodic change law of the cable operating environment during the seasonal alternation process. After generating the seasonal fluctuation eigenvector, the system will normalize it to eliminate the influence of different dimensions and dimensional units, so that the eigenvalues ​​of different quarters are comparable. At the same time, the system will store the generated seasonal fluctuation eigenvector in the database to provide a basis for subsequent dynamic weight compensation processing. Preferably, the seasonal fluctuation characteristics are obtained through the following steps:

[0073] S211: Perform time series reorganization processing on the environmental parameters of the original monitoring data set, align and standardize the ambient temperature data and humidity data along the time axis, and generate a three-dimensional time series input tensor with separated channel dimensions.

[0074] Specifically, when the system reorganizes the environmental parameters of the original monitoring data set in time series, it aligns and standardizes the ambient temperature data and humidity data along the time axis. The system ensures the consistency of temperature and humidity data in the time dimension through precise timestamp matching. The aligned data is processed by the Z-score normalization method so that both temperature and humidity data have zero mean and unit variance, thereby eliminating dimensional differences and numerical range differences, and generating a three-dimensional time series input tensor with channel dimension separation. The three-dimensional time series input tensor structurally includes time steps, number of samples, and feature dimensions, where each time step corresponds to a specific time point, the number of samples covers all data points within the monitoring period, and the feature dimensions correspond to the two channels of temperature and humidity respectively, ensuring that the data structure is regular and meets the input requirements of the deep learning model, laying the foundation for subsequent long-term feature extraction processing.

[0075] S212: Perform long-term feature extraction on the three-dimensional time series input tensor, learn the temporal dependency of seasonal fluctuations through the bidirectional LSTM network layer, and generate an initial time series feature vector containing cross-period coupling features.

[0076] Specifically, the system feeds a three-dimensional time series input tensor into a bidirectional LSTM (Long Short-Term Memory) network layer. The bidirectional LSTM network layer consists of a forward LSTM and a backward LSTM. The forward LSTM processes the data of time steps sequentially from the beginning to the end of the sequence, capturing historical information before each time point. The backward LSTM processes the data of time steps in reverse order from the end to the beginning of the sequence, capturing future information after each time point. At each time step, both the forward LSTM and the backward LSTM output a hidden state, which are concatenated together to form a bidirectional feature representation for that time step. In this way, the bidirectional LSTM can simultaneously consider past and future contextual information and effectively model long-term features in sequence data.

[0077] During the training process of the bidirectional LSTM network layer, the system utilizes a large amount of historical data and continuously adjusts the network weights and biases through a backpropagation algorithm, enabling the network to automatically learn the temporal dependencies of seasonal fluctuations. After processing by the bidirectional LSTM network layer, the system generates an initial time series feature vector. This initial time series feature vector contains a bidirectional feature representation for each time step, capturing the long-term characteristics and seasonal fluctuations in the ambient temperature and humidity data. The system monitors the training status and output quality of the bidirectional LSTM network layer in real time to ensure that the initial time series feature vector accurately reflects the long-term characteristics of the data.

[0078] S213: Perform key period focusing processing on the initial time series feature vector, use the attention mechanism layer to calculate the seasonal fluctuation contribution weights of different time nodes, and generate weighted time series features with time node sensitivity.

[0079] Specifically, based on the initial time series feature vector, the system uses the attention mechanism layer to calculate the seasonal fluctuation contribution weights at different time nodes. The input of the attention mechanism layer is the initial time series feature vector, and the feature vector of each time step is converted into an attention score. The calculation of the attention score is usually implemented through a trainable fully connected network, which maps the feature vector to a lower-dimensional space and then calculates the score in this space. The calculation formula is:

[0080] e t =tanh(W a *h t +b a )

[0081] Among them, e t is the attention score at time step t, h t is the eigenvector of time step t, W a and b ais a trainable weight matrix and bias vector. The system normalizes the attention score to the attention weight through the Softmax function, and the calculation formula is:

[0082]

[0083] Among them, α t is the attention weight for time step t. These attention weights reflect the contribution of different time points to seasonal fluctuations. The system multiplies the initial time series feature vector by the attention weights to obtain weighted time series features. Weighted time series features emphasize the characteristics of key periods and suppress the characteristics of irrelevant periods, making the feature vector sensitive to time points. The system adjusts the parameters of the attention mechanism layer in real time to optimize the calculation of attention weights and ensure that the weighted time series features accurately focus on the characteristics of key periods. In this way, the system can better capture the characteristics of important periods in seasonal fluctuations.

[0084] S214: Perform seasonal feature decoding processing on the weighted time series features, map them to the temperature fluctuation amplitude, humidity change period and seasonal alternation speed feature space through a fully connected network layer, and generate a seasonal fluctuation feature vector. The seasonal fluctuation feature vector is used to indicate the seasonal fluctuation intensity characteristics.

[0085] Specifically, the system performs seasonal feature decoding on the weighted time series features and inputs the weighted time series features into the fully connected network layer. The fully connected network layer consists of multiple neurons, each of which is connected to all elements of the input feature vector. The fully connected network layer performs a nonlinear transformation on the input features through an activation function and maps the features to a specified feature space. In this process, the system trains the weights and biases of the fully connected network layer so that it can map the weighted time series features to the feature space of temperature fluctuation amplitude, humidity change cycle, and seasonal alternation speed. The output seasonal fluctuation feature vector contains three main features: temperature fluctuation amplitude, humidity change cycle, and seasonal alternation speed, which respectively reflect the intensity, periodicity, and change rate of seasonal fluctuations. The system trains the fully connected network layer with a large amount of historical data, so that the output seasonal fluctuation feature vector can accurately indicate the seasonal fluctuation intensity characteristics.

[0086] During training, the system uses a loss function to measure the difference between the predicted seasonal fluctuation feature vector and the true value. The system continuously adjusts the network parameters through a backpropagation algorithm to minimize the loss function. After generating the seasonal fluctuation feature vector, the system verifies and evaluates it to ensure that it accurately reflects the characteristics of seasonal fluctuations. The resulting seasonal fluctuation feature vector is used for subsequent cable operation status assessment and fault warning analysis, providing the system with critical seasonal fluctuation information.

[0087] S22: Manage the seasonal fluctuation characteristic vector, call the historical benchmark data of the preset monitoring strategy parameter library to perform dynamic weight compensation processing, calibrate the offset through the sliding window standard deviation, and generate a benchmark compensation coefficient.

