A high-precision partial discharge monitoring method for high-voltage cable signal denoising

Abstract

The present application relates to the technical field of high-voltage cable state monitoring, and particularly relates to a high-precision partial discharge monitoring method for high-voltage cable signal denoising. According to the time sequence difference characteristics between any two real-time sampling time points in the current monitoring period, and the fluctuation characteristics of the partial discharge voltage and the shield grounding current in the time interval between the any two real-time sampling time points, the partial discharge voltage and the shield grounding current at the buffer sampling time point are combined to analyze the double-channel topology coincidence degree. The adaptive weight vector of the previous real-time sampling time point adjacent to each real-time sampling time point in the current monitoring period is updated by using the double-channel topology coincidence degree, so as to obtain the adaptive weight vector of each real-time sampling time point; and the actual discharge voltage of each real-time sampling time point is determined according to the data characteristics of the shield grounding current of each real-time sampling time point and the historical sampling time point, in combination with the adaptive weight vector and the partial discharge voltage of each real-time sampling time point.

Description

Technical Field

[0001] This invention relates to the field of high-voltage cable condition monitoring technology, specifically to a high-precision partial discharge monitoring method for high-voltage cables with signal denoising. Background Technology

[0002] High-voltage cables are widely used in power transmission scenarios such as substations and underground cable tunnels. In actual field operation environments, the switching operations of adjacent equipment and line corona discharge frequently generate common-mode interference pulses. To monitor the insulation condition of the cable, partial discharge sensors are typically installed at the cable joints to extract high-frequency voltage components, and sheath grounding current transformers are used simultaneously to extract low-frequency current components, thereby capturing the external interference background.

[0003] When processing multi-channel sensor data, existing monitoring gateways typically employ adaptive filtering algorithms such as Normalized Least Mean Square (NLMS). These algorithms continuously update adaptive weight vectors, fitting and subtracting common-mode interference components to extract the true insulation partial discharge voltage. However, conventional adaptive filters, when updating weights, cannot distinguish between internal insulation partial discharge signals and external steady-state background noise. They are prone to misinterpreting true insulation partial discharge signals as common-mode interference signals and discarding them, resulting in low accuracy of the obtained insulation partial discharge voltage and an inability to accurately monitor the insulation status of high-voltage cables. Summary of the Invention

[0004] To address the problem of low accuracy in obtaining partial discharge voltage when monitoring the insulation status of cables using existing technologies, the present invention aims to provide a high-precision partial discharge monitoring method for high-voltage cables with signal denoising. The specific technical solution adopted is as follows: This invention proposes a high-precision partial discharge monitoring method for high-voltage cables with signal denoising, the method comprising: The partial discharge voltage and sheath grounding current are obtained at each sampling moment within the current monitoring period; the sampling moments include buffered sampling moments and real-time sampling moments. Based on the temporal difference characteristics between any two real-time sampling moments within the current monitoring period, as well as the fluctuation characteristics of partial discharge voltage and sheath grounding current within the time interval between the two real-time sampling moments, and combined with the partial discharge voltage and sheath grounding current at the buffer sampling moment, the dual-channel topology overlap degree of the current monitoring period is analyzed. An adaptive weight vector is initialized and set. The adaptive weight vector of the previous real-time sampling time adjacent to each real-time sampling time is updated using the dual-channel topology overlap to obtain the adaptive weight vector of each real-time sampling time. Based on the data characteristics of the sheath grounding current of each real-time sampling time and the historical sampling time, combined with the adaptive weight vector and partial discharge voltage of each real-time sampling time, the actual discharge voltage of each real-time sampling time is determined.

