Flexible cable fault hidden danger real-time identification and early warning method for distribution network operation without power outage

Abstract

The invention, which relates to the technical field of power cable fault detection and early warning, discloses a real-time identification and early warning method for fault hidden troubles of a flexible cable for power distribution network operation without power outage, comprising the following steps: acquiring and calibrating partial discharge, temperature, dielectric loss, humidity and current data of a cable through a distributed sensor; data space-time alignment is realized by combining sensor sensitivity correction; combining with parameter deviation degree dynamic weighting fusion features, and highlighting abnormal parameter contributions; then coupling time-space correlation to analyze the feature correlation degree, and avoiding misjudgment of a single time-space point; the hidden danger membership degree is judged by integrating the correlation degree change rate, and the judgment threshold value is dynamically adjusted by combining the cable load; and finally, yellow, orange and red three-level early warning is output in combination with the short-term prediction deviation. The method overcomes the defects of inaccurate data, fixed weight, rigid threshold and the like in the prior art, can accurately identify hidden dangers in real time and perform graded early warning, provides a clear basis for operation and maintenance, and ensures safe and stable operation of the flexible cable in a distribution network power failure operation scene.

Description

Technical Field

[0001] This invention relates to the field of power cable fault detection and early warning technology, specifically a method for real-time identification and early warning of potential faults in flexible cables used in power distribution network uninterrupted operation. Background Technology

[0002] As the core end-point link connecting users in the power system, the safe and stable operation of the distribution network directly determines the reliability of electricity supply for social production and daily life. Live-line working (LLM) technology, which can completely avoid the economic losses and inconveniences caused by power outages to industrial production and residents, has become the mainstream mode of distribution network operation and maintenance. Flexible cables, with their high flexibility, excellent bending performance, and convenient installation, are widely used in LLM scenarios, mainly for temporary power supply and equipment bypass connections. However, because they are constantly in a energized state, they are susceptible to various factors such as real-time load fluctuations, changes in outdoor humidity, natural aging of the insulation layer, and mechanical wear. These factors can gradually lead to potential faults such as abnormal partial discharge pulse amplitude, abnormally high surface temperature, and increased dielectric loss of the insulation layer. If these potential faults are not identified and warned of in time, they will gradually develop into serious faults such as cable insulation breakdown and short circuits, not only forcing the interruption of LLM operations but also potentially causing partial power outages in the distribution network, threatening the safe operation of the power grid and the safety of personnel.

[0003] Currently, the identification and early warning technologies for fault hazards in flexible cables used in live-line maintenance of power distribution networks still have significant limitations and shortcomings: Firstly, traditional manual inspection methods rely on the experience and judgment of maintenance personnel, which not only has low inspection efficiency and limited coverage, but also poses a risk of electric shock to personnel approaching live flexible cables in live-line maintenance scenarios. Furthermore, it cannot achieve 24-hour real-time monitoring and is difficult to capture short-term and sudden hazard signals. Some technologies only use single-parameter monitoring, such as monitoring only the cable sheath temperature or partial discharge, without comprehensively considering the synergistic effects of environmental humidity and real-time load on the cable status, which can easily lead to missed hazard detection due to "one-sided monitoring".

[0004] Secondly, existing multi-source data fusion identification technologies have significant drawbacks: on the one hand, they often use fixed weights to fuse parameters such as partial discharge and temperature, and cannot dynamically adjust the weights according to the degree of deviation between the parameters and normal thresholds. This results in the feature contribution of abnormal parameters being masked by normal parameters, affecting the accuracy of hazard identification. On the other hand, they have not designed effective processing mechanisms for the differences in the installation positions of multiple sensors and the deviation of the acquisition clock, resulting in a prominent phenomenon of "spatiotemporal asynchrony" of data. For example, data collected by different sensors at the same time may actually have a time difference of seconds, or the spatial correlation of data from sensors at different locations may not be considered, directly leading to the distortion of subsequent analysis results.

[0005] Third, the adaptability of the hazard identification and early warning process is poor: existing technologies mostly use fixed thresholds to identify hazards without considering the dynamic adjustment of the cable load in real time. When the cable insulation withstands high loads, the insulation withstand capability decreases. If the identification thresholds used in low load scenarios are still applied, problems such as "thresholds that are too high and thus fail to identify hazards" or "thresholds that are too low and thus falsely report faults" are likely to occur. At the same time, the early warning is based only on the current monitoring status and lacks the ability to predict the development trend of hazards in the short term. Maintenance personnel cannot predict the speed of hazard deterioration in advance and can only deal with it passively when the hazard is already serious. It is difficult to achieve "early detection and early intervention" and cannot give full play to the safety guarantee advantages of uninterrupted power supply operations.