[0088] Specifically, the historical benchmark data contains the seasonal fluctuation characteristics of the same type of cables in the same region in different quarters over the past years, as well as the corresponding cable operating status parameters. The system compares and analyzes the currently extracted seasonal fluctuation feature vector with the historical benchmark data, and calculates the similarity between the current feature vector and the historical benchmark data, using methods such as cosine similarity or Euclidean distance to determine the position and degree of deviation of the current seasonal fluctuation feature in the historical data. Based on the similarity analysis results, the system extracts the dynamic weight parameters corresponding to the historical benchmark data that best matches the current seasonal fluctuation characteristics from the monitoring strategy parameter library. The system fuses the dynamic weight parameters with the current seasonal fluctuation feature vector through a weighted average algorithm to calculate the initial compensation coefficient. In order to improve the accuracy of the compensation coefficient, the system uses a sliding window standard deviation to calibrate the offset. The system sets a sliding window of fixed length, performs a sliding window calculation on the ambient temperature and humidity data in the original monitoring data set, and obtains the standard deviation within each time window. Preferably, the benchmark compensation coefficient is obtained by the following formula:

[0089]

[0090] Among them, λ is the benchmark compensation coefficient, n is the number of days of the sliding window, and w t The time decay weight factor is used to reflect the decay importance of data in time, usually using the exponential decay function w t =e -αt (α is the attenuation coefficient), which makes the recent data have a greater impact on the compensation coefficient; S t is the characteristic value of the current season in the seasonal fluctuation characteristics, H t is the historical benchmark value of the same period in the historical benchmark data, γ is the preset calibration coefficient, which is used to balance the weight between dynamic weight compensation and standard deviation calibration; σ t The standard deviation of the sliding window reflects the degree of data fluctuation within the sliding window. The system calibrates the initial compensation coefficient based on the standard deviation offset. Linear regression or nonlinear fitting can be used to adjust the initial compensation coefficient to ultimately generate a baseline compensation coefficient. During the generation of the baseline compensation coefficient, the system monitors the stability and convergence of the calculation process in real time to ensure the reliability of the compensation coefficient calculation results. The system also records each generated baseline compensation coefficient, its corresponding seasonal fluctuation characteristics, and historical benchmark data to facilitate subsequent evaluation and optimization of the compensation effect.

[0091] S23: Baseline reconstruction is performed on the original monitoring data set based on the benchmark compensation coefficient, and the long-term trend component of the environmental parameters is reconstructed using the encoder-decoder network to generate a dynamic environmental benchmark data set. The dynamic environmental benchmark data set is used to indicate the basic environmental status parameters of the cable after eliminating seasonal fluctuation interference.

[0092] Specifically, the system uses an encoder-decoder network to reconstruct the environmental parameters in the original monitoring data set. The encoder network is responsible for feature extraction and compression of the input environmental parameter data, mapping the data to a low-dimensional feature space. The encoder is usually composed of multiple neural network layers, such as convolutional layers, pooling layers, or LSTM layers, which are used to capture the time series features and spatial features in the data. The decoder network is responsible for decoding and reconstructing the features extracted by the encoder to generate the long-term trend component of the environmental parameters. The decoder is also composed of multiple neural network layers, and its structure is relatively symmetrical with that of the encoder. It restores the data in the feature space to the dimension of the original data through operations such as upsampling and deconvolution.

[0093] During the reconstruction process, the system uses the baseline compensation coefficient as one of the decoder's input features, combining it with the extracted feature vector to guide the decoder in generating the long-term trend component after eliminating seasonal fluctuations. The system trains the encoder-decoder network to accurately reconstruct the long-term trend component of environmental parameters while suppressing the influence of seasonal fluctuations and other short-term noise. During training, the system uses historical data annotated with long-term trends as training samples. By minimizing the error between the reconstructed data and the true long-term trend, the system adjusts the network's weights and biases to optimize network performance. Ultimately, the system generates a dynamic environmental benchmark dataset that reflects the basic environmental state parameters of the cable after eliminating seasonal fluctuations, providing a more accurate environmental benchmark for subsequent cable operation status assessments.

[0094] During the baseline reconstruction process, the system monitors the data reconstruction results in real time and calculates error metrics such as mean square error and mean absolute error between the reconstructed data and the original data. If the error is excessive, the system automatically adjusts the parameters of the encoder-decoder network or recalculates the baseline compensation coefficient, re-reconstructing the baseline until the error of the reconstructed data meets the preset accuracy requirements. The system stores the dynamic environmental baseline dataset in a database, providing data support for subsequent cable operation status analysis and fault warning. The system also regularly updates and maintains the dynamic environmental baseline dataset to ensure that it promptly reflects the latest changes in the cable operating environment.

[0095] In summary, the present application provides a method for cable online fault location, monitoring and early warning based on a neural network algorithm, which can effectively solve the problem of periodic distortion of monitoring data caused by seasonal fluctuations in ambient temperature and humidity in the background technology through time series feature extraction and dynamic compensation mechanism: first, through quarterly period division and seasonal characteristic value extraction, the periodic law and variation amplitude of temperature and humidity fluctuations can be accurately captured; dynamic weight compensation processing driven by historical benchmark data can achieve accurate calibration of seasonal offsets by sliding window standard deviation, overcoming nonlinear interference that traditional linear models cannot handle; finally, by reconstructing the long-term trend component of environmental parameters through baseline reconstruction technology, the interference effect of seasonal fluctuations on the basic environmental state parameters of the cable can be eliminated. This technical system can significantly improve the accuracy of insulation resistance measurement, provide a benchmark data set that is not affected by seasonal fluctuations for subsequent load-insulation correlation mapping, and fundamentally solve the core defect of "traditional compensation models are difficult to eliminate periodic interference" in the background technology, so that the insulation performance evaluation results can truly reflect the operating status of the cable.

[0096] In one embodiment, Figure 3 As shown, S3 of the cable online operation fault location, monitoring and early warning method based on a neural network algorithm provided by the present invention specifically includes the following steps:

[0097] S31: Perform current load feature extraction processing on the dynamic environment benchmark data set, identify the peak and valley waveforms of the cable load parameters and mark the periods of abnormal load, and generate a load pattern feature matrix reflecting the load change law.

[0098] Specifically, the system preprocesses the current data from the dynamic environment benchmark dataset, removing noise and filling missing values ​​to ensure data integrity and accuracy. The system then applies a wavelet transform to perform a multi-scale decomposition of the current data. The wavelet transform is capable of capturing the sudden changes and fluctuations in the current signal. By setting appropriate thresholds, the system identifies peaks and valleys in the current signal, which typically correspond to sharp changes in cable load. A peak waveform may indicate a sudden increase in load, while a valley waveform may indicate a sudden decrease in load.

[0099] The system time-stamps these waveforms, recording their occurrence time, duration, and amplitude. Simultaneously, by comparing the current load waveform with historical load data, the system identifies periods of abnormal load, where the load value may exceed the preset normal range or change at an abnormal rate. The system integrates this information into a load pattern characteristic matrix. Each row of this matrix represents the load characteristics within a time window, including statistical indicators such as the number of peaks and valleys, their duration, and amplitude, reflecting the changing patterns of cable load. The system updates and maintains this matrix in real time to ensure that it dynamically reflects the latest changes in cable load.

