A cable state anomaly detection method and system based on dynamic modeling

Abstract

This invention discloses a method and system for detecting cable condition anomalies based on dynamic modeling, relating to the field of power system condition monitoring and fault diagnosis technology. The method includes acquiring signals synchronously collected by multiple sensors on the cable, detecting abnormal events using the root mean square value of a sliding window, extracting abnormal signal segments, and calculating the arrival time difference and intensity difference of the same signal between different sensors. These differences are compared with a pre-built database of external interference and internal fault propagation features to determine the signal source. If it is external interference, adaptive filtering is used to remove interference components, obtaining a purified signal. Finally, the purified signal, combined with cable body temperature and load current parameters, is input into a cable condition assessment model to output a health score. This method achieves accurate differentiation and adaptive filtering of external electromagnetic interference and internal faults. Combined with multi-parameter fusion assessment, it significantly reduces the false alarm rate and improves the accuracy and reliability of anomaly detection.

Description

Technical Field

[0001] This invention relates to the field of power system condition monitoring and fault diagnosis technology, and in particular to a method and system for detecting cable condition anomalies based on dynamic modeling. Background Technology

[0002] Existing methods for detecting cable anomalies largely rely on static threshold judgments or single-parameter monitoring, making them difficult to adapt to actual cable operating scenarios. Cable operation is affected by multiple factors, including ambient temperature fluctuations, random load changes, and accumulated mechanical stress. Electrical parameters and insulation characteristics exhibit significant time-varying and nonlinear characteristics. Static thresholds are prone to false alarms during normal operating condition fluctuations and to missed alarms in the early stages of insulation degradation when signals do not reach the threshold. Traditional methods mistakenly attribute all abnormal signals to internal cable faults, ignoring external electromagnetic interference from multi-circuit parallel laying and mixed strong and weak current installations. Transient electromagnetic fields generated by load switching on adjacent lines and equipment start-up and shutdown can induce interference pulses similar to partial discharge signals on monitoring sensors, leading to numerous invalid alarms.

[0003] This inability to distinguish between internal and external interference can trigger a "boy who cried wolf" effect, causing maintenance personnel to lose trust in the system and subjectively filter or block alarms. Once a genuine fault signal is masked by interference or occurs simultaneously with interference, it is easily misjudged, leading to missed opportunities for optimal response and ultimately catastrophic power outages. Therefore, there is an urgent need for an anomaly detection method that can dynamically adapt to changes in operating conditions, accurately distinguish between internal and external interference, effectively purify signals, and comprehensively assess the true health status of cables using multiple parameters. Summary of the Invention

[0004] To address the shortcomings of existing technologies, this invention provides a method and system for detecting cable condition anomalies based on dynamic modeling. It aims to solve the problems of high false alarm rate, weak anti-interference ability, and easy omission of real faults caused by static threshold judgment, inability to distinguish between interference and fault, and insufficient signal purification in existing cable condition detection technologies.

[0005] To achieve the above objectives, the present invention provides a method for detecting cable condition anomalies based on dynamic modeling, comprising the following steps:

[0006] S1. Acquire the characteristic signals collected by multiple signal acquisition units on the target cable, and perform time synchronization processing on the acquired signals to obtain the time-synchronized signals;

[0007] S2. Based on the time-synchronized signal, calculate the root mean square value of the signal within the sliding window for each signal acquisition unit, mark the center time of the window where the root mean square value exceeds the preset abnormal threshold as the abnormal event trigger time, identify the signal segment corresponding to the abnormal event trigger time as an abnormal signal, and calculate the arrival time difference and signal strength difference of the same abnormal signal among multiple signal acquisition units.

[0008] S3. Compare the time of arrival difference and signal strength difference with the pre-built external interference propagation feature library and internal interference propagation feature library, and determine whether the source of the signal is external interference or internal fault based on the comparison results.

[0009] S4. When the source of the signal is determined to be external interference, adaptive filtering is used to remove the interference components from the signal to obtain a purified signal.

[0010] S5. Input the purification signal, combined with the target cable's body temperature and load current parameters, into the pre-built cable condition assessment model, and output the cable condition assessment results.

[0011] Furthermore, the multiple signal acquisition units are arranged along the axial direction of the target cable, and each signal acquisition unit integrates a partial discharge sensor, a high-speed analog-to-digital converter, and a BeiDou / GPS high-precision timing module; the process of acquiring the characteristic signals collected by the multiple signal acquisition units on the target cable and performing time synchronization processing on the acquired signals to obtain the time-synchronized signal includes:

[0012] The original analog electrical signals during the operation of the target cable are acquired using a partial discharge sensor;

[0013] Based on the high-precision time synchronization module of the signal acquisition unit, at the instantaneous sampling moment when the analog-to-digital converter completes sampling and latching of the instantaneous sample value of the original analog electrical signal, the standard timestamp is bound to the instantaneous sample value to generate instantaneous sampled data with standard timestamp;

[0014] The analog-to-digital converter performs digital encoding on instantaneous sampled data with standard timestamps, converting it into raw digital signals with standard timestamps;

[0015] The raw digital signal with standard timestamps is fed into a unified timing reference to achieve nanosecond-level time alignment of multi-channel data, resulting in an aligned digital signal.

[0016] The aligned digital signal is input into a digital bandpass filter to filter out power frequency interference and low-frequency noise outside the passband, generating a filtered multi-channel digital signal.

[0017] Using the maximum absolute value of the filtered multi-channel digital signal as a reference, the amplitude values ​​of all remaining channel signals are scaled to the range of negative one to one to generate a normalized digital signal, which serves as the time-synchronized signal.

[0018] Furthermore, based on the time-synchronized signal, the root mean square (RMS) value of the signal within the sliding window is calculated for each signal acquisition unit. The center time of the window where the RMS value exceeds a preset anomaly threshold is marked as the anomaly event trigger time. The signal segment corresponding to the anomaly event trigger time is identified as an anomalous signal. The arrival time difference and signal strength difference of the same anomalous signal among multiple signal acquisition units are calculated, including:

[0019] For the time-synchronized signal of each signal acquisition unit, calculate the root mean square value of the signal within the sliding window, and mark the center time of the window where the root mean square value exceeds the preset abnormal threshold as the abnormal event trigger time to obtain the abnormal event trigger time.

[0020] Extract the signal segment centered on the trigger time of the abnormal event from the time-synchronized signal to obtain the abnormal signal segment;

[0021] The abnormal signal segment of the first signal acquisition unit that triggered the abnormal event is used as the reference signal segment. The cross-correlation function is calculated between the abnormal signal segments of the other signal acquisition units and the reference signal segment. The sampling point offset when the cross-correlation function reaches its maximum value is obtained, and the arrival time difference of the other signal acquisition units relative to the reference signal acquisition unit is obtained.

[0022] Based on the time of arrival difference, the sampling points of the reference signal segment and the other abnormal signal segments are aligned. The amplitude values ​​of each signal acquisition unit are extracted from the aligned sampling points to obtain the aligned amplitude value sequence.

[0023] Extract the maximum absolute value of the abnormal signal segment from each signal acquisition unit in the aligned amplitude value sequence. Using the maximum absolute value of the reference signal acquisition unit as the denominator, calculate the ratio of the maximum absolute value of the remaining signal acquisition units to the maximum absolute value of the reference signal acquisition unit to obtain the signal strength difference.

[0024] Furthermore, the comparison of arrival time differences and signal strength differences with pre-built external interference propagation feature libraries and internal interference propagation feature libraries, and the determination of whether the signal source is external interference or an internal fault based on the comparison results, includes:

[0025] Obtain a pre-built external interference propagation feature library and an internal interference propagation feature library. The external interference propagation feature library stores multiple first feature vectors, each of which corresponds to a first set of arrival time differences and a first set of signal strength differences of an external interference signal among multiple signal acquisition units. The internal interference propagation feature library stores multiple second feature vectors, each of which corresponds to a second set of arrival time differences and a second set of signal strength differences of an internal fault signal among multiple signal acquisition units.

[0026] The time of arrival difference and signal strength difference are combined into a feature vector to be identified. The first Euclidean distance set between the feature vector to be identified and each first feature vector in the external interference propagation feature library, and the second Euclidean distance set between the feature vector to be identified and each second feature vector in the internal interference propagation feature library are calculated using the Euclidean distance formula.

[0027] The algorithm extracts the minimum value from the first Euclidean distance set as the first minimum distance and the second minimum value as the first minimum distance; it also extracts the minimum value from the second Euclidean distance set as the second minimum distance and the second minimum value as the second minimum distance.

[0028] If the first minimum distance is less than the second minimum distance, the source of the signal is determined to be external interference; if the first minimum distance is greater than the second minimum distance, the source of the signal is determined to be an internal fault.

[0029] If the first minimum distance is equal to the second minimum distance, then calculate the first ratio of the first minimum distance to the first minimum distance, and calculate the second ratio of the second minimum distance to the second minimum distance; if the first ratio is less than the second ratio, then the source of the signal is determined to be external interference; if the second ratio is less than or equal to the first ratio, then the source of the signal is determined to be an internal fault.