[0006] In summary, existing technologies for identifying and warning of fault hazards in flexible cables used in live-line maintenance of distribution networks cannot meet the actual operation and maintenance needs in terms of real-time performance, accuracy, and scenario adaptability. There is an urgent need for a technical solution that can solve the problems of multi-source data calibration and spatiotemporal alignment, dynamic weighted fusion features, combining spatiotemporal correlation and load conditions to determine hazards, and has short-term predictive and graded early warning capabilities, so as to improve the efficiency and reliability of hazard identification and early warning and ensure the safe and stable operation of flexible cables in live-line maintenance scenarios of distribution networks. Summary of the Invention

[0007] The purpose of this invention is to provide a method for real-time identification and early warning of potential faults in flexible cables during live-line maintenance of distribution networks. This method involves collecting and calibrating multi-source data from flexible cables using distributed sensors, achieving spatiotemporal alignment of the data through sensor sensitivity correction, combining dynamic weighted fusion features based on parameter deviation, coupling spatiotemporal correlation analysis features, incorporating the correlation change rate to determine the membership degree of the potential fault, and finally combining short-term prediction deviations to output graded early warnings. This provides real-time and accurate identification and early warning of potential faults in flexible cables during live-line maintenance of distribution networks, assisting maintenance personnel in timely handling to ensure the safe and stable operation of the distribution network.

[0008] To achieve the above objectives, the present invention provides the following technical solution:

[0009] A method for real-time identification and early warning of potential faults in flexible cables during live-line maintenance of power distribution networks, characterized by the following steps:

[0010] S1: Multi-source data acquisition and preliminary calibration to obtain the original data matrix after calibration;

[0011] S2: Spatiotemporal alignment of multi-source data based on sensor sensitivity correction to obtain aligned data;

[0012] S3: Combine the dynamic weighted feature fusion of parameter deviation to calculate the fused feature value;

[0013] S4: Feature correlation analysis of coupled spatiotemporal correlation, outputting feature correlation degree;

[0014] S5: Determine the membership degree of potential faults by incorporating the correlation change rate;

[0015] S6: Combine the tiered early warning system based on short-term forecast deviations to generate an early warning index and push early warning information.

[0016] Step S1: Multi-source data acquisition and preliminary calibration to obtain the calibrated raw data matrix. Specifically, a distributed sensor array is used to acquire the partial discharge pulse amplitude of the flexible cable at fixed acquisition intervals. Skin temperature Insulation layer dielectric loss value Ambient relative humidity and real-time load current The collection time is A preliminary calibration of the sensor's zero-drift error is performed, and the original data matrix after calibration is defined. for:

[0017] ;

[0018] In the formula, For the first The calibrated raw data matrix at each acquisition time includes five types of parameters: partial discharge, temperature, dielectric loss, humidity, and current. For the first Each data collection moment, This represents the total number of data collection points. This refers to the amplitude of the partial discharge pulse in the flexible cable. For zero-drift calibration coefficient, This serves as the zero-drift reference value for the partial discharge sensor. For the first The surface temperature of the flexible cable at each sampling moment. This is the zero-drift calibration coefficient for the temperature sensor. This serves as the zero-drift reference value for the temperature sensor. For the first The dielectric loss value of the flexible cable insulation layer at each sampling time. This is the zero-drift calibration coefficient for the dielectric loss sensor. This is the zero-drift reference value for the dielectric loss sensor. For the first The relative humidity of the environment where the flexible cable is located at each data acquisition moment. This is the zero-drift calibration coefficient for the humidity sensor. This is the zero-drift reference value for the humidity sensor. For the first The real-time load current of the flexible cable at each acquisition moment.

[0019] Step S2 involves spatiotemporal alignment of multi-source data based on sensor sensitivity correction to obtain aligned data. Specifically, this considers differences in sensor installation location, acquisition clock deviation, and sensitivity differences. Perform spatiotemporal alignment, define and obtain the aligned data. Its formula is:

[0020] ;

[0021] In the formula, For the first The spatiotemporally aligned data vector at each acquisition time contains the aforementioned five types of parameters. , For respectively the first , The sensitivity coefficient of each sensor, For the first The collection point for the first Spatiotemporal influence weight of each collection point , For the first The spatial coordinates of each sensor , For the first The spatial coordinates of each sensor For the first Each data collection moment, For the spatiotemporal equivalent propagation speed of data, This represents the average dielectric loss value of the insulation layer within 1 hour during step S1. This represents a small perturbation term.