[0100] S32: Process the temperature gradient data of the dynamic environment benchmark data set based on a linear compensation algorithm to eliminate the temperature drift effect and generate temperature-corrected insulation parameters. The temperature-corrected insulation parameters are used to eliminate insulation resistance measurement deviations caused by ambient temperature fluctuations.

[0101] Specifically, the temperature gradient data in the dynamic environmental benchmark dataset reflects the rate of change of temperature along the cable. This data may be affected by multiple factors, including ambient temperature fluctuations and cable self-heating. The system first calculates the linear trend of the temperature gradient data and fits the temperature gradient data using the least squares method to obtain a linear model for temperature drift. The linear model is expressed as:

[0102] T drift (t) = a*t+b

[0103] Among them, T drift (t) represents the temperature drift value, a is the temperature drift rate, b is the intercept, and t is time. The system calculates the estimated value of temperature drift based on the model and then subtracts it from the original temperature gradient data to obtain the temperature-corrected insulation parameters. This process can be expressed as:

[0104] T corrected (t) = T raw (t)-T drift (t)

[0105] Among them, T corrected (t) is the insulation parameter after temperature correction, T raw (t) represents the raw temperature gradient data. This eliminates the effects of temperature drift caused by ambient temperature fluctuations, ensuring that temperature-corrected insulation parameters more accurately reflect the cable's actual temperature environment, thereby reducing the impact of ambient temperature fluctuations on insulation resistance measurements. The system regularly re-estimates and recalibrates the parameters of the linear compensation algorithm to adapt to changing environmental conditions and ensure accurate temperature correction.

[0106] S33: performing correlation mapping processing on the load mode characteristic matrix and the temperature-corrected insulation parameters, calculating the influence factor of the load change on the insulation performance, and generating insulation performance evaluation data with temperature compensation characteristics.

[0107] Specifically, the system aligns the peak and valley characteristics in the load pattern feature matrix with the temperature-corrected insulation parameters to ensure that both are analyzed on the same time dimension. The system establishes a mapping relationship between load variation characteristics and insulation performance by constructing a multivariate linear regression model or a neural network model. The model inputs include characteristics such as the number of peaks, valley depth, and load duration in the load pattern feature matrix, as well as the temperature-corrected insulation parameters. The output is the impact factor of load variation on insulation performance, which quantifies the degree of negative impact of load variation on insulation performance.

[0108] The system trains the model using extensive historical data and adjusts model parameters through algorithms such as gradient descent, enabling the model to accurately predict the impact of load changes on insulation performance. During model training, the system evaluates model performance through methods such as cross-validation to ensure generalization and predictive accuracy. The resulting temperature-compensated insulation performance evaluation data comprehensively accounts for load changes and temperature effects, more accurately reflecting the cable's true insulation condition and providing more reliable data support for subsequent fault diagnosis and maintenance decisions. The system regularly retrains and updates the model to adapt to changes in cable operating conditions and the accumulation of new data.

[0109] The above-mentioned method for cable online fault location, monitoring and early warning based on a neural network algorithm can fundamentally solve the insulation evaluation distortion problem caused by the coupling of dynamic load and temperature drift in the background technology through a load-temperature dual-dimensional collaborative analysis mechanism: by accurately extracting the characteristics of the load peak and valley waveform, it can capture the dynamic impact law of abnormal load periods on insulation performance; the linear compensation algorithm based on temperature gradient data can effectively remove the effect of ambient temperature drift and eliminate the insulation resistance measurement deviation caused by temperature fluctuations; finally, through the correlation mapping processing of load mode and temperature correction parameters, it can establish a quantitative model of the impact of load changes on insulation performance and generate insulation performance evaluation data with temperature compensation characteristics. This technical system successfully decouples the cross-interference effect of load and temperature to achieve accurate matching of insulation status evaluation values ​​with actual operating conditions, completely overcoming the core defect of the background technology that "dynamic coupling effect causes insulation performance evaluation to deviate from the true value", and providing a reliable judgment criterion for cable safety early warning that is not affected by environmental and load fluctuations.

[0110] In one embodiment, the present invention provides a method for locating, monitoring and warning of cable online faults based on a neural network algorithm, step S4 specifically comprising the following steps:

[0111] S41: Perform high humidity area detection processing on the ambient humidity data of the original monitoring data set, identify continuous exceeding points and generate a spatial cluster distribution map, and generate regional high humidity alarm information. The regional high humidity alarm information is used to indicate the geographical spatial distribution range of abnormal humidity accumulation in the cable channel.

[0112] Specifically, the system first preprocesses the ambient humidity data, removing noise and outliers to ensure data reliability. The system then uses a sliding window technique to traverse the humidity data and identify points where humidity levels continuously exceed a preset threshold. This threshold is typically determined based on historical data about the cable's operating environment and the cable's moisture resistance standards. The length of the sliding window can be customized based on actual needs. For example, a larger window length can be used to identify areas of high humidity that persist for extended periods. For each detected area of ​​high humidity, the system records its starting and ending locations and duration.

[0113] The system further performs spatial clustering analysis on these high-humidity areas, employing the K-means clustering algorithm or another suitable clustering algorithm to group geographically close high-humidity areas together. The clustering algorithm calculates the distances between high-humidity areas based on their latitude and longitude coordinates, grouping similarly located areas into the same cluster. This generates a spatial cluster distribution map, which visually illustrates the geographic distribution of abnormal humidity concentrations within the cable channel. Each cluster represents a relatively concentrated high-humidity area. The system also generates regional high-humidity alarm information, detailing the location, extent, duration, and severity of each high-humidity area, providing a basis for subsequent maintenance decisions.

[0114] S42: Perform sealing failure feature extraction processing on the regional high humidity alarm information, calculate the degradation slope through the humidity change rate over three days, and generate a sealing attenuation index.

[0115] Specifically, the system extracts the humidity data of each high-humidity area in a time series and calculates the humidity change rate within three days, that is, the ratio of the difference between the current humidity value and the humidity value three days ago to the time interval. Subsequently, the system fits the relationship between the humidity change rate and time through linear regression analysis to obtain the degradation slope of the ambient humidity. The degradation slope reflects the rate of humidity deterioration in high-humidity areas. The system calculates the slope between the humidity change rate and time through linear regression to obtain the sealing attenuation index. The sealing attenuation index combines the information of the humidity change rate and the degradation slope to quantify the degree of degradation of the sealing performance of the cable outer sheath waterproof structure. The system will compare and analyze the sealing attenuation index with historical data, evaluate its changing trend, and determine the development speed and severity of sealing failure. The system will also perform spatial distribution analysis on the sealing attenuation index to identify the areas with the most serious sealing failures, providing key information for subsequent maintenance decisions.