[0005] Further, the step of analyzing the dual-channel topology overlap of the current monitoring period based on the temporal difference characteristics between any two real-time sampling moments within the current monitoring period, and the fluctuation characteristics of the partial discharge voltage and the sheath grounding current within the time interval between the two real-time sampling moments, combined with the partial discharge voltage and sheath grounding current at the buffer sampling moment, includes: Based on the distribution range of partial discharge voltage at each buffer sampling time within the current monitoring cycle, the partial discharge voltage at each real-time sampling time is calculated to obtain the standard voltage value at each real-time sampling time. Based on the distribution range of sheath grounding current at the current monitoring cycle and the buffer sampling time, the sheath grounding current at each real-time sampling time is calculated to obtain the standard current value at each real-time sampling time. Based on the temporal difference characteristics between any two real-time sampling moments within the current monitoring period, and combined with the fluctuation characteristics of the standard voltage value within the time interval between any two real-time sampling moments, a partial discharge correlation matrix is ​​constructed. Based on the time sequence difference characteristics between any two real-time sampling moments within the current monitoring period, and combined with the fluctuation characteristics of the standard current value within the time interval between any two real-time sampling moments, a sheath grounding correlation matrix is ​​constructed. By integrating the partial discharge correlation matrix and the sheath grounding correlation matrix, the dual-channel topology overlap for the current monitoring period is obtained.

[0006] Further, the step of calculating the partial discharge voltage at each real-time sampling moment based on the distribution range of the partial discharge voltage at the buffered sampling moment within the current monitoring period to obtain the standard voltage value at each real-time sampling moment includes: The maximum value of the partial discharge voltage at all buffer sampling times within the current monitoring period is taken as the maximum voltage value. The minimum value of the partial discharge voltage at all buffer sampling times within the current monitoring period is taken as the minimum voltage value; The partial discharge voltage at each real-time sampling moment is normalized using the maximum and minimum voltage values ​​to obtain the standard voltage value at each real-time sampling moment.

[0007] Furthermore, the partial discharge correlation matrix is ​​constructed based on the temporal difference characteristics between any two real-time sampling moments within the current monitoring period, combined with the fluctuation characteristics of the standard voltage value within the time interval between any two real-time sampling moments, including: The real-time sampling time includes a sequence index; based on the sequence index of each real-time sampling time within the current monitoring period, a partial discharge correlation matrix is ​​constructed; the rows and columns of the partial discharge correlation matrix are both sequence indices of each real-time sampling time arranged in ascending numerical order. For any position in the partial discharge correlation matrix, if the absolute value of the difference between the row index and the column index of the position is equal to 1, then the element of the position is set to 1. If the absolute value of the difference between the row index and column index of the position is greater than the preset maximum visible threshold, or if the absolute value of the difference is equal to 0, then the element at the position is set to 0. If the absolute value of the difference between the row index and column index of the location is greater than 1, and the absolute value of the difference is less than or equal to the preset maximum visible threshold, then the element of the location is determined based on the fluctuation characteristics of the standard voltage value within the time interval between the two real-time sampling times corresponding to the row index and column index of the location.

[0008] Further, determining the element at the position based on the fluctuation characteristics of the standard voltage value within the time interval between the two real-time sampling times corresponding to the row index and column index of the position includes: The real-time sampling time corresponding to the minimum value between the row index and column index of the position is taken as the starting time, and the real-time sampling time corresponding to the maximum value between the row index and column index of the position is taken as the ending time. Linear fitting is performed on the standard voltage values ​​at the start and end times, and the result of the linear fitting is used as the global rate of change. Linear fitting is performed on the standard voltage value at any real-time sampling time within the time interval between the start time and the end time, and the result of the linear fitting is used as the local rate of change. If the difference between all local rates of change and the global rate of change in the time interval between the start time and the end time is less than or equal to a preset fluctuation tolerance threshold, then the element at the position is set to 1; otherwise, the element at the position is set to 0.

[0009] Further, the fusion of the partial discharge correlation matrix and the sheath grounding correlation matrix to obtain the dual-channel topology overlap for the current monitoring period includes: The similarity between the partial discharge correlation matrix and the sheath grounding correlation matrix is ​​measured using the Dessian similarity coefficient algorithm, and the result is used as the dual-channel topology overlap.