[0022] Step S3 combines the dynamic weighted feature fusion of parameter deviation to calculate the fused feature value, specifically based on... Effective features of five types of parameters are extracted, and the deviation of each parameter from the normal threshold is introduced as a weight adjustment factor to calculate the fusion feature value. Its formula is:

[0023] ;

[0024] In the formula, For the first The fusion feature value at each acquisition time, with a value range of [value range missing]. , For parameter category index, These correspond to five types of parameters: partial discharge pulse amplitude, skin temperature, insulation layer dielectric loss, ambient relative humidity, and real-time load current. For the first The first data collection time Dynamic weights of class parameters For the first The base weight of the class parameter, initially set to... , , , , , For the first The deviation weighting coefficient of the class parameter has a range of values. , For the first The first data collection time Deviation of class parameters For the first The first data collection time Aligned data values ​​of class parameters For the first The minimum value for normal operation of class parameters. For the first The maximum normal value of the class parameter. For the first The characteristic nonlinear adjustment index of the class parameter has a range of values. .

[0025] Basic weights in step S3 The correction is made every 12 hours based on the accuracy of hazard identification; the correction amount is... , For the first The accuracy of identifying potential hazards individually based on class parameters. The revised version must meet the following requirements. .

[0026] Step S4 involves performing feature correlation analysis on the coupled spatiotemporal correlation, outputting the feature correlation degree, specifically by combining... The correlation between time series data and the spatial distribution of sensors is used to calculate the feature correlation degree. The formula for quantifying the coupling relationship between current features and historical and adjacent spatial features is as follows:

[0027] ;

[0028] In the formula, For the first The coupling feature correlation degree at each acquisition time, with a value range of [value range missing]. , A value close to 1 indicates a high correlation, while a value close to -1 indicates a low correlation. The length of the time backtracking window. The length of the spatial neighborhood window, with a value range of [value missing]. , For the time backtracking step, For the spatial neighborhood step size, For the spatiotemporal neighborhood weights, satisfying , For the first Each data collection time and its corresponding The average value of spatiotemporal neighborhood fusion feature values The spatiotemporal decay coefficient has a value range of 1. .

[0029] Step S5 incorporates the correlation change rate to determine the membership degree of fault hazards, specifically based on... and Introducing the correlation degree change rate Quantify the development trend of potential hazards and define the membership degree of potential faults. for:

[0030] ;

[0031] Among them, the rate of change of correlation for:

[0032] ;

[0033] In the formula, For the first The membership degree of potential faults at each data acquisition moment. This is the steepness coefficient of the membership curve, with a value range of [value range missing]. , The threshold value for fusion feature values ​​is set to a range of [value range missing]. ,when At that time, the eigenvalues ​​tend to favor the direction of potential hazards. The weight for the rate of change of correlation is [value missing], and its value range is [value missing]. , This is the membership offset coefficient, with a value range of [value range missing]. , For the first The rate of change of correlation at each data collection time. For the first Feature correlation at each data collection time The threshold for determining the membership degree of potential faults. ;when hour, ;when hour, .

[0034] Rate of change of correlation in step S5 If the threshold is exceeded This triggers a second membership check, with the check formula being: .

[0035] Step S6 combines the tiered early warning system based on short-term forecast deviations to generate an early warning index and push out early warning information, specifically based on... and its rate of change Introducing the ARIMA model to predict membership levels for the next 10 minutes ( Deviation from actual value Calculate the early warning index The warning is tiered as follows:

[0036] ;

[0037] Among the prediction bias The formula is as follows:

[0038] ;

[0039] The ARIMA model formula is as follows:

[0040] ;

[0041] In the formula, For the first The early warning index at each data collection time, with a value range of [value missing]. , For the first Rate of change of the membership degree of the hidden danger at each data collection time For the first The membership degree of potential faults at each data acquisition moment. The membership rate change weighting coefficient has a value range of [value range missing]. , This represents the deviation between the predicted membership degree value for the next 10 minutes and the current actual membership degree value. For the prediction of the first based on the ARIMA model The membership degree of potential faults at each data acquisition moment. To predict the step size, As of the date The maximum membership degree prediction deviation at each acquisition time. The prediction bias weighting coefficient has a range of values. , These are the autoregressive coefficients of the ARIMA model. These are the moving average coefficients of the ARIMA model. This represents the error term of the ARIMA model;

[0042] When the warning level is When a yellow alert is issued, An orange alert was issued when A red alert is issued, and the alert information includes the location of the potential hazard, parameter values, and predicted values, which are then pushed through the power distribution network dispatch cloud platform.