[0030] Furthermore, the pre-construction process of the external interference propagation feature library and the internal interference propagation feature library includes:

[0031] During the load fluctuation of the interference source cable laid in the same cable trench as the target cable and parallel to the target cable, the time-synchronized signals of multiple signal acquisition units are collected simultaneously to obtain multiple interference source operation signals.

[0032] Sliding window energy detection is performed on the operating signal of each interference source. The center time of the window when the root mean square value of the signal exceeds the interference triggering threshold is marked as the interference triggering time. A fixed length signal segment is extracted with the interference triggering time as the center to obtain multiple interference signal segments.

[0033] For each of the multiple interference signal segments, the time difference of arrival between the multiple signal acquisition units is calculated by the cross-correlation function. The maximum absolute value of the interference signal segment on each signal acquisition unit is extracted and the ratio is calculated to obtain the signal strength difference. The time difference of arrival and the signal strength difference are combined into an interference feature vector to obtain multiple interference feature vectors.

[0034] Multiple interference feature vectors are stored in an external interference propagation feature library;

[0035] Cut a cable sample of the same model and specifications as the target cable and simulate a partial discharge fault on the cable sample;

[0036] The same number and spacing of signal acquisition units as the target cable are arranged along the axial direction of the cable sample to synchronously acquire the characteristic signals when the fault occurs, and obtain the fault simulation signal.

[0037] Sliding window energy detection is performed on the fault simulation signal. The center time of the window where the root mean square value of the signal exceeds the fault triggering threshold is marked as the fault triggering time. A fixed length signal segment is extracted with the fault triggering time as the center to obtain multiple fault signal segments.

[0038] For each fault signal segment among multiple fault signal segments, the time difference of arrival between multiple signal acquisition units is calculated by cross-correlation function. The maximum absolute value of the fault signal segment on each signal acquisition unit is extracted and the ratio is calculated to obtain the signal strength difference. The time difference of arrival and the signal strength difference are combined into a fault feature vector to obtain multiple fault feature vectors.

[0039] Multiple fault feature vectors are stored in an internal interference propagation feature library.

[0040] Furthermore, when the source of the signal is determined to be external interference, adaptive filtering is used to remove the interference components from the signal to obtain a purified signal, including:

[0041] Based on the difference in arrival time, the signal acquisition unit with the earliest abnormal event trigger time is selected from multiple signal acquisition units as the reference sensor, and the time-synchronized signal of the reference sensor is used as the reference input signal.

[0042] Use the time-synchronized signal from any of the other signal acquisition units besides the reference sensor as the main input signal;

[0043] The filter weight vector is initialized to zero. The reference input signal is input to the finite impulse response filter. The finite impulse response filter performs a weighted summation of the reference input signal based on the current filter weight vector and outputs an estimated interference signal.

[0044] The error signal is obtained by subtracting the estimated interference signal from the main input signal;

[0045] The error signal and the reference input signal are used to calculate the weight adjustment amount using the least mean square algorithm, and the filter weight vector is updated. This process is repeated until the mean square value of the error signal converges. The error signal at the end of the iteration is used as the clean signal.

[0046] Furthermore, the step of calculating the weight adjustment amount and updating the filter weight vector by using the least mean square algorithm on the error signal and the reference input signal includes:

[0047] Extract the filter weight vector at the current sampling time and the error signal at the current sampling time;

[0048] Extract the signal amplitudes at the current sampling time and previous sampling times from the reference input signal to construct the reference input signal vector;

[0049] Multiply the preset step size factor by the error signal, and then multiply by the reference input signal vector to obtain the weight adjustment vector;

[0050] Add the weight adjustment vector to the filter weight vector at the current sampling time to obtain the filter weight vector at the next sampling time.

[0051] Furthermore, the cable condition assessment model includes a partial discharge phase distribution map generation module, a convolutional neural network classification module, and a multi-parameter fusion judgment module. The partial discharge phase distribution map generation module converts the purified partial discharge pulse sequence into a discharge number distribution vector and a maximum discharge quantity distribution vector, constructing a two-dimensional map. The convolutional neural network classification module extracts features from the two-dimensional map using pre-trained convolutional layers, pooling layers, and fully connected layers, outputting a discharge type probability distribution. The multi-parameter fusion judgment module concatenates the discharge type probability distribution with the body temperature state code and the load current state code, then outputs a cable health score using a Sigmoid activation function. The purification signal, combined with the target cable's body temperature and load current parameters, is input into the pre-constructed cable condition assessment model to output the cable condition assessment results, including:

[0052] Peak detection is performed on the purification signal to extract the peak amplitude and occurrence time of each pulse in the purification signal, and a pulse sequence containing the peak amplitude and occurrence time is generated.

[0053] The power grid synchronization signal is obtained as the power frequency cycle reference. Based on the phase difference between the occurrence time of each pulse in the pulse sequence and the zero crossing point of the power grid synchronization signal, each pulse is assigned to the corresponding power frequency phase interval. The partial discharge phase distribution map is generated by the partial discharge phase distribution map generation module.

[0054] The partial discharge phase distribution map is input into the pre-trained convolutional layer of the convolutional neural network classification module. Local features are extracted through the convolutional layer, dimensionality is reduced through the pooling layer, and the discharge type probability distribution is output through the fully connected layer to obtain the discharge type probability distribution vector.

[0055] The system acquires the target cable's body temperature and load current values. It then compares the body temperature value with the average of historical temperature data. If the body temperature value is less than the average of historical temperature data, it outputs a temperature status code 0; otherwise, it outputs a temperature status code 1, generating a temperature status code. Similarly, it compares the load current value with the average of historical current data. If the load current value is less than the average of historical current data, it outputs a current status code 0; otherwise, it outputs a current status code 1, generating a current status code.

[0056] The discharge type probability distribution vector, temperature status code, and current status code are input into the multi-parameter fusion judgment module for vector concatenation to form a comprehensive feature vector. The comprehensive feature vector is then used to output the cable health score through the Sigmoid activation function, which constitutes the cable status assessment result.

[0057] Furthermore, the generation of the partial discharge phase distribution map includes:

[0058] The power frequency voltage signal is obtained from the bus of the target cable through a voltage transformer, and the power frequency voltage signal is zero-crossing detected to generate a square wave synchronization signal.

[0059] The duration of a complete cycle of the square wave synchronization signal is obtained as the power frequency cycle duration, and the power frequency cycle duration is divided into multiple phase intervals.

[0060] Calculate the time difference between the occurrence time of each pulse in the pulse sequence and the rising edge of the square wave synchronization signal. Divide the time difference by the power frequency cycle duration and multiply by the total number of phase intervals to obtain the phase interval number of the pulse.

[0061] Initialize a counting array and a maximum value array of the same length as the total number of phase intervals, with each element initially set to zero;

[0062] Traverse the pulse sequence, locate the index position of the counting array based on the phase interval number of each pulse, and increment the count value at the index position by one;

[0063] Compare the peak amplitude of the pulse with the current value at the same index position in the maximum value array, and update the index position of the maximum value array with the larger value;

[0064] After the traversal is complete, the count array is used as the distribution vector of the number of discharges, and the maximum value array is used as the distribution vector of the maximum discharge amount.

[0065] The discharge number distribution vector and the maximum discharge quantity distribution vector are used as two channels to construct a two-dimensional matrix, which constitutes a partial discharge phase distribution map.

[0066] A cable condition anomaly detection system based on dynamic modeling includes:

[0067] Signal acquisition module: used to acquire characteristic signals collected by multiple signal acquisition units on the target cable, and to perform time synchronization processing on the acquired signals to obtain time-synchronized signals;

[0068] Difference Calculation Module: Based on the time-synchronized signal, it calculates the root mean square value of the signal within the sliding window for each signal acquisition unit, marks the center time of the window where the root mean square value exceeds the preset abnormal threshold as the abnormal event trigger time, identifies the signal segment corresponding to the abnormal event trigger time as an abnormal signal, and calculates the arrival time difference and signal strength difference of the same abnormal signal among multiple signal acquisition units.

[0069] Signal source determination module: It is used to compare the time of arrival difference and signal strength difference with the pre-built external interference propagation feature library and internal interference propagation feature library, and determine whether the source of the signal is external interference or internal fault based on the comparison results.

[0070] Interference Removal Module: When the source of a signal is determined to be external interference, it uses adaptive filtering to remove the interference components from the signal and obtain a purified signal.

[0071] Condition assessment module: This module combines the purification signal with the target cable's body temperature and load current parameters and inputs them into a pre-built cable condition assessment model to output the cable condition assessment results.

[0072] The present invention provides a cable condition anomaly detection method and system based on dynamic modeling, the advantages of which are mainly reflected in the following aspects:

[0073] 1. Significantly reduces false alarm rate and accurately distinguishes between external interference and internal faults. This invention utilizes multiple high-precision synchronous sensors deployed along the cable axis. By calculating the differences in arrival time and signal strength of the same abnormal signal between different sensors, it constructs the "spatiotemporal footprint" feature of the signal and compares it with pre-built external interference feature libraries and internal fault feature libraries for judgment. This mechanism can effectively identify external electromagnetic interference generated by load fluctuations of nearby lines, high-frequency communication, etc., avoiding misjudging it as internal insulation defects in the cable, thereby greatly reducing invalid alarms, eliminating the "crying wolf" effect, and increasing the trust of maintenance personnel in the monitoring system.