[0010] Further, the step of determining the actual discharge voltage at each real-time sampling moment based on the data characteristics of the sheath grounding current at each real-time sampling moment and historical sampling moments, combined with the adaptive weight vector and partial discharge voltage at each real-time sampling moment, includes: Take any real-time sampling moment within the current monitoring period as the current real-time sampling moment, and construct the sheath grounding data vector based on the sheath grounding current at each sampling moment within the preset backtracking window of the current real-time sampling moment; The interference estimation voltage is obtained by multiplying the sheath grounding data vector with the adaptive weight vector at the current real-time sampling time. The difference between the partial discharge voltage at the current real-time sampling moment and the estimated interference voltage is taken as the actual discharge voltage at the current real-time sampling moment.

[0011] Further, the initialization setting of the adaptive weight vector involves updating the adaptive weight vector of the previous real-time sampling time adjacent to each real-time sampling time using the dual-channel topology overlap, to obtain the adaptive weight vector for each real-time sampling time, including: Based on the temporal characteristics of the current monitoring period and combined with the dual-channel topological overlap, the dynamic step size control factor of the current monitoring period is determined. The adaptive weight vector is initialized to a zero vector. The normalized least mean square algorithm is used to process the sheath grounding data vector of the previous real-time sampling time adjacent to the current real-time sampling time, the adaptive weight vector of the previous real-time sampling time, the actual discharge voltage of the previous real-time sampling time, and the dynamic step size control factor to obtain the adaptive weight vector of the current real-time sampling time.

[0012] Furthermore, the determination of the dynamic step size control factor for the current monitoring period based on the time-series characteristics of the current monitoring period and the dual-channel topological overlap includes: If the total number of current monitoring cycles and historical monitoring cycles is less than or equal to the preset number of cold start convergence cycles, then the preset base step size is used as the dynamic step size control factor for the current monitoring cycle; if the total number of current monitoring cycles and historical monitoring cycles is greater than the preset number of cold start convergence cycles, then the product of the preset base step size and the dual-channel topology overlap is used as the dynamic step size control factor for the current monitoring cycle.

[0013] Furthermore, the method for obtaining the monitoring period is as follows: Any sampling time is taken as a candidate time. If the partial discharge voltage of the candidate time is greater than the preset hardware trigger threshold, then the candidate time is taken as the transient event trigger time. The start time of the cycle is obtained by subtracting the trigger time of the transient event from the preset backtracking time; the sampling time within the time interval between the start time of the cycle and the trigger time of the transient event is used as the cache sampling time. The transient event trigger time is added to the preset duration to obtain the cycle end time; the sampling time within the time interval between the transient event trigger time and the cycle end time is used as the real-time sampling time. The time interval between the start time and the end time of the cycle is taken as the monitoring cycle.

[0014] The present invention has the following beneficial effects: This invention analyzes the dual-channel topological overlap of the current monitoring period based on the temporal difference characteristics between any two real-time sampling moments within the current monitoring period, as well as the fluctuation characteristics of partial discharge voltage and sheath grounding current within the time interval between these two real-time sampling moments. It combines the partial discharge voltage and sheath grounding current at the buffer sampling moment of the current monitoring period. By combining the time span constraint with the fluctuation characteristics of the signal within the time span, the influence of absolute time misalignment is eliminated, enabling the partial discharge monitoring gateway to quickly and accurately identify the essential differences in waveform morphology even in complex industrial electromagnetic interference environments. Simultaneously, the one-dimensional temporal signals of the high-frequency voltage channel and low-frequency current channel are standardized using a clean, stable-state electrical noise baseline without abnormal abrupt changes before the partial discharge transient pulse trigger. This avoids the problem of excessive compression of waveform characteristics caused by high-amplitude abnormal pulses and prevents numerical distortion of the dual-channel topological overlap caused by high-amplitude abnormal pulse distortion.