[0043] The coefficients of the ARIMA model in step S6 , Retraining every 6 hours using historical membership data, with a training sample size of no less than [number missing]. Data points.

[0044] The specific mechanism of this method is as follows:

[0045] First, a distributed sensor array is used to synchronously collect data on the partial discharge pulse amplitude, sheath temperature, insulation dielectric loss, ambient relative humidity, and real-time load current of the flexible cable. These parameters correspond to the cable's insulation state, thermal state, environmental influencing factors, and operating load, respectively, and are the core basis for identifying potential hazards. Due to the zero-drift error of the sensors, a zero-drift calibration coefficient is used to correct the original data, resulting in a calibrated data matrix. This eliminates hardware error interference for subsequent analysis, ensuring the accuracy of the data foundation. To address the issues of different sensor installation locations, asynchronous acquisition clocks, and inconsistent sensitivity, a sensor sensitivity coefficient is introduced to correct data amplitude deviations. The contribution of data from different acquisition points is adjusted through spatiotemporal influence weighting, and the spatiotemporal dimension measurement is unified by combining the data's spatiotemporal equivalent propagation speed. Finally, spatiotemporally aligned data is output. Subsequently, based on the spatiotemporally aligned data, effective features of various parameters are extracted: on the one hand, the weights are dynamically adjusted through parameter deviation, with parameters exhibiting greater deviation receiving higher weights, thus focusing on abnormal parameters; on the other hand, the contribution of abnormal features is enhanced through nonlinear adjustment exponents, preventing normal parameters from masking abnormal signals. Finally, the features of various parameters are fused into a single fused feature value, achieving the condensation of multi-dimensional information and providing a comprehensive feature basis for hazard identification. Next, the correlation between the current feature and the historical-spatial feature is calculated: the influence of long-term and distant data is reduced by the spatiotemporal neighborhood weight, and the correlation between recent and nearby data is further strengthened by the spatiotemporal decay coefficient. The final output correlation can quantify the trend consistency between the current feature and the surrounding and historical states, and avoid misjudgment caused by the abnormality of a single spatiotemporal point data.

[0046] Using fusion feature value, feature correlation degree, and correlation degree change rate as input, the S-shaped membership function is used to quantitatively determine potential hazards: the closer the fusion feature value is to 1, the higher the degree of abnormality; the larger the absolute value of the correlation degree; the more significant the correlation degree change rate; and the closer the membership degree value is to 1. At the same time, the membership degree threshold is dynamically adjusted according to the real-time load of the cable. The higher the load, the lower the threshold, because cables are more likely to trigger potential hazards under high load. When the membership degree exceeds the threshold, it is determined that there is a potential fault, thus achieving accurate quantitative identification of potential hazards.

[0047] Finally, based on the membership degree of the hidden danger, the rate of change of membership degree, and the membership degree deviation predicted by the ARIMA model for the next 10 minutes, the early warning index is calculated: the higher the membership degree, the faster the rate of change, and the greater the prediction deviation, the higher the early warning index. Then, according to the early warning index, three levels of early warning are divided into yellow, orange, and red. The location of the hidden danger, the current parameter value, and the predicted value are pushed through the power distribution network dispatch cloud platform to provide operation and maintenance personnel with accurate risk level and handling basis, so as to realize timely early warning and response to hidden dangers.

[0048] Compared with the prior art, the beneficial effects of the present invention are:

[0049] By using a spatiotemporal alignment design for multi-source data acquisition calibration and sensor sensitivity correction, the problem of inaccurate data foundation caused by large zero-drift error, spatiotemporal asynchrony, and inconsistent sensitivity of multi-sensor data in the existing technology is solved, which significantly improves the data reliability of subsequent analysis.

[0050] By combining dynamic weighted feature fusion of parameter deviation with feature correlation analysis of coupled spatiotemporal correlation, the limitations of fixed weight fusion and isolated spatiotemporal point judgment in existing technologies are overcome. This approach can highlight the contribution of abnormal parameters to the judgment of hidden dangers, avoid misjudgment caused by single spatiotemporal data anomalies, and improve the accuracy of hidden danger identification.

[0051] By incorporating dynamic membership threshold determination based on correlation change rate and hierarchical early warning combined with short-term prediction deviation, this technology overcomes the shortcomings of existing technologies, such as fixed threshold determination and early warning based only on the current state. It can be adapted to different cable operation load scenarios and can predict the development trend of hidden dangers in advance and quantify the early warning risks. This helps operation and maintenance personnel to handle hidden dangers more timely and accurately, minimize power outage losses caused by flexible cable faults in power distribution network uninterrupted operation scenarios, and ensure the safe and stable operation of the power distribution network. Attached Figure Description

[0052] Figure 1 This is a flowchart of a method for real-time identification and early warning of potential faults in flexible cables for uninterrupted power distribution network operations, according to the present invention. Detailed Implementation

[0053] The technical solutions of the present invention will now be described in detail with reference to the accompanying drawings.