[0074] 2. Effectively purifies interference signals and achieves multi-parameter fusion health assessment. For signals identified as external interference, this invention employs adaptive filtering to remove interference components, obtaining a purified signal reflecting the cable's own condition. Then, the partial discharge spectrum, body temperature state code, and load current state code generated from the purified signal are vector-fused, and a cable health score is output through an evaluation model. This scheme eliminates the contamination of discharge feature extraction by interference, while introducing operating condition parameters such as temperature and current as auxiliary criteria, making the evaluation results more accurate and better adaptable to the dynamic changes in cable operating conditions, significantly improving the reliability of early insulation degradation warnings. Attached Figure Description

[0075] Figure 1 This is a flowchart illustrating a cable condition anomaly detection method based on dynamic modeling according to the present invention.

[0076] Figure 2 This is a functional block diagram of a cable condition anomaly detection system based on dynamic modeling according to the present invention. Detailed Implementation

[0077] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0078] Please see Figure 1 This invention provides a method for detecting cable condition anomalies based on dynamic modeling, comprising the following steps:

[0079] S1. Acquire the characteristic signals collected by multiple signal acquisition units on the target cable, and perform time synchronization processing on the acquired signals to obtain the time-synchronized signals;

[0080] In this embodiment, the multiple signal acquisition units are arranged along the axial direction of the target cable. Each signal acquisition unit integrates a partial discharge sensor, a high-speed analog-to-digital converter, and a BeiDou / GPS high-precision timing module. The process of acquiring the characteristic signals collected by the multiple signal acquisition units on the target cable and performing time synchronization processing on the acquired signals to obtain a time-synchronized signal includes:

[0081] The original analog electrical signals during the operation of the target cable are collected using a partial discharge sensor.

[0082] Partial discharge sensors are devices used to detect weak pulse discharges in cable insulation, such as ultra-high frequency sensors. The raw analog electrical signal refers to the unprocessed voltage waveform directly output by the sensor, containing information about internal cable faults and external electromagnetic interference. In this step, multiple sensors spaced along the cable axis synchronously acquire analog signals. Because each sensor is spatially positioned relative to nearby interference sources, this provides a physical basis for subsequently distinguishing interference sources using propagation delay and amplitude differences.

[0083] The high-precision time synchronization module based on the signal acquisition unit binds the standard timestamp with the instantaneous sample value at the instantaneous sampling moment when the analog-to-digital converter completes sampling and latching of the original analog electrical signal, generating instantaneous sampling data with the standard timestamp.

[0084] A high-precision time synchronization module refers to a globally unified clock that integrates BeiDou or GPS dual-mode time synchronization circuitry, providing nanosecond-level accuracy. The instantaneous sampling moment is the physical moment of the voltage value latched by the analog-to-digital converter. The standard timestamp is an absolute time tag strictly aligned with that moment, obtained from the time synchronization module. This step hardware-binds the timestamp to the sampled value at the instant of sampling, avoiding synchronization errors caused by software delays and ensuring that each sampling point has a time coordinate accurate to the nanosecond.

[0085] The analog-to-digital converter (ADC) performs digital encoding on instantaneous sampled data with standard timestamps, converting it into raw digital signals with standard timestamps.

[0086] An analog-to-digital converter (ADC) converts instantaneous sampled analog voltage values ​​into binary values. After digitization, each digital sample point retains its original timestamp, forming a (time, value) sequence. This step transforms a continuous analog signal into a discrete digital stream that a computer can process, while fully preserving nanosecond-level time information.

[0087] The raw digital signal with standard timestamps is fed into a unified timing reference to achieve nanosecond-level time alignment of multi-channel data, resulting in an aligned digital signal.

[0088] A unified timing reference means that all acquisition units use the satellite clock as a reference, ensuring a consistent time standard. Nanosecond-level time alignment involves arranging sampling points with the same or closest timestamps from each channel onto the same time axis. This step, through merge sorting and interpolation matching, ensures that the residual time deviation of the multi-channel signals is less than 1 nanosecond, laying the foundation for subsequent accurate calculation of microsecond delays.

[0089] The aligned digital signal is input into a digital bandpass filter to filter out power frequency interference and low-frequency noise outside the passband, generating a filtered multi-channel digital signal.

[0090] Digital bandpass filters allow signals within a specific frequency range to pass through. For partial discharge detection, a passband of 100MHz to 500MHz effectively preserves high-frequency pulses from faults and interference while suppressing 50Hz power frequency and its harmonics, as well as low-frequency vibration noise. The signal-to-noise ratio is significantly improved after filtering, making subsequent event detection and time delay calculations more reliable.

[0091] Using the maximum absolute value of the filtered multi-channel digital signal as a reference, the amplitude values ​​of all remaining channel signals are scaled to the range of negative one to one to generate a normalized digital signal, which serves as the time-synchronized signal.

[0092] The maximum absolute value refers to the maximum absolute amplitude value among all sampling points across all channels within the current time window. Normalization divides the amplitude of each sampling point by this maximum value, linearly mapping all values ​​to the interval [-1, 1]. This step eliminates the absolute energy differences between different events, allowing subsequent delay calculations and amplitude comparisons to focus only on the relative relationships between channels, resulting in a uniformly scaled and stable signal.

[0093] S2. Based on the time-synchronized signal, calculate the root mean square value of the signal within the sliding window for each signal acquisition unit, mark the center time of the window where the root mean square value exceeds the preset abnormal threshold as the abnormal event trigger time, identify the signal segment corresponding to the abnormal event trigger time as an abnormal signal, and calculate the arrival time difference and signal strength difference of the same abnormal signal among multiple signal acquisition units.

[0094] In this embodiment, based on the time-synchronized signal, the root mean square (RMS) value of the signal within a sliding window is calculated for each signal acquisition unit. The center time of the window where the RMS value exceeds a preset anomaly threshold is marked as the anomaly event trigger time. The signal segment corresponding to the anomaly event trigger time is identified as an anomalous signal. The arrival time difference and signal strength difference of the same anomalous signal among multiple signal acquisition units are calculated, including:

[0095] For each signal acquisition unit after time synchronization, the root mean square value of the signal within the sliding window is calculated. The center time of the window where the root mean square value exceeds the preset abnormal threshold is marked as the abnormal event trigger time, thus obtaining the abnormal event trigger time.

[0096] A sliding window refers to a fixed-length time interval that moves along the time axis. The root mean square (RMS) value is the square root of the sum of the squares of all samples within the window, reflecting the average energy of the signal during that time period. The preset anomaly threshold is a threshold value set based on the statistical characteristics of normal operating signals; exceeding this threshold indicates a possible abnormal event. This step involves independently sliding the window on each sensor channel and calculating the RMS. When the RMS exceeds the threshold, the center time of the window is recorded as the trigger time of the abnormal event. This method can effectively capture sudden pulse signals, and the center time is more representative of the true occurrence time of the event than the edge times.

[0097] The abnormal signal segment is obtained by extracting a signal segment centered on the time of the abnormal event from the time-synchronized signal.

[0098] A signal segment refers to a fixed-length discrete sampling sequence extracted from the original continuous signal. This step takes a certain number of sampling points forward and backward (e.g., a total length of 100 microseconds) centered on the trigger moment to construct a complete event waveform. This segment contains sufficient information about the events before and after the occurrence of the abnormal event, facilitating subsequent comparison of the waveform characteristics of the same event on different channels.

[0099] The abnormal signal segment of the first signal acquisition unit that triggered the abnormal event is used as the reference signal segment. The cross-correlation function of the abnormal signal segments of the other signal acquisition units and the reference signal segment is calculated. The sampling point offset when the cross-correlation function reaches its maximum value is obtained, and the arrival time difference of the other signal acquisition units relative to the reference signal acquisition unit is obtained.

[0100] The cross-correlation function is used to measure the similarity between two signals at different time offsets. For a reference signal segment... and another signal segment Their cross-correlation function is defined as:

[0101]

[0102] In the formula, It is the time offset. Representing time, it is a continuous variable. This step calculates this function for each anomalous signal segment in the non-reference channel and the reference segment, finding the one that makes... Take the maximum value Value. This refers to the time difference (delay) between the arrival of a signal from a reference sensor to another sensor. The sensor that first triggers the event is closest to the source of interference or the fault point, and using it as a reference yields the most accurate relative time delay.

[0103] Based on the time-of-arrival difference, the sampling points of the reference signal segment and the other abnormal signal segments are aligned. The amplitude values ​​of each signal acquisition unit are extracted from the aligned sampling points to obtain the aligned amplitude value sequence.