[0015] An adaptive weight vector is initialized and updated using the dual-channel topology overlap of the current monitoring period, updating the adaptive weight vector of the previous real-time sampling time adjacent to each real-time sampling time. This results in an adaptive weight vector for each real-time sampling time, which can track external interference characteristics while preventing erroneous learning of internal real discharge pulses. When a real partial discharge occurs, the dual-channel topology overlap drops sharply, causing the adaptive weight vector to stop iterating, thus preventing the underlying algorithm from misinterpreting the real discharge pulse as an external common-mode interference signal and canceling it out.

[0016] Based on the data characteristics of the sheath grounding current at each real-time sampling moment and historical sampling moment, combined with the adaptive weight vector and partial discharge voltage at each real-time sampling moment, the actual discharge voltage at each real-time sampling moment is determined. This fully reconstructs the time lag effect and waveform distortion law caused by multiple reflections of the common-mode interference signal, achieving high-precision filtering of external common-mode interference signals and obtaining a true and accurate insulation partial discharge voltage, facilitating precise assessment of the insulation status of high-voltage cables. Attached Figure Description

[0017] Figure 1 The flowchart illustrates the steps of a high-precision partial discharge monitoring method for high-voltage cables with signal denoising, as provided in one embodiment of the present invention. Figure 2 This is a flowchart illustrating the steps of a method for obtaining dual-channel topology overlap according to an embodiment of the present invention. Figure 3 This is a flowchart illustrating the steps of an adaptive weight vector update method provided in an embodiment of the present invention. Detailed Implementation

[0018] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.

[0019] The following description, in conjunction with the accompanying drawings, details the specific scheme of a high-precision partial discharge monitoring method for high-voltage cables for signal denoising provided by the present invention.

[0020] Please see Figure 1 The diagram illustrates a flowchart of a high-precision partial discharge monitoring method for high-voltage cables with signal denoising according to an embodiment of the present invention. The method includes: Step S10: Obtain the partial discharge voltage and sheath grounding current at each sampling moment within the current monitoring period; the sampling moment includes the buffered sampling moment and the real-time sampling moment.

[0021] This invention operates inside the partial discharge monitoring gateway of a high-voltage cable, receiving the partial discharge voltage and sheath grounding current at each sampling moment output by the analog-to-digital converter of the partial discharge monitoring gateway.

[0022] The partial discharge monitoring gateway integrates an analog-to-digital converter (ADC) module and a trigger control circuit. A capacitively coupled sensor installed at the high-voltage cable joint continuously senses and outputs a high-frequency voltage analog signal; a current transformer installed on the grounding wire continuously senses and outputs a low-frequency current analog signal. After these continuous analog electrical signals are connected to the partial discharge monitoring gateway, the ADC, driven by a shared clock, performs synchronous discretization sampling of the continuous analog signals from the high-frequency voltage channel and the low-frequency current channel at a fixed sampling frequency: the high-frequency voltage analog signal is converted from analog to digital to obtain the partial discharge voltage at each sampling moment; the low-frequency current analog signal is also converted from analog to digital to obtain the sheath grounding current at each sampling moment.

[0023] In one specific implementation of this invention, the fixed sampling frequency of the analog-to-digital converter is set to 50 MHz.

[0024] Preferably, in some implementations of the embodiments of the present invention, the method for obtaining the monitoring period is as follows: Any sampling time is taken as a candidate time. If the partial discharge voltage of the candidate time is greater than the preset hardware trigger threshold, then the candidate time is taken as the transient event trigger time. The start time of the cycle is obtained by subtracting the trigger time of the transient event from the preset backtracking time; the sampling time within the time interval between the start time of the cycle and the trigger time of the transient event is used as the cache sampling time. The transient event trigger time is added to the preset duration to obtain the cycle end time; the sampling time within the time interval between the transient event trigger time and the cycle end time is used as the real-time sampling time. The time interval between the start time and the end time of the cycle is taken as the monitoring cycle.

[0025] The preset hardware trigger threshold represents the upper limit of the maximum voltage level that common-mode noise and ambient white noise can reach under normal operating conditions without partial discharge in a high-voltage cable. When the partial discharge voltage exceeds the preset hardware trigger threshold, it is determined that a partial discharge transient pulse has occurred in the high-voltage cable, and the data interception operation of the monitoring cycle is initiated at this time.