[0116] S43: Perform threshold deviation analysis on the sealing attenuation index, call the insulation resistance threshold data in the monitoring strategy parameter library to quantify the degree of degradation of the waterproof performance, and generate waterproof degradation level data reflecting the degree of degradation of the waterproof sealing performance. The waterproof degradation level data is used to indicate the aging decay level of the cable outer sheath waterproof structure.

[0117] Specifically, the system uses insulation resistance threshold data from the monitoring strategy parameter library. This threshold data is determined based on cable design standards, operating experience, and historical fault data, reflecting the safe lower limit of insulation resistance under different humidity conditions. The system establishes a quantitative relationship model between the seal attenuation index and insulation resistance, mapping the seal attenuation index to an assessment indicator of waterproof performance degradation. This model, typically trained using extensive experimental and field monitoring data, accurately reflects the impact of seal attenuation on waterproof performance. Based on the degree of threshold deviation—the difference between the seal attenuation index and the insulation resistance threshold—the system categorizes waterproof performance degradation into different levels, generating waterproof degradation grade data. This waterproof degradation grade data intuitively reflects the degree of aging and degradation of the cable outer sheath waterproof structure, providing maintenance personnel with a clear basis for repair decision-making. The system regularly recalibrates and optimizes the threshold deviation analysis model to adapt to changes in the cable operating environment and the accumulation of new data. The system also integrates this waterproof degradation grade data with a geographic information system (GIS) to generate a spatial distribution map of waterproof performance degradation, helping maintenance personnel more intuitively understand the overall status of waterproof sealing within the cable channel.

[0118] In one embodiment, Figure 4 As shown, S5 of the cable online operation fault location, monitoring and early warning method based on a neural network algorithm provided by the present invention specifically includes the following steps:

[0119] S51: performing waveform feature decomposition processing on the sheath circulating current value data of the original monitoring data set, separating the fundamental component and the harmonic characteristic parameters, and generating the circulating current phase amplitude characteristics.

[0120] Specifically, the system first preprocesses the sheath circulation value data, including noise removal and normalization, to ensure data accuracy and comparability. To remove noise, the system can use wavelet threshold denoising. This method decomposes the signal into wavelet coefficients of different scales and thresholds these coefficients to eliminate noise components. Normalization linearly transforms the data to the [0, 1] interval to eliminate the influence of different dimensions. The system then applies a Fourier transform to perform frequency domain analysis on the processed sheath circulation value data. The Fourier transform converts the time domain signal into the frequency domain signal, allowing the system to identify the different frequency components in the signal.

[0121] By setting an appropriate frequency range and resolution, the system separates the fundamental component and harmonic characteristic parameters. The fundamental component corresponds to the main frequency component of the signal, reflecting the basic trend and average amplitude of the sheath circulating current value; the harmonic characteristic parameters capture the periodic fluctuations and high-frequency details in the signal, such as the second harmonic and third harmonic. The system further calculates the amplitude and phase information of each frequency component to generate a circulating current phase amplitude characteristic matrix. Each row of this matrix represents the circulating current characteristics within a time window, containing information such as the fundamental amplitude, fundamental phase, and the amplitude and phase of each harmonic. The system updates and maintains this characteristic matrix in real time to ensure that it can dynamically reflect the latest changes in the sheath circulating current value. The system also performs statistical analysis on the circulating current phase amplitude characteristics, calculating statistical indicators such as its mean and standard deviation to assess the stability of the sheath circulating current value.

[0122] For example, by calculating the standard deviation of the fundamental amplitude, the system can determine whether the fluctuation of the sheath circulating current value is within the normal range; by analyzing the changing trends of the harmonic amplitudes, the system can identify potential fault signs. The system uses these statistical indicators to further enrich the circulating current phase amplitude feature matrix, providing more comprehensive feature information for subsequent multi-dimensional feature fusion processing. During the feature extraction process, the system automatically records the extraction time, extraction method, and related parameter settings for each feature, allowing for traceability and optimization in subsequent data analysis.

[0123] S52: Perform multi-dimensional feature fusion processing on the waterproof degradation level data and the circulation phase amplitude characteristics, construct a risk association feature space, and generate a multi-source risk feature matrix.

[0124] Specifically, the system aligns the waterproof degradation level data and the circulation phase amplitude feature matrix in chronological order to ensure that the two are analyzed in the same time dimension, and uses the interpolation method to match the timestamps so that the waterproof degradation level data and the circulation phase amplitude features at each time point can correspond to each other. Preferably, the system can use the principal component analysis (PCA) method to fuse these two different types of data. Specifically, the system calculates the covariance matrix and eigenvalue decomposition of the fused data to extract the principal components, which can explain the largest variance in the data, thereby achieving data dimensionality reduction and feature extraction.

[0125] The system uses a trained model or clustering algorithm to identify feature combinations related to cable fault risks and construct a multi-source risk feature matrix. The multi-source risk feature matrix integrates the multi-dimensional features of waterproof degradation and sheath circulation, providing a comprehensive feature basis for subsequent fault location and risk assessment. The system will regularly retrain and optimize the feature fusion model to adapt to changes in cable operating conditions and the accumulation of new data, ensuring the accuracy and effectiveness of the feature fusion results. During the feature fusion process, the system automatically monitors the stability and consistency of the fusion results, and evaluates the fusion effect by calculating the data correlation and error indicators before and after fusion. If the fusion effect is found to be poor, the system will automatically adjust the parameters of the fusion algorithm or select a different fusion method and re-perform feature fusion until the fusion result meets the preset performance indicators.

[0126] S53: Perform cable topology analysis on the multi-source risk feature matrix, locate the cable fault coordinate point through the space vector, generate fault point coordinate data, perform credibility verification on the fault point coordinate data, calculate the confidence interval of the fault point coordinate data, and generate risk probability value data of the cable fault point. The risk probability value data is used to indicate the confidence level of the fault occurrence and the maintenance priority.

[0127] Specifically, the system combines cable geographic information system (GIS) data and network topology to correlate the eigenvectors in the multi-source risk signature matrix with the actual physical location of the cable. GIS data provides spatial information such as the cable's geographic coordinates, direction, and joint locations, while the network topology describes the connections and electrical characteristics between cables. Using a spatial vector location algorithm, the system leverages spatially relevant features in the eigenvectors, such as distance and direction, to pinpoint the precise location of the fault within the cable network.

[0128] The spatial vector location algorithm constructs a spatial vector model of the cable network and matches the risk signatures in the feature vector with the spatial vector to locate the fault point. The system verifies the reliability of the fault point coordinate data by comparing it with known information such as historical fault data and cable joint locations. The system uses statistical methods and machine learning models to calculate confidence intervals for the fault point coordinate data and determine the probability range of the fault.