[0054] like Figure 1 As shown, it includes the following steps:

[0055] Step S1: Multi-source data acquisition and preliminary calibration to obtain the calibrated raw data matrix. Specifically, a distributed sensor array is used to acquire the partial discharge pulse amplitude of the flexible cable at fixed acquisition intervals. Skin temperature Insulation layer dielectric loss value Ambient relative humidity and real-time load current The collection time is A preliminary calibration of the sensor's zero-drift error is performed, and the original data matrix after calibration is defined. for:

[0056] ;

[0057] In the formula, For the first The calibrated raw data matrix at each acquisition time includes five types of parameters: partial discharge, temperature, dielectric loss, humidity, and current. For the first Each data collection moment, This represents the total number of data collection points. This refers to the amplitude of the partial discharge pulse in the flexible cable. For zero-drift calibration coefficient, This serves as the zero-drift reference value for the partial discharge sensor. For the first The surface temperature of the flexible cable at each sampling moment. This is the zero-drift calibration coefficient for the temperature sensor. This serves as the zero-drift reference value for the temperature sensor. For the first The dielectric loss value of the flexible cable insulation layer at each sampling time. This is the zero-drift calibration coefficient for the dielectric loss sensor. This is the zero-drift reference value for the dielectric loss sensor. For the first The relative humidity of the environment where the flexible cable is located at each data acquisition moment. This is the zero-drift calibration coefficient for the humidity sensor. This is the zero-drift reference value for the humidity sensor. For the first The real-time load current of the flexible cable at each acquisition moment.

[0058] Step S2 involves spatiotemporal alignment of multi-source data based on sensor sensitivity correction to obtain aligned data. Specifically, this considers differences in sensor installation location, acquisition clock deviation, and sensitivity differences. Perform spatiotemporal alignment, define and obtain the aligned data. Its formula is:

[0059] ;

[0060] In the formula, For the first The spatiotemporally aligned data vector at each acquisition time contains the aforementioned five types of parameters. , For respectively the first , The sensitivity coefficient of each sensor, For the first The collection point for the first Spatiotemporal influence weight of each collection point , For the first The spatial coordinates of each sensor , For the first The spatial coordinates of each sensor For the first Each data collection moment, For the spatiotemporal equivalent propagation speed of data, This represents the average dielectric loss value of the insulation layer within 1 hour during step S1. This represents a small perturbation term.

[0061] Step S3 combines the dynamic weighted feature fusion of parameter deviation to calculate the fused feature value, specifically based on... Effective features of five types of parameters are extracted, and the deviation of each parameter from the normal threshold is introduced as a weight adjustment factor to calculate the fusion feature value. Its formula is:

[0062] ;

[0063] In the formula, For the first The fusion feature value at each acquisition time, with a value range of [value range missing]. , For parameter category index, These correspond to five types of parameters: partial discharge pulse amplitude, skin temperature, insulation layer dielectric loss, ambient relative humidity, and real-time load current. For the first The first data collection time Dynamic weights of class parameters For the first The base weight of the class parameter, initially set to... , , , , , For the first The deviation weighting coefficient of the class parameter has a range of values. , For the first The first data collection time Deviation of class parameters For the first The first data collection time Aligned data values ​​of class parameters For the first The minimum value for normal operation of class parameters. For the first The maximum normal value of the class parameter. For the first The characteristic nonlinear adjustment index of the class parameter has a range of values. .

[0064] Basic weights in step S3 The correction is made every 12 hours based on the accuracy of hazard identification; the correction amount is... , For the first The accuracy of identifying potential hazards individually based on class parameters. The revised version must meet the following requirements. .

[0065] Step S4 involves performing feature correlation analysis on the coupled spatiotemporal correlation, outputting the feature correlation degree, specifically by combining... The correlation between time series data and the spatial distribution of sensors is used to calculate the feature correlation degree. The formula for quantifying the coupling relationship between current features and historical and adjacent spatial features is as follows:

[0066] ;

[0067] In the formula, For the first The coupling feature correlation degree at each acquisition time, with a value range of [value range missing]. , A value close to 1 indicates a high correlation, while a value close to -1 indicates a low correlation. The length of the time backtracking window. The length of the spatial neighborhood window, with a value range of [value missing]. , For the time backtracking step, For the spatial neighborhood step size, For the spatiotemporal neighborhood weights, satisfying , For the first Each data collection time and its corresponding The average value of spatiotemporal neighborhood fusion feature values The spatiotemporal decay coefficient has a value range of 1. .