[0026] The preset backtracking duration represents the time span required to extract the steady-state background before triggering and construct historical data slices of the underlying filter. The preset duration represents the complete time span of a typical partial discharge transient pulse and its accompanying common-mode interference, from occurrence, oscillation to decay. The monitoring cycle duration, composed of the preset backtracking duration and the preset duration, covers a period of stable background noise without abnormalities before the partial discharge transient pulse is triggered, as well as the complete dynamic process after triggering, from abrupt change, oscillation to decay. It fully preserves the rising edge of the partial discharge transient pulse from calm to abrupt change, preventing signal rise edge loss due to trigger delay of hardware circuitry, and truly reflects the steady-state electrical noise floor reference of the cable at the moment of partial discharge transient pulse outbreak. At the same time, it can capture the complete partial discharge characteristic waveform and the accompanying common-mode interference waveform, avoiding partial discharge signal truncation and ensuring the integrity of feature extraction.

[0027] For any given monitoring period, the sampling times within the time interval between the start of the period and the triggering time of the transient event are used as buffered sampling times to provide steady-state background clutter without abnormal changes, so as to extract the background reference value and provide a true continuous historical data slice for subsequent updates of the adaptive weight vector using the adaptive filter; the sampling times within the time interval between the triggering time of the transient event and the end of the period are used as real-time sampling times to capture the complete abnormal transient pulse and the accompanying common-mode interference signal, which are used for subsequent analysis of the common-mode interference component in the partial discharge voltage and to perform interference cancellation operations.

[0028] To facilitate the subsequent discretization and time-series traversal calculation of the acquired data from the high-frequency voltage channel and the low-frequency current channel, this embodiment assigns a sequence number index from 1 to N to each real-time sampling moment within the monitoring period in chronological order.

[0029] In this embodiment, the latest monitoring period obtained under the absolute system time of the partial discharge monitoring gateway is used as the current monitoring period.

[0030] In one specific implementation of this invention, the method for obtaining the preset hardware trigger threshold is as follows: During the initial installation and operation of the partial discharge monitoring gateway, and after confirming that the monitored high-voltage cable is in a normal operating state without insulation defects, a continuous background noise signal for a duration (e.g., 200 milliseconds) is continuously collected. The maximum value of the partial discharge voltage is multiplied by a preset margin coefficient to obtain the preset hardware trigger threshold. In this embodiment, the preset hardware trigger threshold is 150 millivolts.

[0031] The preset margin factor represents a safety buffer margin reserved to cope with random fluctuations in background noise in the power grid environment. Since the peak value of steady-state background clutter in actual industrial environments is not absolutely constant, directly using the absolute maximum value of the partial discharge voltage collected within the preset time period as the preset hardware trigger threshold is easily triggered by extremely small conventional electromagnetic fluctuations. Multiplying it by the preset margin factor ensures that normal background fluctuations are filtered out, responding only to real abnormal transient pulses with significant steps in energy and amplitude. The preset margin factor is conventionally taken as 1.2 to 2.0, and in this embodiment, it is preferably 1.5.

[0032] In one specific implementation of this invention, the preset backtracking duration is obtained as follows: based on the requirement of the adaptive filter to fully fit the multipath propagation delay characteristics of the common-mode interference signal in the high-voltage cable sheath, the preset backtracking duration typically corresponds to the time span spanned by 64 to 256 sampling moments output by the analog-to-digital converter. In this embodiment, combined with a fixed sampling frequency of 50MHz, the preset backtracking duration preferably corresponds to 128 sampling moments, i.e., set to 2.56 microseconds.

[0033] In one specific implementation of this invention, the preset duration is obtained as follows: based on the physical characteristic that typical high-frequency partial discharge signals usually decay completely within a few microseconds to tens of microseconds, the preset duration typically corresponds to the time span spanned by 512 to 4096 sampling moments output by the analog-to-digital converter. In this embodiment, combined with a fixed sampling frequency of 50MHz, the preset duration preferably corresponds to 1024 sampling moments, i.e., set to 20.48 microseconds.