[0129] Confidence interval calculations are based on the uncertainty estimates of the eigenvectors and the confidence level of the model predictions, generating risk probability data for the cable fault point. This risk probability data not only indicates the likelihood of a fault but also reflects repair priorities, helping maintenance personnel rationally allocate repair resources and sequence. The system visualizes the fault point coordinates and risk probability data on a cable topology map, providing an intuitive basis for maintenance decisions. The system also continuously updates the cable topology analysis model to adapt to the expansion and changes of the cable network, ensuring accurate and reliable fault location. During the fault location process, the system automatically records the location time, location method, and related parameter settings for each fault point, enabling traceability and optimization during subsequent maintenance work. Furthermore, the system regularly evaluates and verifies the fault point coordinates and risk probability data, continuously improving the accuracy and reliability of the fault location algorithm through comparative analysis with actual repair results.

[0130] In summary, the present application provides a method for online cable fault location, monitoring and early warning based on a neural network algorithm, which can systematically solve the core defects of multi-source data timing inaccuracy and insufficient fault location accuracy in the background technology through the collaborative innovation of signal analysis and spatial modeling: based on the refined decomposition and processing of the fundamental and harmonic characteristics of the circulating current waveform, it can achieve the purpose of separating the implicit fault phase information in the sheath circulating current and eliminate the interference of signal mixing on the positioning accuracy; through the multi-dimensional fusion of the waterproof degradation level and the circulating current phase characteristics, it can construct a risk correlation feature space covering the sealing status and electrical parameters, and realize the quantitative correlation modeling of humidity change and insulation degradation; finally, combined with the spatial vector analysis of the cable topology structure, it can accurately locate the fault coordinate point and generate a risk probability value quantified by the confidence interval, so as to realize the scientific judgment of fault location accuracy and maintenance priority. This technical system successfully opens up the full-chain analysis path from waterproof sealing status monitoring to precise fault location, significantly improves the reliability of spatial coordinate positioning, and provides a decision-making basis for power grid operation and maintenance with both spatial accuracy and confidence assessment.

[0131] In one embodiment, S6 of a method for locating, monitoring and warning of cable online faults based on a neural network algorithm provided by the present invention specifically includes the following steps:

[0132] S61: Perform false alarm pattern recognition processing on historical alarm records, extract common feature items in high-frequency false alarm events, and generate false alarm feature templates.

[0133] Specifically, the system extracts historical alarm records from the database. These records contain information such as alarm time, alarm type, alarm location, and subsequent verification results. The system preprocesses the alarm records, including removing duplicate records, filling in missing fields, and correcting erroneous data to ensure data integrity and accuracy. Subsequently, the system uses a clustering analysis algorithm to cluster high-frequency false alarm events. Clustering is based on the feature vectors of the alarm events, which include factors such as the periodicity of alarm time, the combination of alarm types, and the concentration of alarm locations. By setting appropriate clustering parameters, the system clusters similar false alarm events into different groups. For each group, the system extracts common features, which may represent a specific time period, a specific combination of alarm types, or concentrated alarms in a specific area. The system integrates these common features into false alarm feature templates. These false alarm feature templates are stored in the system as rule sets for subsequent alarm filtering and identification.

[0134] Optimally, the system regularly updates and maintains false alarm signature templates to adapt to changes in the cable operating environment and the accumulation of new alarm data. The system also evaluates the effectiveness of false alarm signature template recognition, calculating the accuracy and recall rate of false alarm recognition to optimize clustering algorithm parameters and feature extraction methods, thereby improving false alarm pattern recognition performance. When processing historical alarm records, the system automatically records the handling process and results of each alarm event, allowing for tracing and optimization in subsequent analysis.

[0135] S62: Perform deviation quantification processing on the risk probability value data and the insulation resistance threshold data of the original monitoring data set, calculate the accuracy offset between the predicted value and the actual alarm, and generate a parameter credibility score.

[0136] Specifically, the system aligns the risk probability data and the insulation resistance data from the original monitoring dataset in chronological order, ensuring that the two are compared over the same time dimension. The system calculates the deviation between the predicted risk probability value and the actual monitored insulation resistance value to quantify the accuracy offset between the predicted value and the actual alarm. Furthermore, the system calculates the average deviation and standard deviation of all data points to assess the overall accuracy of the prediction. Based on the size and distribution of the deviation, the system generates a parameter credibility score.

[0137] The system stores parameter credibility scores for subsequent updates and optimization of the monitoring strategy parameter library. The system regularly evaluates the results of the deviation quantification process and verifies the rationality of the parameter credibility scores by comparing and analyzing them with actual fault data. If the scores are found to be inconsistent with the actual situation, the system automatically adjusts the deviation calculation formula or scoring method and reprocesses them until the scores accurately reflect the accuracy relationship between the predicted value and the actual alarm. During the deviation quantification process, the system automatically records the deviation value and credibility score of each data point for traceability and optimization in subsequent analysis.

[0138] S63: Adaptively update the monitoring strategy parameter library based on the gradient descent algorithm and the parameter credibility score, and update the benchmark weight and insulation resistance threshold data of the monitoring strategy parameter library.

[0139] Specifically, the system extracts the current baseline weight and insulation resistance threshold data from the monitoring strategy parameter library. Using the parameter credibility score as the objective function of the gradient descent algorithm, the system adjusts the baseline weight and insulation resistance threshold through an optimization algorithm to maximize the parameter credibility score. Through multiple iterative updates, the system gradually adjusts the baseline weight and insulation resistance threshold to adapt to changes in the cable's operating status.

[0140] During the update process, the system monitors the effects of parameter adjustments in real time and evaluates the rationality of the update by calculating the matching degree between the updated parameter credibility score and the actual alarm data. If the updated parameters cause the credibility score to drop or the matching degree with the actual alarm data to decrease, the system automatically reduces the learning rate or adjusts the direction of gradient descent and re-updates. The system also stores the updated baseline weight and insulation resistance threshold data in the monitoring strategy parameter library and implements version control on the parameter library to enable rollback to the previous version if necessary.

[0141] During the adaptive update process, the system automatically records each updated parameter value, credibility score, and update time for traceability and optimization in subsequent analysis. The system also regularly conducts comprehensive evaluations of the monitoring strategy parameter library, verifying the rationality and effectiveness of the parameters through comparative analysis with historical fault data and actual operating conditions, ensuring that the monitoring strategy parameter library can continue to support reliable cable operation.