[0104] Sampling point alignment refers to shifting signal segments from other channels along the time axis based on the calculated time delay (in units of sampling points), so that the physical events corresponding to each sampling moment are matched in time. In this step, the reference segment is kept stationary, and the remaining segments are shifted forward or backward by a certain number of sampling points (the shift amount equals the time delay divided by the sampling period), and then the amplitude values ​​under the same time index are extracted.

[0105] Extract the maximum absolute value of the abnormal signal segment from each signal acquisition unit in the aligned amplitude value sequence. Using the maximum absolute value of the reference signal acquisition unit as the denominator, calculate the ratio of the maximum absolute value of the remaining signal acquisition units to the maximum absolute value of the reference signal acquisition unit to obtain the signal strength difference.

[0106] The maximum absolute value refers to the maximum absolute value of the amplitudes of all sampling points within a signal segment, representing the peak intensity of the event on the corresponding sensor. Signal intensity difference refers to the ratio of the peak values ​​of other channels to the peak value of the reference channel. This step calculates the peak-to-peak ratio for each non-reference channel. Since external electromagnetic interference attenuates with increasing distance when propagating through the air, and internal faults attenuate differently when propagating along the cable, this ratio, along with time delay, constitutes a key characteristic for distinguishing the source of the signal.

[0107] S3. Compare the time of arrival difference and signal strength difference with the pre-built external interference propagation feature library and internal interference propagation feature library, and determine whether the source of the signal is external interference or internal fault based on the comparison results.

[0108] In this embodiment, comparing the time-of-arrival difference and signal strength difference with pre-built external interference propagation feature libraries and internal interference propagation feature libraries, and determining whether the signal source is external interference or an internal fault based on the comparison results, includes:

[0109] Obtain a pre-constructed external interference propagation feature library and an internal interference propagation feature library. The external interference propagation feature library stores multiple first feature vectors, each of which corresponds to a first set of arrival time differences and a first set of signal strength differences of an external interference signal among multiple signal acquisition units. The internal interference propagation feature library stores multiple second feature vectors, each of which corresponds to a second set of arrival time differences and a second set of signal strength differences of an internal fault signal among multiple signal acquisition units.

[0110] The feature vector is a multidimensional array composed of arrival time differences and signal strength differences. The external interference propagation feature library is a set of typical feature vectors obtained through actual measurement and calculation under known interference source operation, while the internal interference propagation feature library is a set of typical feature vectors obtained through laboratory simulation of internal cable faults. This step reads these two pre-built feature libraries from the storage unit to provide reference templates for subsequent comparisons.

[0111] The time of arrival difference and signal strength difference are combined into a feature vector to be identified. The first Euclidean distance set between the feature vector to be identified and each first feature vector in the external interference propagation feature library, and the second Euclidean distance set between the feature vector to be identified and each second feature vector in the internal interference propagation feature library are calculated using the Euclidean distance formula.

[0112] Euclidean distance is used to measure the similarity between two feature vectors; the smaller the distance, the more similar they are. Let the feature vector to be identified be... The template feature vector is The Euclidean distance calculation formula is:

[0113]

[0114] In the formula, This represents the feature vector to be identified. This represents the template feature vector in the feature library. The dimension of the feature vector (i.e., the sum of the number of time-of-arrival differences and the number of signal strength differences). and Let represent the values ​​of the vector to be identified and the template vector in the i-th dimension, respectively. The sum of squared differences across dimensions is used to calculate the Euclidean distance, which is obtained by taking the square root of the sum of squared differences. In this step, the Euclidean distance is calculated between the vector to be identified and each first feature vector in the external feature library, resulting in the first set of Euclidean distances. Similarly, the Euclidean distance is calculated for each second feature vector in the internal feature library, resulting in the second set of Euclidean distances. These two sets reflect the global similarity between the signal to be identified and known interference or fault templates.

[0115] The algorithm extracts the minimum value from the first Euclidean distance set as the first minimum distance and the second minimum value as the first minimum distance; it also extracts the minimum value from the second Euclidean distance set as the second minimum distance and the second minimum value as the second minimum distance.

[0116] The sorting algorithm refers to the operation of arranging a set of values ​​in ascending order. The first minimum distance is the distance between the signal to be identified and the most similar template in the external feature library; the first minimum distance is the distance with the second most similar template; the second minimum distance and the second minimum distance correspond to the closest and second closest templates in the internal feature library. This step extracts these four key values ​​for subsequent comparison and judgment.

[0117] If the first minimum distance is less than the second minimum distance, the source of the signal is determined to be external interference; if the first minimum distance is greater than the second minimum distance, the source of the signal is determined to be an internal fault.

[0118] The judgment logic is based on the nearest neighbor principle: the signal to be identified is classified into the category of the feature library with the smallest distance to the nearest template. That is, if the minimum distance to the external library is less than the minimum distance to the internal library, it means that the signal is closer to the external interference feature and is judged as external interference; otherwise, it is judged as an internal fault.

[0119] If the first minimum distance is equal to the second minimum distance, then calculate the first ratio of the first minimum distance to the first minimum distance, and calculate the second ratio of the second minimum distance to the second minimum distance; if the first ratio is less than the second ratio, then the source of the signal is determined to be external interference; if the second ratio is less than or equal to the first ratio, then the source of the signal is determined to be an internal fault.

[0120] When the nearest distances are equal, simply comparing the minimum value is insufficient to distinguish them. Therefore, the distance information of the second nearest neighbor is introduced. The first ratio is... The second ratio is The smaller the ratio, the more significant the leading advantage of the most recent template over the second most recent template, and the higher the classification confidence. If the first ratio is less than the second ratio, it indicates that the most recent template from the external library has a more significant advantage, and is judged as external interference; otherwise, it is judged as an internal fault.

[0121] In this embodiment, the pre-construction process of the external interference propagation feature library and the internal interference propagation feature library includes:

[0122] During the load fluctuation of the interference source cable laid in the same cable trench as the target cable and parallel to the target cable, the time-synchronized signals of multiple signal acquisition units are collected simultaneously to obtain multiple interference source operation signals.

[0123] Interference source cables refer to lines laid parallel to the target cable at close range, such as 35 kV data center cables or fiber optic composite cables. Load fluctuation periods refer to the operating times of interference source cables during transient load changes, equipment start-ups / shutdowns, or data migrations that generate strong electromagnetic interference. Time-synchronized signals refer to multi-channel digital signals that have undergone synchronous processing. This step involves synchronous acquisition during the load fluctuation period of the interference source cable to ensure that the captured signals primarily reflect external interference characteristics, rather than internal cable faults. The acquired operating signals from multiple interference sources constitute the raw data for constructing the external feature library.

[0124] For each interference source's operating signal, a sliding window energy detection is performed. The center time of the window where the root mean square value of the signal exceeds the interference trigger threshold is marked as the interference trigger time. A fixed-length signal segment is extracted with the interference trigger time as the center to obtain multiple interference signal segments.

[0125] Sliding window energy detection involves sliding a fixed-length time window along the signal time axis and calculating the arithmetic square root (root mean square) of the sum of the squares of all samples within each window. The interference trigger threshold is a threshold value set based on the background noise level of the interference source's operating signal, typically 3 to 5 times the noise root mean square value. In this step, a sliding window is applied to each interference source's operating signal. When the root mean square value exceeds the threshold, the center time of that window is marked as the interference trigger time, and a fixed-length (e.g., 100 microseconds) signal segment is extracted centered on this point. These segments are called interference signal segments, and each segment represents an independent interference event waveform.

[0126] For each of the multiple interference signal segments, the time difference of arrival between the multiple signal acquisition units is calculated using the cross-correlation function. The maximum absolute value of the interference signal segment on each signal acquisition unit is extracted and the ratio is calculated to obtain the signal strength difference. The time difference of arrival and the signal strength difference are combined into an interference feature vector to obtain multiple interference feature vectors.

[0127] The cross-correlation function is used to calculate the arrival time difference of the same interference signal between different sensors. The maximum absolute value refers to the maximum amplitude within the signal segment on each sensor, and the signal strength difference refers to the ratio of the peak value of each sensor to the peak value of the reference sensor. The interference feature vector is a multi-dimensional array composed of a time delay set and an amplitude ratio set. For example, when the system deploys three signal acquisition units numbered S1, S2, and S3, taking S1 as the reference unit, the interference feature vector can be expressed as... ,in, This represents the time difference of arrival of the same interference signal from signal acquisition unit S1 to S2. This represents the time difference of arrival of the same interference signal from signal acquisition unit S1 to S3. These represent the peak amplitudes of the same interference signal on signal acquisition units S1, S2, and S3, respectively. The ratio of the peak amplitudes of S2 to S1. This represents the peak amplitude ratio of S3 to S1. This eigenvector comprehensively reflects the spatiotemporal propagation characteristics of external interference signals among multiple sensors. This step independently calculates the eigenvector for each interference signal segment, obtaining a set of vectors that characterize the propagation characteristics of this type of external interference.

[0128] Multiple interference feature vectors are stored in an external interference propagation feature library;

[0129] This step stores all the interference feature vectors calculated in the previous step into a database or storage unit to form an external interference propagation feature library.