[0034] Step S20: Based on the time sequence difference characteristics between any two real-time sampling moments in the current monitoring period, and the fluctuation characteristics of partial discharge voltage and sheath grounding current in the time interval between the two real-time sampling moments, combined with the partial discharge voltage and sheath grounding current at the buffer sampling moment, analyze the dual-channel topology overlap of the current monitoring period.

[0035] Dual-channel topology coincidence represents the degree of similarity in the macroscopic waveform fluctuations of the one-dimensional time-series signals of the high-frequency voltage channel and the low-frequency current channel within the duration of the partial discharge transient pulse in the current monitoring period. Dual-channel topology coincidence is used at the algorithmic level to distinguish between external common-mode interference (such as switching operations or corona pulses) and actual internal partial discharge. When external common-mode interference occurs, the interference pulse simultaneously enters both the high-frequency voltage channel and the low-frequency current channel, at which point the dual-channel topology coincidence approaches 1. When actual insulation partial discharge occurs inside the high-voltage cable, the high-frequency abrupt pulse is significant only in the high-frequency voltage channel, causing the connectivity network of the high-frequency voltage channel to be severely interrupted by the abrupt peak, while the connectivity network of the low-frequency current channel remains intact. At this point, the connectivity network structures of the high-frequency voltage channel and the low-frequency current channel exhibit significant differences, and the dual-channel topology coincidence approaches 0.

[0036] The timing difference between any two real-time sampling moments within the current monitoring period represents the time span distance between these two real-time sampling moments. The fluctuation characteristics of the partial discharge voltage and the sheath grounding current within the time interval between any two real-time sampling moments represent the signal fluctuation and oscillation evolution law of the one-dimensional timing signal of the high-frequency voltage channel and the low-frequency current channel due to external common-mode interference or internal real partial discharge within this time span. The partial discharge voltage and sheath grounding current at all buffered sampling moments within the current monitoring period represent the clean, steady-state electrical noise baseline without abnormal changes before the partial discharge transient pulse is triggered. By combining these three factors to analyze the dual-channel topology overlap degree of the current monitoring period, and combining the time span constraint with the signal fluctuation characteristics within the time span, the influence of absolute time misalignment can be eliminated. This allows the partial discharge monitoring gateway to quickly and accurately identify the essential differences in waveform morphology even in complex industrial electromagnetic interference environments. Meanwhile, the one-dimensional timing signals of the high-frequency voltage channel and the low-frequency current channel are standardized using a steady-state electrical noise floor reference, which avoids the problem of excessive compression of waveform characteristics caused by high-amplitude abnormal pulses and prevents numerical distortion of the dual-channel topology overlap caused by high-amplitude abnormal pulse distortion.

[0037] Preferably, such as Figure 2 As shown, in some implementations of this invention, based on the timing difference characteristics between any two real-time sampling moments within the current monitoring period, and the fluctuation characteristics of the partial discharge voltage and the sheath grounding current within the time interval between the two real-time sampling moments, combined with the partial discharge voltage and sheath grounding current at the buffer sampling moment, the dual-channel topology overlap degree of the current monitoring period is analyzed, including: Step S201: Based on the distribution range of partial discharge voltage at each real-time sampling moment within the current monitoring cycle, calculate the partial discharge voltage at each real-time sampling moment to obtain the standard voltage value at each real-time sampling moment.

[0038] The standard voltage value represents the relative fluctuation intensity of the partial discharge voltage at a specific real-time sampling moment relative to the steady-state local voltage noise after eliminating the difference between the DC bias and dimensional gain of the sensor hardware. The value ranges from 0 to 1.

[0039] The distribution range of partial discharge voltage at the current monitoring cycle reflects the upper and lower boundaries of the steady-state local voltage noise floor when the high-voltage cable is operating normally and without partial discharge. Based on this, the standard voltage value at each real-time sampling moment is calculated, eliminating the DC bias error introduced by the inherent properties of the sensor and the dimensional differences between different channels, and truly restoring the real fluctuation characteristics of the partial discharge signal relative to the steady-state local voltage noise floor.