[0142] Preferably, if Figure 5 As shown, the present invention provides a cable online fault location monitoring and early warning system 700 based on a neural network algorithm. The system is configured with the following modules:

[0143] The data acquisition and preprocessing module 710 is used to collect the ambient temperature data, ambient humidity data, sheath circulating current data, and partial discharge pulse signal data of the cable operating environment through multi-source sensors deployed along the cable, generate multi-source environmental data, and perform noise reduction and time series alignment on the multi-source environmental data to generate the original monitoring data set;

[0144] Data seasonality analysis module 720, for performing seasonal fluctuation intensity analysis on the original monitoring data set based on a neural network algorithm, calling a preset monitoring strategy parameter library to perform dynamic baseline compensation processing, and generating a dynamic environmental baseline data set that removes seasonal effects;

[0145] The load-insulation mapping module 730 is used to perform load-insulation correlation mapping processing on the dynamic environment reference data set, calculate the insulation resistance correction value, and generate insulation performance evaluation data with temperature compensation characteristics;

[0146] The sealing and waterproofing test module 740 is used to perform a quantitative waterproofing and sealing evaluation on the insulation performance evaluation data and the original monitoring data set, call the insulation resistance threshold data in the monitoring strategy parameter library to perform deviation verification, and generate waterproofing degradation level data reflecting the degradation of the sealing performance;

[0147] Fault risk analysis module 750, used to process the waterproof degradation level data and the original monitoring data set, extract the risk characteristics of the cable operation status and perform joint feature analysis to generate fault point coordinate data and risk probability value data;

[0148] The parameter updating module 760 is used to perform credibility verification processing on the risk probability value data, and update the benchmark weight and insulation resistance threshold data of the monitoring strategy parameter library based on the accuracy deviation between the historical alarm records and the fault point coordinate data.

[0149] In summary, the cable online fault location, monitoring and early warning system based on the neural network algorithm provided by the present application can effectively overcome the core defects of the existing cable monitoring technology through systematic data processing and closed-loop optimization mechanism: first, with the help of noise reduction and time alignment processing of multi-source sensor data, the spatiotemporal synchronous fusion of heterogeneous signals such as temperature and circulation can be achieved, eliminating the feature dislocation problem caused by sampling frequency differences; through seasonal fluctuation intensity analysis and dynamic benchmark compensation processing of the original monitoring data, the nonlinear correction of periodic interference of environmental temperature and humidity can be achieved, and the insulation resistance measurement deviation can be significantly suppressed; the temperature compensation mechanism based on load-insulation correlation mapping is used to decouple the coupling effect of dynamic load and temperature drift, so that the insulation performance evaluation value truly reflects the actual state of the cable; through quantitative evaluation of waterproof sealing and verification of insulation resistance threshold deviation, a quantitative correlation model of humidity change and insulation degradation can be established, and the sealing state monitoring can be converted into a quantifiable maintenance basis; finally, combined with the closed-loop verification of historical alarm records of fault point coordinate accuracy deviation analysis, dynamic optimization and update of warning thresholds and compensation parameters can be achieved, forming an intelligent early warning system that adaptively evolves with the cable aging process. This full-process technical system can comprehensively improve the accuracy of fault location and the reliability of insulation status assessment, and provide monitoring support with continuous evolution capabilities for the power grid active defense system.

[0150] Preferably, the data acquisition and preprocessing module 710 provided in this application is configured with the following units:

[0151] A multi-source data acquisition unit is used to collect ambient temperature data, ambient humidity data, sheath circulating current value data and partial discharge pulse signal data of the cable operating environment through multi-source sensors deployed along the cable to generate multi-source environmental data;

[0152] The timing deviation compensation unit is used to process the ambient temperature data and ambient humidity data of multi-source environmental data, mark the timestamps and perform adaptive window function compensation to eliminate the timing deviation caused by the sampling frequency differences of multiple sensors and generate time-aligned timing environmental parameters;

[0153] The status data purification unit is used to perform sliding window filtering on the sheath circulating current value data and partial discharge pulse signal data of the cable operation status in the multi-source environmental data, eliminate high-frequency interference through wavelet threshold noise reduction, and generate purified status data;

[0154] The multidimensional data fusion unit is used to perform multidimensional data fusion processing on the time series environmental parameters and purification status data, build a monitoring data matrix with a unified time stamp, and generate an original monitoring data set.

[0155] Preferably, the data seasonality analysis module 720 provided in this application is configured with the following units:

[0156] Seasonal feature extraction unit, used to divide the original monitoring data set into quarterly periods, extract seasonal characteristic values ​​of ambient temperature data and ambient humidity data, and generate seasonal fluctuation characteristics. Seasonal fluctuation characteristics are used to indicate the temperature and humidity fluctuation amplitude and periodic variation pattern of the cable operating environment during the seasonal transition process;

[0157] The benchmark compensation coefficient generation unit is used to process the seasonal fluctuation characteristic vector, call the historical benchmark data of the preset monitoring strategy parameter library to perform dynamic weight compensation processing, calibrate the offset through the sliding window standard deviation, and generate the benchmark compensation coefficient;

[0158] The dynamic environmental benchmark generation unit is used to perform baseline reconstruction on the original monitoring data set based on the benchmark compensation coefficient, reconstruct the long-term trend component of the environmental parameters, and generate a dynamic environmental benchmark data set. The dynamic environmental benchmark data set is used to indicate the basic environmental status parameters of the cable after eliminating seasonal fluctuation interference.

[0159] Preferably, the seasonal feature extraction unit includes a time series data normalization subunit, a long-period feature learning subunit, a time node weighting subunit, and a seasonal feature decoding subunit. Among them, the time series data normalization subunit is used to perform time series reorganization processing on the environmental parameters of the original monitoring data set, align and standardize the environmental temperature data and humidity data according to the time axis, and generate a three-dimensional time series input tensor with channel dimension separation; the long-period feature learning subunit is used to perform long-period feature extraction processing on the three-dimensional time series input tensor, learn the temporal dependency of seasonal fluctuations through a bidirectional LSTM network layer, and generate an initial time series feature vector containing cross-period coupling features; the time node weighting subunit is used to perform key period focus processing on the initial time series feature vector, calculate the seasonal fluctuation contribution weights of different time nodes using the attention mechanism layer, and generate weighted time series features with time node sensitivity; the seasonal feature decoding subunit is used to perform seasonal feature decoding processing on the weighted time series features, map them to the temperature fluctuation amplitude, humidity change cycle, and seasonal alternation speed feature space through a fully connected network layer, and generate a seasonal fluctuation feature vector, which is used to indicate the seasonal fluctuation intensity characteristics.

[0160] Preferably, the load-insulation mapping module 730 provided in this application is configured with the following units:

[0161] The load feature extraction unit is used to extract the current load features of the dynamic environment benchmark data set, identify the peak and valley waveforms of the cable load parameters, mark the periods of abnormal load, and generate a load pattern feature matrix reflecting the load change law;

[0162] The temperature drift correction unit is used to process the temperature gradient data of the dynamic environmental reference data set based on a linear compensation algorithm to eliminate the temperature drift effect and generate temperature-corrected insulation parameters. The temperature-corrected insulation parameters are used to eliminate insulation resistance measurement deviations caused by ambient temperature fluctuations.

[0163] The insulation performance evaluation unit is used to perform correlation mapping processing on the load mode characteristic matrix and the temperature-corrected insulation parameters, calculate the impact factor of load change on insulation performance, and generate insulation performance evaluation data with temperature compensation characteristics.