[0068] Step S5 incorporates the correlation change rate to determine the membership degree of fault hazards, specifically based on... and Introducing the correlation degree change rate Quantify the development trend of potential hazards and define the membership degree of potential faults. for:

[0069] ;

[0070] Among them, the rate of change of correlation for:

[0071] ;

[0072] In the formula, For the first The membership degree of potential faults at each data acquisition moment. This is the steepness coefficient of the membership curve, with a value range of [value range missing]. , The threshold value for fusion feature values ​​is set to a range of [value range missing]. ,when At that time, the eigenvalues ​​tend to favor the direction of potential hazards. The weight for the rate of change of correlation is [value missing], and its value range is [value missing]. , This is the membership offset coefficient, with a value range of [value range missing]. , For the first The rate of change of correlation at each data collection time. For the first Feature correlation at each data collection time The threshold for determining the membership degree of potential faults. ;when hour, ;when hour, .

[0073] Rate of change of correlation in step S5 If the threshold is exceeded This triggers a second membership check, with the check formula being: .

[0074] Step S6 combines the tiered early warning system based on short-term forecast deviations to generate an early warning index and push out early warning information, specifically based on... and its rate of change Introducing the ARIMA model to predict membership levels for the next 10 minutes ( Deviation from actual value Calculate the early warning index The warning is tiered as follows:

[0075] ;

[0076] Among the prediction bias The formula is as follows:

[0077] ;

[0078] The ARIMA model formula is as follows:

[0079] ;

[0080] In the formula, For the first The early warning index at each data collection time, with a value range of [value missing]. , For the first Rate of change of the membership degree of the hidden danger at each data collection time For the first The membership degree of potential faults at each data acquisition moment. The membership rate change weighting coefficient has a value range of [value range missing]. , This represents the deviation between the predicted membership degree value for the next 10 minutes and the current actual membership degree value. For the prediction of the first based on the ARIMA model The membership degree of potential faults at each data acquisition moment. To predict the step size, As of the date The maximum membership degree prediction deviation at each acquisition time. The prediction bias weighting coefficient has a range of values. , These are the autoregressive coefficients of the ARIMA model. These are the moving average coefficients of the ARIMA model. This represents the error term of the ARIMA model;

[0081] When the warning level is When a yellow alert is issued, An orange alert was issued when A red alert is issued, and the alert information includes the location of the potential hazard, parameter values, and predicted values, which are then pushed through the power distribution network dispatch cloud platform.

[0082] The coefficients of the ARIMA model in step S6 , Retraining every 6 hours using historical membership data, with a training sample size of no less than [number missing]. Data points.

[0083] This embodiment uses 10kV distribution network uninterrupted bypass operation as the application scenario. The flexible cable used in the operation is model YJV22-10kV-1×300mm², with a rated current of [missing information]. The specific implementation process and results are as follows: A distributed array containing 4 sets of sensors was used, and data was collected for 2 hours at a fixed acquisition interval of 1 second, with a total of 7200 acquisition points. The raw data was calibrated according to the sensor zero drift calibration coefficient and reference value set in the document. At the 100th acquisition time, For example, the original data matrix after calibration at that moment is obtained. The real-time load current is 450A. Considering the differences in sensor installation locations, the spatial coordinates of the four sensor sets are as follows: , , , At time 100, corresponding to the second group of sensors, the clock deviation and sensitivity difference, sensor sensitivity correction should be performed according to the document requirements, such as the sensitivity coefficient of the second group of sensors. Spatiotemporal equivalent propagation speed calibration, ultimately The spatiotemporal influence weights of the first group of sensors on the second group of sensors are calculated. Finally, the aligned data is obtained. According to the normal threshold and characteristic nonlinear adjustment index of each parameter set in the document, such as partial discharge parameters... Temperature parameters The dynamic weights of the first type of parameters with a partial discharge parameter deviation of 0.427 were calculated. Thus, the fusion feature values ​​are obtained. Backtrack window according to the set time Spatial Neighborhood Window Time and space decay coefficient The feature correlation degree at time 100 is calculated. Step 1: Introduce the rate of change of correlation degree, and calculate... Then, by combining the membership degree set in the document, the relevant parameters are calculated to obtain the membership degree of the potential fault at that moment. Because the current real-time load current of the cable is Determined according to the file dynamic threshold rule Therefore, a potential fault was identified; the rate of change of membership degree was calculated according to the document method. Predicting the hazard membership degree in the next 10 minutes using the ARIMA model Prediction bias Then, combined with the warning index calculation parameters set in the document, , Received early warning index A yellow alert was issued based on the document's early warning grading standards, and the location of the potential hazard was pushed to the power distribution network dispatch cloud platform as required by the document. The location was at the second group of sensors, with coordinates... The current multi-source parameter values ​​and the predicted membership values ​​for the next 10 minutes guide maintenance personnel to conduct targeted inspections. This embodiment, verified by actual parameters, shows that the method can effectively achieve real-time identification and hierarchical early warning of fault hazards in flexible cables during power outage operations in distribution networks. The data processing flow is clear, the results are accurate, and it is suitable for actual engineering application needs.