[0130] A cable sample of the same model and specifications as the target cable was cut, and a partial discharge fault was simulated on the cable sample.

[0131] The cable sample refers to a test section that is identical to the target cable in terms of structure, materials, and insulation class. Simulating partial discharge faults involves artificially creating insulation defects in the cable sample, such as inserting metal needles into the insulation layer, creating air gaps, or applying overvoltage to generate realistic partial discharge pulses. This step simulates internal faults in a controlled laboratory environment, avoiding interference from the complex electromagnetic environment of the field, thus obtaining pure fault characteristics.

[0132] Signal acquisition units of the same number and spacing as the target cable are arranged along the axial direction of the cable sample to synchronously acquire characteristic signals when a fault occurs, thereby obtaining fault simulation signals.

[0133] This step involves deploying the same number of partial discharge sensors on the cable sample at the exact same geometric positions as in the field, ensuring that the signal propagation path lengths are identical. Nanosecond-level synchronization is achieved using a BeiDou or GPS timing module. The acquired fault simulation signals record the waveforms of the partial discharge pulses propagating along the cable's interior to each sensor, providing data for building an internal feature library.

[0134] A sliding window energy detection is performed on the fault simulation signal. The center time of the window where the root mean square value of the signal exceeds the fault triggering threshold is marked as the fault triggering time. A fixed length signal segment is extracted with the fault triggering time as the center to obtain multiple fault signal segments.

[0135] The fault trigger threshold is set based on the background noise of the fault simulation signal, and is usually slightly lower than the interference trigger threshold because the internal fault signal attenuates significantly when propagating inside the cable, and its amplitude may be small. This step is the same as the segment extraction method for external interference, resulting in multiple fault signal segments, each segment corresponding to the waveform of a simulated partial discharge event.

[0136] For each fault signal segment among multiple fault signal segments, the time difference of arrival between multiple signal acquisition units is calculated by cross-correlation function. The maximum absolute value of the fault signal segment on each signal acquisition unit is extracted and the ratio is calculated to obtain the signal strength difference. The time difference of arrival and the signal strength difference are combined into a fault feature vector to obtain multiple fault feature vectors.

[0137] This step is entirely consistent with the method for calculating the eigenvectors of external interference, both using time delay and amplitude ratio to construct the eigenvectors. Because internal fault signals propagate along the cable insulation or conductor at a speed (typically 50% to 70% of the speed of light) lower than the speed of external electromagnetic waves in air, and their attenuation modes differ, the calculated time delay and amplitude ratio differ significantly from those of external interference. These differences allow the eigenvectors to distinguish between the two signal sources.

[0138] Multiple fault feature vectors are stored in an internal interference propagation feature library.

[0139] This step stores all the fault feature vectors obtained in the previous step into the database, forming an internal interference propagation feature library.

[0140] S4. When the source of the signal is determined to be external interference, adaptive filtering is used to remove the interference components from the signal to obtain a purified signal.

[0141] In this embodiment, when the source of the signal is determined to be external interference, adaptive filtering is used to remove the interference components from the signal to obtain a purified signal, including:

[0142] Based on the difference in arrival time, the signal acquisition unit with the earliest trigger time of the abnormal event is selected from multiple signal acquisition units as the reference sensor, and the time-synchronized signal of the reference sensor is used as the reference input signal.

[0143] The time-of-arrival difference is the calculated time delay of each sensor relative to the reference sensor. The sensor that triggers the abnormal event earliest is closest to the interference source, resulting in the strongest interference component and the highest signal-to-noise ratio in its signal. The reference sensor is the selected sensor whose time-synchronized signal is used as the reference input signal for the adaptive filter. The purpose of this step is to provide a high-quality interference reference source for subsequent interference cancellation.

[0144] Use the time-synchronized signal from any of the other signal acquisition units besides the reference sensor as the main input signal.

[0145] The main input signal refers to a mixed signal containing interference components and potential internal fault components, denoted as... Since the reference sensor is closest to the interference source, the interference signals received by the other sensors experience time delay and amplitude attenuation relative to the reference sensor, but are still highly correlated with the reference signal. This step selects the synchronization signal of any non-reference sensor as the main input. Subsequently, adaptive filtering will be used to subtract the estimated interference component from the main input to obtain the purified signal.

[0146] The filter weight vector is initialized to zero. The reference input signal is input to the finite impulse response filter. The finite impulse response filter performs a weighted summation of the reference input signal based on the current filter weight vector and outputs an estimated interference signal.

[0147] Filter weight vector It is a length of The coefficient array is initially all zeros. The Finite Impulse Response (FIR) filter responds to the current and past states of the reference input signal. The sampling points are weighted and summed to output the estimated interference signal. The calculation formula is:

[0148] ;

[0149] In the formula, This indicates the reference input signal at the current sampling time. amplitude, Indicates the past number The amplitude of each sampling point For the first Each filter coefficient As the reference input signal vector, This is the transpose of the weight vector. This step generates an estimate of the interference components in the main input through a linear combination of filters.

[0150] The error signal is obtained by subtracting the estimated interference signal from the main input signal.

[0151] Error signal The calculation formula is:

[0152] ;

[0153] In the formula, It is the amplitude of the main input signal at the current sampling moment. This step estimates the interference signal. If the estimated interference signal is completely consistent with the actual interference components in the main input, then the error signal will only contain the internal fault signal and residual noise. The error signal generated in this step is both the output (cleaned signal) of the adaptive filter and the basis for subsequent adjustment of the filter weights.

[0154] The error signal and the reference input signal are used to calculate the weight adjustment amount using the least mean square algorithm, and the filter weight vector is updated. This process is repeated until the mean square value of the error signal converges. The error signal at the end of the iteration is used as the clean signal.

[0155] The Least Mean Square (LMS) algorithm is an adaptive filtering algorithm that progressively optimizes the filter weights by minimizing the mean square value of the error signal. The weight update formula is:

[0156] ;

[0157] In the formula, Indicates the current sampling time The filter weight vector, This represents the weight vector at the next sampling time. It is the step size factor (a small positive constant that controls the convergence speed and steady-state error). This is the current error signal. This is the reference input signal vector. In each iteration, the weight adjustment is... This step repeats the above update process until the mean square value of the error signal tends to stabilize (i.e., convergence). At this point, the filter can estimate the interference components well, and the final error signal... Interference is significantly suppressed, and the signal is output as a purification signal.

[0158] In this embodiment, the step of calculating the weight adjustment amount and updating the filter weight vector by using the least mean square algorithm on the error signal and the reference input signal includes:

[0159] Extract the filter weight vector at the current sampling time and the error signal at the current sampling time.

[0160] The current sampling time is recorded as Filter weight vector It is the set of coefficients of the adaptive filter at the current moment. Error signal. This is the calculated difference between the current main input signal and the estimated interference signal. This step reads these two values ​​from the filter storage unit as the basis for weight updates.

[0161] Extract the signal amplitudes at the current sampling time and previous sampling times from the reference input signal to construct the reference input signal vector.

[0162] Reference input signal This is the time synchronization signal from the reference sensor. Reference input signal vector. From the current sampling point and its predecessor It consists of historical sampling points, with a length equal to the filter order. This step constructs the vector, which will be used in subsequent operations with the error signal and weight vector.

[0163] Multiply the preset step size factor by the error signal, and then multiply by the reference input signal vector to obtain the weight adjustment vector.

[0164] Step size factor This is a pre-set positive constant used to control the convergence speed and steady-state error of the adaptive filter. The step size factor is set according to the filter order. and the average power of the reference input signal Select satisfy This ensures the algorithm converges and remains stable. It is typically set to a value of... to The decimals between [a certain value] and [a certain value]. The formula for calculating the weight adjustment vector is:

[0165] ;

[0166] In the formula, This represents the weight adjustment vector. It is the step size factor. It is the error signal at the current sampling time. This is the reference input signal vector. The adjustment calculated in this step determines the direction and step size of the filter weights in each iteration.

[0167] Add the weight adjustment vector to the filter weight vector at the current sampling time to obtain the filter weight vector at the next sampling time.

[0168] The formula for calculating the filter weight vector at the next sampling time is:

[0169] ;

[0170] In the formula, This represents the weight vector at the next sampling time. It is the weight vector at the current sampling time. This is the weight adjustment vector calculated in the previous step. This step completes one iteration update, adjusting the filter coefficients in a direction that reduces the mean square value of the error signal. This process is repeated until the mean square value of the error signal converges.

[0171] S5. Input the purification signal, combined with the target cable's body temperature and load current parameters, into the pre-built cable condition assessment model, and output the cable condition assessment results.