[0164] Preferably, the sealing and waterproof testing module 740 provided in this application is configured with the following units:

[0165] The high humidity area detection unit is used to detect and process the ambient humidity data in the original monitoring data set, identify the points where the humidity exceeds the standard continuously, generate a spatial cluster distribution map, and generate regional high humidity alarm information. The regional high humidity alarm information is used to indicate the geographical spatial distribution range of abnormal humidity accumulation in the cable channel;

[0166] The seal failure feature extraction unit is used to extract and process seal failure features from regional high humidity alarm information, calculate the degradation slope based on the humidity change rate over three days, and generate a seal attenuation index;

[0167] The waterproof degradation level assessment unit is used to perform threshold deviation analysis on the sealing attenuation index, call the insulation resistance threshold data in the monitoring strategy parameter library to quantify the degree of degradation of the waterproof performance, and generate waterproof degradation level data reflecting the degree of degradation of the waterproof sealing performance. The waterproof degradation level data is used to indicate the aging decline level of the cable outer sheath waterproof structure.

[0168] Preferably, the fault risk analysis module 750 provided in this application is configured with the following units:

[0169] The circulation characteristic decomposition unit is used to perform waveform characteristic decomposition processing on the sheath circulation value data of the original monitoring data set, separate the fundamental component and harmonic characteristic parameters, and generate the circulation phase amplitude characteristics;

[0170] The multi-source risk feature fusion unit is used to perform multi-dimensional feature fusion processing on the waterproof degradation level data and the circulation phase amplitude characteristics, construct the risk association feature space, and generate the multi-source risk feature matrix;

[0171] The fault location and verification unit is used to perform cable topology analysis on the multi-source risk feature matrix, locate the cable fault coordinate point through the space vector, generate the fault point coordinate data, perform credibility verification on the fault point coordinate data, calculate the confidence interval of the fault point coordinate data, and generate the risk probability value data of the cable fault point. The risk probability value data is used to indicate the confidence level of the fault occurrence and the maintenance priority.

[0172] Preferably, the parameter updating module 760 provided in this application is configured with the following units:

[0173] The false alarm feature extraction unit is used to perform false alarm pattern recognition processing on historical alarm records, extract common feature items in high-frequency false alarm events, and generate false alarm feature templates;

[0174] A credibility score generating unit is used to perform deviation quantification processing on the risk probability value data and the insulation resistance threshold data of the original monitoring data set, calculate the accuracy offset between the predicted value and the actual alarm, and generate a parameter credibility score;

[0175] The monitoring strategy updating unit is used to adaptively update the monitoring strategy parameter library based on the gradient descent algorithm and the parameter credibility score, and update the benchmark weight and insulation resistance threshold data of the monitoring strategy parameter library.

[0176] In one embodiment, the present application also provides a computer device including a memory and a processor, wherein the memory stores a computer program, and when the processor executes the computer program, the above-mentioned cable online operation fault location, monitoring and early warning method based on the neural network algorithm is implemented.

[0177] In one embodiment, the present application also provides a computer-readable storage medium having a computer program stored thereon. When the computer program is executed by a processor, the above-mentioned cable online operation fault location, monitoring and early warning method based on the neural network algorithm is implemented.

[0178] In the description of this specification, the reference terms "one embodiment," "some embodiments," "example," "specific example," or "some examples" mean that the specific features, structures, materials, or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present application. Moreover, the specific features, structures, materials, or characteristics described may be combined in any appropriate manner in any one or more embodiments or examples. In addition, those skilled in the art may combine and integrate different embodiments or examples described in this specification, as well as features of different embodiments or examples, unless they are mutually inconsistent.

[0179] For the device embodiments, since they basically correspond to the method embodiments, the relevant parts can be referred to the partial description of the method embodiments. The device embodiments described above are merely illustrative, wherein the components described as separate parts may or may not be physically separated, and the parts displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules can be selected according to actual needs to achieve the purpose of the disclosed solution. A person of ordinary skill in the art can understand and implement it without expending creative work.

[0180] The above description is merely a specific embodiment of the present application, but the scope of protection of the present application is not limited thereto. Any person skilled in the art can easily conceive of various modifications or substitutions within the technical scope disclosed in this application, and such modifications or substitutions should be included within the scope of protection of the present application. Therefore, the scope of protection of the present application should be based on the scope of protection of the claims.

Claims

1. A method for locating, monitoring and warning of cable online faults based on a neural network algorithm, characterized in that: The following steps are involved: S1: Multi-source sensors deployed along the cable collect ambient temperature data, ambient humidity data, sheath circulating current data, and partial discharge pulse signal data of the cable operating environment to generate multi-source environmental data. The multi-source environmental data is then subjected to noise reduction and time series alignment to generate an original monitoring data set. S2: Perform seasonal fluctuation intensity analysis on the original monitoring data set based on a neural network algorithm, call a preset monitoring strategy parameter library to perform dynamic benchmark compensation processing, and generate a dynamic environmental benchmark data set that removes seasonal effects; S3: performing load-insulation correlation mapping processing on the dynamic environment benchmark data set, calculating the insulation resistance correction value, and generating insulation performance evaluation data with temperature compensation characteristics; S4: performing waterproof sealing quantitative evaluation processing on the insulation performance evaluation data and the original monitoring data set, calling the insulation resistance threshold data of the monitoring strategy parameter library to perform deviation verification, and generating waterproof degradation level data reflecting the degradation of sealing performance; S5: Processing the waterproof degradation level data and the original monitoring data set, extracting the risk characteristics of the cable operation status and performing joint feature analysis to generate fault point coordinate data and risk probability value data; S6: Perform credibility verification processing on the risk probability value data, and update the benchmark weight and insulation resistance threshold data of the monitoring strategy parameter library based on the accuracy deviation between the historical alarm record and the fault point coordinate data.

2. The method according to claim 1, characterized in that Said S1 comprises: S11: Multi-source sensors deployed along the cable collect ambient temperature data, ambient humidity data, sheath circulating current data, and partial discharge pulse signal data of the cable operating environment to generate multi-source environmental data. S12: Processing the ambient temperature data and ambient humidity data of the multi-source environmental data, marking timestamps and performing adaptive window function compensation to eliminate timing deviations caused by differences in multi-sensor sampling frequencies, and generating time-aligned timing environmental parameters; S13: performing sliding window filtering on the sheath circulating current value data and partial discharge pulse signal data of the cable operation status in the multi-source environmental data, eliminating high-frequency interference through wavelet threshold noise reduction, and generating purification status data; S14: performing multi-dimensional data fusion processing on the time series environmental parameters and the purification status data, constructing a monitoring data matrix with a unified time stamp, and generating an original monitoring data set.