Claims

1. A method for real-time identification and early warning of potential faults in flexible cables used in live-line power distribution network operations, characterized in that, The following steps are included: S1: Multi-source data acquisition and preliminary calibration to obtain the original data matrix after calibration; S2: Spatiotemporal alignment of multi-source data based on sensor sensitivity correction to obtain aligned data; S3: Combine the dynamic weighted feature fusion of parameter deviation to calculate the fused feature value; S4: Feature correlation analysis of coupled spatiotemporal correlation, outputting feature correlation degree; S5: Determine the membership degree of potential faults by incorporating the correlation change rate; S6: Combine the tiered early warning system based on short-term forecast deviations to generate an early warning index and push early warning information.

2. The method for real-time identification and early warning of potential faults in flexible cables for live-line power distribution network operations according to claim 1, characterized in that, Step S1: Multi-source data acquisition and preliminary calibration to obtain the calibrated raw data matrix. Specifically, a distributed sensor array is used to acquire the partial discharge pulse amplitude of the flexible cable at fixed acquisition intervals. Skin temperature Insulation layer dielectric loss value Ambient relative humidity and real-time load current The collection time is A preliminary calibration of the sensor's zero-drift error is performed, and the original data matrix after calibration is defined. for: ； In the formula, For the first The calibrated raw data matrix at each acquisition time includes five types of parameters: partial discharge, temperature, dielectric loss, humidity, and current. For the first Each data collection moment, This represents the total number of data collection points. This refers to the amplitude of the partial discharge pulse in the flexible cable. For zero-drift calibration coefficient, This serves as the zero-drift reference value for the partial discharge sensor. For the first The surface temperature of the flexible cable at each sampling moment. This is the zero-drift calibration coefficient for the temperature sensor. This serves as the zero-drift reference value for the temperature sensor. For the first The dielectric loss value of the flexible cable insulation layer at each sampling time. This is the zero-drift calibration coefficient for the dielectric loss sensor. This is the zero-drift reference value for the dielectric loss sensor. For the first The relative humidity of the environment where the flexible cable is located at each data acquisition moment. This is the zero-drift calibration coefficient for the humidity sensor. This is the zero-drift reference value for the humidity sensor. For the first The real-time load current of the flexible cable at each acquisition moment.

3. The method for real-time identification and early warning of potential faults in flexible cables for live-line power distribution network operations according to claim 1, characterized in that, Step S2 involves spatiotemporal alignment of multi-source data based on sensor sensitivity correction to obtain aligned data. Specifically, this considers differences in sensor installation location, acquisition clock deviation, and sensitivity differences. Perform spatiotemporal alignment, define and obtain the aligned data. Its formula is: ； In the formula, For the first The spatiotemporally aligned data vector at each acquisition time contains the aforementioned five types of parameters. , For respectively the first , The sensitivity coefficient of each sensor, For the first The collection point for the first Spatiotemporal influence weight of each collection point , For the first The spatial coordinates of each sensor , For the first The spatial coordinates of each sensor For the first Each data collection moment, For the spatiotemporal equivalent propagation speed of data, This represents the average dielectric loss value of the insulation layer within 1 hour during step S1. This represents a small perturbation term.

4. The method for real-time identification and early warning of potential faults in flexible cables for live-line power distribution network operations according to claim 1, characterized in that, Step S3 combines the dynamic weighted feature fusion of parameter deviation to calculate the fused feature value, specifically based on... Effective features of five types of parameters are extracted, and the deviation of each parameter from the normal threshold is introduced as a weight adjustment factor to calculate the fusion feature value. Its formula is: ； In the formula, For the first The fusion feature value at each acquisition time, with a value range of [value range missing]. , For parameter category index, These correspond to five types of parameters: partial discharge pulse amplitude, skin temperature, insulation layer dielectric loss, ambient relative humidity, and real-time load current. For the first The first data collection time Dynamic weights of class parameters For the first The base weight of the class parameter, initially set to... , , , , , For the first The deviation weighting coefficient of the class parameter has a range of values. , For the first The first data collection time Deviation of class parameters For the first The first data collection time Aligned data values ​​of class parameters For the first The minimum value for normal operation of class parameters. For the first The maximum normal value of the class parameter. For the first The characteristic nonlinear adjustment index of the class parameter has a range of values. .