[0172] In this embodiment, the cable condition assessment model includes a partial discharge phase distribution map generation module, a convolutional neural network classification module, and a multi-parameter fusion judgment module. The partial discharge phase distribution map generation module converts the purified partial discharge pulse sequence into a discharge number distribution vector and a maximum discharge quantity distribution vector, constructing a two-dimensional map. The convolutional neural network classification module extracts features from the two-dimensional map using pre-trained convolutional layers, pooling layers, and fully connected layers, outputting a discharge type probability distribution. The multi-parameter fusion judgment module concatenates the discharge type probability distribution with the body temperature state code and the load current state code, then outputs a cable health score using a Sigmoid activation function. The purification signal, combined with the target cable's body temperature and load current parameters, is input into the pre-constructed cable condition assessment model to output the cable condition assessment result, including:

[0173] First, the pre-construction method of the cable condition assessment model is explained. This model is pre-trained through the following process: A large amount of data on purified partial discharge signals, cable body temperature, and load current under known cable conditions (such as normal operation, partial discharge, joint overheating, etc.) is collected, and a health score is assigned based on the actual condition (0 indicates healthy, 1 indicates severe abnormality). The pulse sequence is converted into a two-dimensional spectrum using a partial discharge phase distribution map generation module. This spectrum is then input into a convolutional neural network for training, learning the mapping from the spectrum to the discharge type probability distribution. Simultaneously, the temperature and current state codes are concatenated with the discharge type probability distribution and trained through a fully connected layer with a Sigmoid activation function, enabling the model to output a health score between 0 and 1. After training, the model parameters are fixed for actual assessment.

[0174] Peak detection is performed on the purification signal to extract the peak amplitude and occurrence time of each pulse in the purification signal, and a pulse sequence containing the peak amplitude and occurrence time is generated.

[0175] Peak detection refers to identifying pulses exceeding a set threshold from a continuous purified signal waveform and recording the maximum amplitude of each pulse and its corresponding time position. The purified signal is the output signal that has been freed from external interference. The pulse sequence is a list, with each element being a binary tuple (peak amplitude, occurrence time). This step discretizes the continuous waveform into a series of independent events, providing the basic data for subsequent generation of the phase distribution map.

[0176] The power grid synchronization signal is obtained as the power frequency cycle reference. Based on the phase difference between the occurrence time of each pulse in the pulse sequence and the zero-crossing point of the power grid synchronization signal, each pulse is assigned to the corresponding power frequency phase interval. The partial discharge phase distribution map is generated by the partial discharge phase distribution map generation module.

[0177] The power grid synchronization signal is the power frequency voltage signal (50Hz or 60Hz) obtained from the voltage transformer of the target cable bus. A zero-crossing point refers to the instantaneous point at which the signal changes from negative to positive or vice versa. Phase difference refers to the time offset of the pulse occurrence relative to the zero-crossing point, converted to angles (0 to 360 degrees). The power frequency phase interval divides a complete power frequency cycle into several equal parts (e.g., 1 degree per interval, totaling 360 intervals).

[0178] This step calculates the time difference between the occurrence time and the nearest zero-crossing point for each pulse in the pulse sequence, divides it by the power frequency cycle duration, and multiplies it by the total number of intervals to obtain the phase interval number to which the pulse belongs. Then, the number of pulses in each interval is accumulated to form a discharge frequency distribution vector, and the maximum peak amplitude in each interval is recorded as the maximum discharge quantity distribution vector. Finally, these two vectors are concatenated as two channels to form a two-dimensional matrix, i.e., the partial discharge phase distribution map. This map intuitively reflects the correlation between partial discharge and power frequency voltage phase, and is a key feature for identifying discharge types.

[0179] The partial discharge phase distribution map is input into the pre-trained convolutional layer of the convolutional neural network classification module. Local features are extracted through the convolutional layer, dimensionality is reduced through the pooling layer, and the discharge type probability distribution is output through the fully connected layer to obtain the discharge type probability distribution vector.

[0180] The convolutional neural network classification module contains multiple convolutional layers, pooling layers, and fully connected layers. Convolutional layers use small convolutional kernels to slide across a two-dimensional map, extracting local spatial patterns (e.g., discharge clustering features within specific phase intervals). Pooling layers downsample the feature map output by the convolutions (e.g., max pooling), reducing the number of parameters and enhancing translation invariance.

[0181] The fully connected layer maps the pooled feature vectors to the output space, with the output dimension equal to the preset number of discharge types (e.g., no discharge, internal discharge, surface discharge, corona discharge, etc.), and converts them into a probability distribution using the Softmax function. This step inputs the two-dimensional map generated in the previous step into the pre-trained CNN. After forward propagation, a probability distribution vector is obtained, where each element represents the probability of the map corresponding to a certain discharge type. This vector quantifies the discharge mode information implicit in the map.

[0182] Obtain the target cable's body temperature and load current value. Compare the body temperature value with the average value of historical temperature data. If the body temperature value is less than the average value of historical temperature data, output temperature status code 0; if it is greater than or equal to the average value, output temperature status code 1, generating a temperature status code. Compare the load current value with the average value of historical current data. If the load current value is less than the average value of historical current data, output current status code 0; if it is greater than or equal to the average value, output current status code 1, generating a current status code.

[0183] The body temperature value refers to the temperature of the target cable's insulation or sheath measured by distributed optical fiber or point temperature sensors. The average historical temperature data is the long-term average temperature statistically analyzed during normal operation. The load current value is the effective value of the current currently transmitted by the target cable obtained through a current transformer. The average historical current data is the average current under normal load conditions. The status code is the result of binarizing continuous parameters: values ​​below the historical average are considered normal (code 0), and values ​​above or equal to the historical average are considered abnormally high (code 1). This step generates two single-bit codes, reflecting the cable's current thermal state and load state, respectively.

[0184] The discharge type probability distribution vector, temperature status code, and current status code are input into the multi-parameter fusion judgment module for vector concatenation to form a comprehensive feature vector. The comprehensive feature vector is then used to output the cable health score through the Sigmoid activation function, which constitutes the cable status assessment result.

[0185] Vector concatenation refers to joining multiple vectors or scalars into a longer one-dimensional vector. Let the probability distribution vector of discharge type be... ( (Number of discharge types), temperature state code The current state is coded as Then the comprehensive feature vector is The Sigmoid activation function maps any real number to... The interval, the formula is:

[0186] ;

[0187] In the formula, This indicates the output cable health score. This represents the scalar value of the composite eigenvector after a linear transformation (weighted summation plus bias). It is the base of the natural logarithm. This step inputs the comprehensive feature vector into the multi-parameter fusion judgment module, which first performs a linear transformation on the vector:

[0188] ;

[0189] In the formula, For the weight vector, For bias, then The cable health score is obtained by mapping the value to a range of 0 to 1 using the Sigmoid function. A score closer to 1 indicates a more severe abnormality, while a score closer to 0 indicates a healthier cable. This score is the final cable condition assessment result.

[0190] In this embodiment, generating the partial discharge phase distribution map includes:

[0191] The power frequency voltage signal is obtained from the bus of the target cable through a voltage transformer, and the power frequency voltage signal is zero-crossing detected to generate a square wave synchronization signal.

[0192] A voltage transformer is a measuring device that converts high voltage to low voltage to safely acquire the power frequency voltage waveform on cable busbars. Zero-crossing detection refers to identifying the zero-crossing points where the power frequency voltage signal changes from negative to positive or from positive to negative, and outputting a pulse at each zero-crossing point. The square wave synchronization signal is a digital signal that jumps at zero-crossing points; its rising edge corresponds to the zero-crossing point where the voltage changes from negative to positive, and its falling edge corresponds to the zero-crossing point where the voltage changes from positive to negative. This step converts the continuous power frequency voltage into a discrete square wave synchronization signal, providing a unified time base reference for subsequent pulse phase positioning.

[0193] The duration of a complete cycle of the square wave synchronization signal is obtained as the power frequency cycle duration, and the power frequency cycle duration is divided into multiple phase intervals.

[0194] A complete cycle duration refers to the time interval between two consecutive rising edges of a square wave synchronization signal; this is 20 milliseconds for a 50Hz power frequency and 16.67 milliseconds for a 60Hz power frequency. A phase interval is a small segment into which the 360-degree power frequency phase angle is evenly divided. The total number of phase intervals is a preset value, usually set according to the analysis precision; for example, 360 intervals per 1 degree, or 180 (per 2 degrees) or 720 (per 0.5 degrees). This step determines the power frequency cycle length and interval division precision, preparing for assigning a specific phase interval number to each pulse.

[0195] Calculate the time difference between the occurrence time of each pulse in the pulse sequence and the rising edge of the square wave synchronization signal. Divide the time difference by the power frequency cycle duration and multiply by the total number of phase intervals to obtain the phase interval number of the pulse.

[0196] The pulse sequence is a list (peak amplitude, occurrence time) extracted from the purified signal. The time difference is denoted as... This refers to the time interval between the pulse occurrence and the most recent rising edge of the square wave. The duration of the power frequency cycle is denoted as... The total number of phase intervals is denoted as (Preset value, such as 360). Phase interval number. The calculation formula is:

[0197] ;

[0198] In the formula, This represents the time difference between the pulse occurrence and the rising edge of the square wave synchronization signal. Indicates the duration of the power frequency cycle. Indicates the total number of phase intervals. This indicates the floor function. Initialize a count array and a maximum value array of the same length as the total number of phase intervals, with each element initialized to zero.