3. The method according to claim 1, characterized in that The S2 includes: S21: Inputting the original monitoring data set into a pre-trained neural network model for quarterly period processing, extracting seasonal characteristic values ​​of the ambient temperature data and the ambient humidity data, and generating seasonal fluctuation characteristics, wherein the seasonal fluctuation characteristics are used to indicate the temperature and humidity fluctuation amplitude and periodic variation pattern of the cable operating environment during the seasonal transition process; S22: managing the seasonal fluctuation characteristic vector, calling the historical benchmark data of the preset monitoring strategy parameter library to perform dynamic weight compensation processing, calibrating the offset through the sliding window standard deviation, and generating a benchmark compensation coefficient; S23: Baseline reconstruction processing is performed on the original monitoring data set based on the benchmark compensation coefficient, and the long-term components of the environmental parameters are reconstructed using an encoder-decoder network to generate a dynamic environmental benchmark data set, which is used to indicate the basic environmental status parameters of the cable after eliminating seasonal fluctuation interference.

4. The method according to claim 3, characterized in that The S21 includes: S211: Performing time series reorganization processing on the environmental parameters of the original monitoring data set, aligning and standardizing the environmental temperature data and humidity data along the time axis, and generating a three-dimensional time series input tensor with channel dimension separation; S212: performing long-period feature extraction processing on the three-dimensional time series input tensor, learning the temporal dependency of seasonal fluctuations through a bidirectional LSTM network layer, and generating an initial time series feature vector containing cross-period coupling features; S213: performing key period focusing processing on the initial time series feature vector, calculating the seasonal fluctuation contribution weights of different time nodes using the attention mechanism layer, and generating a weighted time series feature with time node sensitivity; S214: Perform seasonal feature decoding processing on the weighted time series features, map them to the temperature fluctuation amplitude, humidity change period and seasonal alternation speed feature space through a fully connected network layer, and generate a seasonal fluctuation feature vector. The seasonal fluctuation feature vector is used to indicate the seasonal fluctuation intensity characteristics.

5. The method according to claim 3, characterized in that The reference compensation coefficient is obtained by the following formula: Among them, λ is the benchmark compensation coefficient, n is the number of days of the sliding window, and w t is the time decay weight factor, S t is the characteristic value of the current season in the seasonal fluctuation characteristics, H t is the historical benchmark value of the same period in the historical benchmark data, γ is the preset calibration coefficient, σ t is the sliding window standard deviation.

6. The method according to claim 1, characterized in that The S3 includes: S31: performing current load feature extraction processing on the dynamic environment benchmark data set, identifying the peak and valley waveforms of the cable load parameters and marking the periods of abnormal load, thereby generating a load pattern feature matrix reflecting the load variation law; S32: Processing the temperature gradient data of the dynamic environment reference data set based on a linear compensation algorithm to eliminate temperature drift effects and generate temperature-corrected insulation parameters, wherein the temperature-corrected insulation parameters are used to eliminate insulation resistance measurement deviations caused by ambient temperature fluctuations; S33: performing correlation mapping processing on the load mode characteristic matrix and the temperature-corrected insulation parameters, calculating the impact factor of load change on insulation performance, and generating insulation performance evaluation data with temperature compensation characteristics.

7. The method according to claim 1, characterized in that The S4 includes: S41: Performing high humidity area detection processing on the ambient humidity data of the original monitoring data set, identifying points where humidity exceeds the standard continuously and generating a spatial cluster distribution map, and generating regional high humidity alarm information, wherein the regional high humidity alarm information is used to indicate the geographical spatial distribution range of abnormal humidity accumulation in the cable channel; S42: extracting seal failure features from the high humidity warning information of the area, calculating a degradation slope based on the humidity change rate over three days, and generating a seal attenuation index; S43: Perform threshold deviation analysis on the sealing attenuation index, call the insulation resistance threshold data of the monitoring strategy parameter library to quantify the degree of degradation of the waterproof performance, and generate waterproof degradation level data reflecting the degree of degradation of the waterproof sealing performance. The waterproof degradation level data is used to indicate the aging decay level of the cable outer sheath waterproof structure.

8. The method according to claim 1, characterized in that The S5 includes: S51: performing waveform feature decomposition processing on the sheath circulating current value data of the original monitoring data set, separating the fundamental component and the harmonic characteristic parameters, and generating the circulating current phase amplitude characteristics; S52: performing multi-dimensional feature fusion processing on the waterproof degradation level data and the circulating current phase amplitude characteristics, constructing a risk association feature space, and generating a multi-source risk feature matrix; S53: Perform cable topology analysis on the multi-source risk feature matrix, locate the cable fault coordinate point through the space vector, generate fault point coordinate data, perform credibility verification on the fault point coordinate data, calculate the confidence interval of the fault point coordinate data, and generate risk probability value data of the cable fault point. The risk probability value data is used to indicate the confidence level of the fault occurrence and the maintenance priority.

9. The method according to any one of claims 1 to 8, characterized in that The S6 includes: S61: Perform false alarm pattern recognition on historical alarm records, extract common feature items in high-frequency false alarm events, and generate false alarm feature templates; S62: performing deviation quantification processing on the risk probability value data and the insulation resistance threshold data of the original monitoring data set, calculating the accuracy offset between the predicted value and the actual alarm, and generating a parameter credibility score; S63: Adaptively update the monitoring strategy parameter library based on the gradient descent algorithm and the parameter credibility score, and update the benchmark weight and insulation resistance threshold data of the monitoring strategy parameter library.

10. A cable online fault location monitoring and early warning system based on a neural network algorithm, characterized in that: The system comprises: The data acquisition and preprocessing module is used to collect the ambient temperature data, ambient humidity data, sheath circulating current value data and partial discharge pulse signal data of the cable operating environment through multi-source sensors deployed along the cable, generate multi-source environmental data, and perform noise reduction and time series alignment on the multi-source environmental data to generate the original monitoring data set; A data seasonality analysis module is used to analyze the seasonal fluctuation intensity of the original monitoring data set based on a neural network algorithm, call a preset monitoring strategy parameter library to perform dynamic benchmark compensation processing, and generate a dynamic environmental benchmark data set that removes seasonal effects; a load-insulation mapping module, configured to perform load-insulation correlation mapping processing on the dynamic environment benchmark data set, calculate insulation resistance correction values, and generate insulation performance evaluation data with temperature compensation characteristics; A sealing and waterproof testing module is used to perform a waterproof and seal quantitative evaluation process on the insulation performance evaluation data and the original monitoring data set, call the insulation resistance threshold data of the monitoring strategy parameter library to perform deviation verification, and generate waterproof degradation level data reflecting the degradation of sealing performance; A fault risk analysis module is used to process the waterproof degradation level data and the original monitoring data set, extract the risk characteristics of the cable operation status and perform joint feature analysis to generate fault point coordinate data and risk probability value data; A parameter updating module is used to perform credibility verification processing on the risk probability value data, and update the benchmark weight and insulation resistance threshold data of the monitoring strategy parameter library based on the accuracy deviation between the historical alarm records and the fault point coordinate data.

Images