5. The method for real-time identification and early warning of potential faults in flexible cables for live-line power distribution network operations according to claim 1, characterized in that, Basic weights in step S3 The correction is made every 12 hours based on the accuracy of hazard identification; the correction amount is... , For the first The accuracy of identifying potential hazards individually based on class parameters. The revised version must meet the following requirements. .

6. The method for real-time identification and early warning of potential faults in flexible cables for live-line power distribution network operations according to claim 1, characterized in that, Step S4 involves performing feature correlation analysis on the coupled spatiotemporal correlation, outputting the feature correlation degree, specifically by combining... The correlation between time series data and the spatial distribution of sensors is used to calculate the feature correlation degree. The formula for quantifying the coupling relationship between current features and historical and adjacent spatial features is as follows: ； In the formula, For the first The coupling feature correlation degree at each acquisition time, with a value range of [value range missing]. , A value close to 1 indicates a high correlation, while a value close to -1 indicates a low correlation. The length of the time backtracking window. The length of the spatial neighborhood window, with a value range of [value missing]. , For the time backtracking step, For the spatial neighborhood step size, For the spatiotemporal neighborhood weights, satisfying , For the first Each data collection time and its corresponding The average value of spatiotemporal neighborhood fusion feature values The spatiotemporal decay coefficient has a value range of 1. .

7. The method for real-time identification and early warning of potential faults in flexible cables for live-line power distribution network operations according to claim 1, characterized in that, Step S5 incorporates the correlation change rate to determine the membership degree of fault hazards, specifically based on... and Introducing the correlation degree change rate Quantify the development trend of potential hazards and define the membership degree of potential faults. for: ； Among them, the rate of change of correlation for: ； In the formula, For the first The membership degree of potential faults at each data acquisition moment. This is the steepness coefficient of the membership curve, with a value range of [value range missing]. , The threshold value for fusion feature values ​​is set to a range of [value range missing]. ,when At that time, the eigenvalues ​​tend to favor the direction of potential hazards. The weight for the rate of change of correlation is [value missing], and its value range is [value missing]. , This is the membership offset coefficient, with a value range of [value range missing]. , For the first The rate of change of correlation at each data collection time. For the first Feature correlation at each data collection time The threshold for determining the membership degree of potential faults. ;when hour, ;when hour, .

8. The method for real-time identification and early warning of potential faults in flexible cables for live-line power distribution network operations according to claim 1, characterized in that, Rate of change of correlation in step S5 If the threshold is exceeded This triggers a second membership check, with the check formula being: .

9. The method for real-time identification and early warning of potential faults in flexible cables for live-line power distribution network operations according to claim 1, characterized in that, Step S6 combines the tiered early warning system based on short-term forecast deviations to generate an early warning index and push out early warning information, specifically based on... and its rate of change Introducing the ARIMA model to predict membership levels for the next 10 minutes ( Deviation from actual value Calculate the early warning index The warning is tiered as follows: ； Among the prediction bias The formula is as follows: ； The ARIMA model formula is as follows: ； In the formula, For the first The early warning index at each data collection time, with a value range of [value missing]. , For the first Rate of change of the membership degree of the hidden danger at each data collection time For the first The membership degree of potential faults at each data acquisition moment. The membership rate change weighting coefficient has a value range of [value range missing]. , This represents the deviation between the predicted membership degree value for the next 10 minutes and the current actual membership degree value. For the prediction of the first based on the ARIMA model The membership degree of potential faults at each data acquisition moment. To predict the step size, As of the date The maximum membership degree prediction deviation at each acquisition time. The prediction bias weighting coefficient has a range of values. , These are the autoregressive coefficients of the ARIMA model. These are the moving average coefficients of the ARIMA model. This represents the error term of the ARIMA model; When the warning level is When a yellow alert is issued, An orange alert was issued when A red alert is issued, and the alert information includes the location of the potential hazard, parameter values, and predicted values, which are then pushed through the power distribution network dispatch cloud platform.

10. A method for real-time identification and early warning of potential faults in flexible cables for live-line power distribution network operations according to claim 1, characterized in that, The coefficients of the ARIMA model in step S6 , Retraining every 6 hours using historical membership data, with a training sample size of no less than [number missing]. Data points.

Images