[0199] The counting array is a array of length 1000. An integer array is used to count the number of pulse occurrences within each phase interval. The maximum value array is a long integer array. A floating-point array is used to record the maximum value of the pulse peak amplitude within each phase interval. This step creates two arrays filled with zeros to prepare for subsequent cumulative statistics.

[0200] Traverse the pulse sequence, locate the index position of the counting array based on the phase interval number of each pulse, and increment the count value at the index position by one.

[0201] Traversal refers to processing each pulse in the pulse sequence sequentially. For each pulse, its phase interval is numbered. As the index of the counting array, execute This step accumulates the number of pulse occurrences according to the phase interval to obtain the discharge frequency distribution.

[0202] The peak amplitude of the pulse is compared with the current value at the same index position in the maximum value array, and the larger value is used to update the index position in the maximum value array.

[0203] For each pulse, its phase interval number is determined. Read the current value of the maximum value array The peak amplitude of the current pulse and In comparison, if Replace if larger This step records the maximum pulse amplitude within each phase interval to obtain the maximum discharge quantity distribution.

[0204] After the traversal is complete, the count array is used as the distribution vector of the number of discharges, and the maximum value array is used as the distribution vector of the maximum discharge amount.

[0205] The discharge number distribution vector is a vector of length . The first array is a vector, where each element represents the number of pulse occurrences within the corresponding phase interval. The second array is a vector of the same length, where each element represents the maximum amplitude of a pulse within the corresponding phase interval. This step converts both arrays into vector form to facilitate subsequent graph construction.

[0206] The discharge number distribution vector and the maximum discharge quantity distribution vector are used as two channels to construct a two-dimensional matrix, which constitutes a partial discharge phase distribution map.

[0207] The rows of the two-dimensional matrix correspond to the phase interval index, and the columns correspond to two channels (the first channel represents the number of discharges, and the second channel represents the maximum discharge amount). This matrix can be visualized as a two-row or two-column image, where the color intensity represents the numerical value.

[0208] Please see Figure 2 This invention provides a cable condition anomaly detection system based on dynamic modeling, comprising:

[0209] Signal acquisition module: used to acquire characteristic signals collected by multiple signal acquisition units on the target cable, and to perform time synchronization processing on the acquired signals to obtain time-synchronized signals;

[0210] Difference Calculation Module: Based on the time-synchronized signal, it calculates the root mean square value of the signal within the sliding window for each signal acquisition unit, marks the center time of the window where the root mean square value exceeds the preset abnormal threshold as the abnormal event trigger time, identifies the signal segment corresponding to the abnormal event trigger time as an abnormal signal, and calculates the arrival time difference and signal strength difference of the same abnormal signal among multiple signal acquisition units.

[0211] Signal source determination module: It is used to compare the time of arrival difference and signal strength difference with the pre-built external interference propagation feature library and internal interference propagation feature library, and determine whether the source of the signal is external interference or internal fault based on the comparison results.

[0212] Interference Removal Module: When the source of a signal is determined to be external interference, it uses adaptive filtering to remove the interference components from the signal and obtain a purified signal.

[0213] Condition assessment module: This module combines the purification signal with the target cable's body temperature and load current parameters and inputs them into a pre-built cable condition assessment model to output the cable condition assessment results.

[0214] The above embodiments can be implemented, in whole or in part, by software, hardware, firmware, or any other combination thereof. When implemented in software, the above embodiments can be implemented, in whole or in part, as a computer program product. Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution.

[0215] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0216] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application.

Claims

1. A method for detecting cable condition anomalies based on dynamic modeling, characterized in that, Includes the following steps: S1. Acquire the characteristic signals collected by multiple signal acquisition units on the target cable, and perform time synchronization processing on the acquired signals to obtain the time-synchronized signals; S2. Based on the time-synchronized signal, calculate the root mean square value of the signal within the sliding window for each signal acquisition unit, mark the center time of the window where the root mean square value exceeds the preset abnormal threshold as the abnormal event trigger time, identify the signal segment corresponding to the abnormal event trigger time as an abnormal signal, and calculate the arrival time difference and signal strength difference of the same abnormal signal among multiple signal acquisition units. S3. Compare the time of arrival difference and signal strength difference with the pre-built external interference propagation feature library and internal interference propagation feature library, and determine whether the source of the signal is external interference or internal fault based on the comparison results. S4. When the source of the signal is determined to be external interference, adaptive filtering is used to remove the interference components from the signal to obtain a purified signal. S5. Input the purification signal, combined with the target cable's body temperature and load current parameters, into the pre-built cable condition assessment model, and output the cable condition assessment results.

2. The cable condition anomaly detection method based on dynamic modeling according to claim 1, characterized in that, The multiple signal acquisition units are arranged along the axial direction of the target cable. Each signal acquisition unit integrates a partial discharge sensor, a high-speed analog-to-digital converter, and a BeiDou / GPS high-precision timing module. The process of acquiring the characteristic signals collected by the multiple signal acquisition units on the target cable and performing time synchronization processing on the acquired signals to obtain a time-synchronized signal includes: The original analog electrical signals during the operation of the target cable are acquired using a partial discharge sensor; Based on the high-precision time synchronization module of the signal acquisition unit, at the instantaneous sampling moment when the analog-to-digital converter completes sampling and latching of the instantaneous sample value of the original analog electrical signal, the standard timestamp is bound to the instantaneous sample value to generate instantaneous sampled data with standard timestamp; The analog-to-digital converter performs digital encoding on instantaneous sampled data with standard timestamps, converting it into raw digital signals with standard timestamps; The raw digital signal with standard timestamps is fed into a unified timing reference to achieve nanosecond-level time alignment of multi-channel data, resulting in an aligned digital signal. The aligned digital signal is input into a digital bandpass filter to filter out power frequency interference and low-frequency noise outside the passband, generating a filtered multi-channel digital signal. Using the maximum absolute value of the filtered multi-channel digital signal as a reference, the amplitude values ​​of all remaining channel signals are scaled to the range of negative one to one to generate a normalized digital signal, which serves as the time-synchronized signal.

3. The cable condition anomaly detection method based on dynamic modeling according to claim 1, characterized in that, Based on the time-synchronized signal, the root mean square (RMS) value of the signal within a sliding window is calculated for each signal acquisition unit. The center time of the window where the RMS value exceeds a preset anomaly threshold is marked as the anomaly event trigger time. The signal segment corresponding to the anomaly event trigger time is identified as an anomalous signal. The arrival time difference and signal strength difference of the same anomalous signal among multiple signal acquisition units are calculated, including: For the time-synchronized signal of each signal acquisition unit, calculate the root mean square value of the signal within the sliding window, and mark the center time of the window where the root mean square value exceeds the preset abnormal threshold as the abnormal event trigger time to obtain the abnormal event trigger time. Extract the signal segment centered on the trigger time of the abnormal event from the time-synchronized signal to obtain the abnormal signal segment; The abnormal signal segment of the first signal acquisition unit that triggered the abnormal event is used as the reference signal segment. The cross-correlation function is calculated between the abnormal signal segments of the other signal acquisition units and the reference signal segment. The sampling point offset when the cross-correlation function reaches its maximum value is obtained, and the arrival time difference of the other signal acquisition units relative to the reference signal acquisition unit is obtained. Based on the time of arrival difference, the sampling points of the reference signal segment and the other abnormal signal segments are aligned. The amplitude values ​​of each signal acquisition unit are extracted from the aligned sampling points to obtain the aligned amplitude value sequence. Extract the maximum absolute value of the abnormal signal segment from each signal acquisition unit in the aligned amplitude value sequence. Using the maximum absolute value of the reference signal acquisition unit as the denominator, calculate the ratio of the maximum absolute value of the remaining signal acquisition units to the maximum absolute value of the reference signal acquisition unit to obtain the signal strength difference.

4. The cable condition anomaly detection method based on dynamic modeling according to claim 1, characterized in that, The step of comparing the time-of-arrival difference and signal strength difference with pre-built external interference propagation feature libraries and internal interference propagation feature libraries, and determining whether the signal source is external interference or an internal fault based on the comparison results, includes: Obtain a pre-built external interference propagation feature library and an internal interference propagation feature library. The external interference propagation feature library stores multiple first feature vectors, each of which corresponds to a first set of arrival time differences and a first set of signal strength differences of an external interference signal among multiple signal acquisition units. The internal interference propagation feature library stores multiple second feature vectors, each of which corresponds to a second set of arrival time differences and a second set of signal strength differences of an internal fault signal among multiple signal acquisition units. The time of arrival difference and signal strength difference are combined into a feature vector to be identified. The first Euclidean distance set between the feature vector to be identified and each first feature vector in the external interference propagation feature library, and the second Euclidean distance set between the feature vector to be identified and each second feature vector in the internal interference propagation feature library are calculated using the Euclidean distance formula. The algorithm extracts the minimum value from the first Euclidean distance set as the first minimum distance and the second minimum value as the first minimum distance; it also extracts the minimum value from the second Euclidean distance set as the second minimum distance and the second minimum value as the second minimum distance. If the first minimum distance is less than the second minimum distance, the source of the signal is determined to be external interference; if the first minimum distance is greater than the second minimum distance, the source of the signal is determined to be an internal fault. If the first minimum distance is equal to the second minimum distance, then calculate the first ratio of the first minimum distance to the first minimum distance, and calculate the second ratio of the second minimum distance to the second minimum distance; if the first ratio is less than the second ratio, then the source of the signal is determined to be external interference; if the second ratio is less than or equal to the first ratio, then the source of the signal is determined to be an internal fault.

5. The cable condition anomaly detection method based on dynamic modeling according to claim 4, characterized in that, The pre-construction process of the external interference propagation feature library and the internal interference propagation feature library includes: During the load fluctuation of the interference source cable laid in the same cable trench as the target cable and parallel to the target cable, the time-synchronized signals of multiple signal acquisition units are collected simultaneously to obtain multiple interference source operation signals. Sliding window energy detection is performed on the operating signal of each interference source. The center time of the window when the root mean square value of the signal exceeds the interference triggering threshold is marked as the interference triggering time. A fixed length signal segment is extracted with the interference triggering time as the center to obtain multiple interference signal segments. For each of the multiple interference signal segments, the time difference of arrival between the multiple signal acquisition units is calculated by the cross-correlation function. The maximum absolute value of the interference signal segment on each signal acquisition unit is extracted and the ratio is calculated to obtain the signal strength difference. The time difference of arrival and the signal strength difference are combined into an interference feature vector to obtain multiple interference feature vectors. Multiple interference feature vectors are stored in an external interference propagation feature library; Cut a cable sample of the same model and specifications as the target cable and simulate a partial discharge fault on the cable sample; The same number and spacing of signal acquisition units as the target cable are arranged along the axial direction of the cable sample to synchronously acquire the characteristic signals when the fault occurs, and obtain the fault simulation signal. Sliding window energy detection is performed on the fault simulation signal. The center time of the window where the root mean square value of the signal exceeds the fault triggering threshold is marked as the fault triggering time. A fixed length signal segment is extracted with the fault triggering time as the center to obtain multiple fault signal segments. For each fault signal segment among multiple fault signal segments, the time difference of arrival between multiple signal acquisition units is calculated by cross-correlation function. The maximum absolute value of the fault signal segment on each signal acquisition unit is extracted and the ratio is calculated to obtain the signal strength difference. The time difference of arrival and the signal strength difference are combined into a fault feature vector to obtain multiple fault feature vectors. Multiple fault feature vectors are stored in an internal interference propagation feature library.

6. The cable condition anomaly detection method based on dynamic modeling according to claim 1, characterized in that, When the source of the signal is determined to be external interference, adaptive filtering is used to remove the interference components from the signal to obtain a purified signal, including: Based on the difference in arrival time, the signal acquisition unit with the earliest abnormal event trigger time is selected from multiple signal acquisition units as the reference sensor, and the time-synchronized signal of the reference sensor is used as the reference input signal. Use the time-synchronized signal from any of the other signal acquisition units besides the reference sensor as the main input signal; The filter weight vector is initialized to zero. The reference input signal is input to the finite impulse response filter. The finite impulse response filter performs a weighted summation of the reference input signal based on the current filter weight vector and outputs an estimated interference signal. The error signal is obtained by subtracting the estimated interference signal from the main input signal; The error signal and the reference input signal are used to calculate the weight adjustment amount using the least mean square algorithm, and the filter weight vector is updated. This process is repeated until the mean square value of the error signal converges. The error signal at the end of the iteration is used as the clean signal.

7. The cable condition anomaly detection method based on dynamic modeling according to claim 6, characterized in that, The step of calculating the weight adjustment amount and updating the filter weight vector by using the least mean square algorithm on the error signal and the reference input signal includes: Extract the filter weight vector at the current sampling time and the error signal at the current sampling time; Extract the signal amplitudes at the current sampling time and previous sampling times from the reference input signal to construct the reference input signal vector; Multiply the preset step size factor by the error signal, and then multiply by the reference input signal vector to obtain the weight adjustment vector; Add the weight adjustment vector to the filter weight vector at the current sampling time to obtain the filter weight vector at the next sampling time.

8. The cable condition anomaly detection method based on dynamic modeling according to claim 1, characterized in that, The cable condition assessment model includes a partial discharge phase distribution map generation module, a convolutional neural network classification module, and a multi-parameter fusion judgment module. The partial discharge phase distribution map generation module converts the purified partial discharge pulse sequence into a discharge number distribution vector and a maximum discharge quantity distribution vector, constructing a two-dimensional map. The convolutional neural network classification module extracts features from the two-dimensional map using pre-trained convolutional layers, pooling layers, and fully connected layers, outputting a discharge type probability distribution. The multi-parameter fusion judgment module concatenates the discharge type probability distribution with the cable's body temperature state code and load current state code, then outputs a cable health score using a Sigmoid activation function. The purified signal, combined with the target cable's body temperature and load current parameters, is input into the pre-constructed cable condition assessment model to output the cable condition assessment results, including: Peak detection is performed on the purification signal to extract the peak amplitude and occurrence time of each pulse in the purification signal, and a pulse sequence containing the peak amplitude and occurrence time is generated. The power grid synchronization signal is obtained as the power frequency cycle reference. Based on the phase difference between the occurrence time of each pulse in the pulse sequence and the zero crossing point of the power grid synchronization signal, each pulse is assigned to the corresponding power frequency phase interval. The partial discharge phase distribution map is generated by the partial discharge phase distribution map generation module. The partial discharge phase distribution map is input into the pre-trained convolutional layer of the convolutional neural network classification module. Local features are extracted through the convolutional layer, dimensionality is reduced through the pooling layer, and the discharge type probability distribution is output through the fully connected layer to obtain the discharge type probability distribution vector. The system acquires the target cable's body temperature and load current values. It then compares the body temperature value with the average of historical temperature data. If the body temperature value is less than the average of historical temperature data, it outputs a temperature status code 0; otherwise, it outputs a temperature status code 1, generating a temperature status code. Similarly, it compares the load current value with the average of historical current data. If the load current value is less than the average of historical current data, it outputs a current status code 0; otherwise, it outputs a current status code 1, generating a current status code. The discharge type probability distribution vector, temperature status code, and current status code are input into the multi-parameter fusion judgment module for vector concatenation to form a comprehensive feature vector. The comprehensive feature vector is then used to output the cable health score through the Sigmoid activation function, which constitutes the cable status assessment result.

9. The cable condition anomaly detection method based on dynamic modeling according to claim 8, characterized in that, The generation of the partial discharge phase distribution map includes: The power frequency voltage signal is obtained from the bus of the target cable through a voltage transformer, and the power frequency voltage signal is zero-crossing detected to generate a square wave synchronization signal. The duration of a complete cycle of the square wave synchronization signal is obtained as the power frequency cycle duration, and the power frequency cycle duration is divided into multiple phase intervals. Calculate the time difference between the occurrence time of each pulse in the pulse sequence and the rising edge of the square wave synchronization signal. Divide the time difference by the power frequency cycle duration and multiply by the total number of phase intervals to obtain the phase interval number of the pulse. Initialize a counting array and a maximum value array of the same length as the total number of phase intervals, with each element initially set to zero; Traverse the pulse sequence, locate the index position of the counting array based on the phase interval number of each pulse, and increment the count value at the index position by one; Compare the peak amplitude of the pulse with the current value at the same index position in the maximum value array, and update the index position of the maximum value array with the larger value; After the traversal is complete, the count array is used as the distribution vector of the number of discharges, and the maximum value array is used as the distribution vector of the maximum discharge amount. The discharge number distribution vector and the maximum discharge quantity distribution vector are used as two channels to construct a two-dimensional matrix, which constitutes a partial discharge phase distribution map.

10. A cable condition anomaly detection system based on dynamic modeling, used with the method described in any one of claims 1 to 9, characterized in that, include: Signal acquisition module: used to acquire characteristic signals collected by multiple signal acquisition units on the target cable, and to perform time synchronization processing on the acquired signals to obtain time-synchronized signals; Difference Calculation Module: Based on the time-synchronized signal, it calculates the root mean square value of the signal within the sliding window for each signal acquisition unit, marks the center time of the window where the root mean square value exceeds the preset abnormal threshold as the abnormal event trigger time, identifies the signal segment corresponding to the abnormal event trigger time as an abnormal signal, and calculates the arrival time difference and signal strength difference of the same abnormal signal among multiple signal acquisition units. Signal source determination module: It is used to compare the time of arrival difference and signal strength difference with the pre-built external interference propagation feature library and internal interference propagation feature library, and determine whether the source of the signal is external interference or internal fault based on the comparison results. Interference Removal Module: When the source of a signal is determined to be external interference, it uses adaptive filtering to remove the interference components from the signal and obtain a purified signal. Condition assessment module: This module combines the purification signal with the target cable's body temperature and load current parameters and inputs them into a pre-built cable condition assessment model to output the cable condition assessment results.

Images