[0040] More specifically, based on the distribution range of partial discharge voltage at each buffered sampling time within the current monitoring period, the partial discharge voltage at each real-time sampling time is calculated to obtain the standard voltage value for each real-time sampling time, including: The maximum value of the partial discharge voltage at all buffer sampling times within the current monitoring period is taken as the maximum voltage value. The minimum value of the partial discharge voltage at all buffer sampling times within the current monitoring period is taken as the minimum voltage value; The partial discharge voltage at each real-time sampling moment is normalized using the maximum and minimum voltage values ​​to obtain the standard voltage value at each real-time sampling moment.

[0041] This embodiment utilizes maximum and minimum value normalization to process the partial discharge voltage at each real-time sampling moment, obtaining a standard voltage value for each real-time sampling moment. Taking any real-time sampling moment as an example, the difference between the partial discharge voltage and the minimum voltage at that real-time sampling moment is used as the numerator; the difference between the maximum voltage and the minimum voltage is added to a preset zero-prevention constant (such as 10). -8 The result of the numerator is used as the denominator, and the ratio of the numerator to the denominator is used as the standard voltage value at that real-time sampling moment.

[0042] It should be understood that the method of normalizing the original data using maximum and minimum value normalization is a well-known technique in this field and will not be elaborated further.

[0043] Step S202: Based on the distribution range of the sheath grounding current at each real-time sampling moment within the current monitoring cycle, calculate the sheath grounding current at each real-time sampling moment to obtain the standard current value at each real-time sampling moment.

[0044] The standard current value represents the relative fluctuation intensity of the sheath grounding current signal at the real-time sampling moment relative to the steady-state grounding current noise after eliminating the static zero drift and dimensional gain differences introduced by the sensor hardware. The value ranges from 0 to 1.

[0045] The distribution range of the sheath grounding current at the current monitoring cycle reflects the upper and lower boundaries of the steady-state grounding current noise floor of the measurement circuit when the high-voltage cable sheath is operating normally and without high-frequency transient changes. Based on this, the standard current value at each real-time sampling moment is calculated, eliminating the steady-state grounding current noise floor introduced by the inherent properties of the sensor and the dimensional differences between different channels, and truly restoring the real fluctuation characteristics of the sheath grounding current signal relative to the steady-state grounding current noise floor.

[0046] The standard current value for each real-time sampling moment is calculated using the same method as in step S201 for calculating the standard voltage value at each sampling moment. It should be understood that this means the standard current value for each real-time sampling moment is obtained by replacing the partial discharge voltage with the sheath grounding current and the standard voltage value with the standard current value, following the same method used in step S201 for calculating the standard voltage value.

[0047] Step S203: Based on the temporal difference characteristics between any two real-time sampling moments within the current monitoring period, and combined with the fluctuation characteristics of the standard voltage value within the time interval between any two real-time sampling moments, construct a partial discharge correlation matrix.

[0048] The partial discharge correlation matrix represents the state matrix showing the interconnections between partial discharge signals at any two different real-time sampling moments within the duration of the transient pulse of partial discharge in the current monitoring period. Elements in the partial discharge correlation matrix represent whether there is a direct morphological correlation between partial discharge signals at any two real-time sampling moments, and their values ​​are either 0 or 1. A value of 1 indicates that the fluctuations of the partial discharge signals at any two real-time sampling moments are continuous and directly correlated, belonging to a smooth or complete macroscopic fluctuation structure; a value of 0 indicates that the fluctuations of the partial discharge signals at any two real-time sampling moments are truncated and uncorrelated, reflecting that the partial discharge signal underwent violent high-frequency oscillations or abrupt changes within that time span.

[0049] The timing difference between any two real-time sampling moments represents the time span between them. The fluctuation of the standard voltage value within the time interval between any two real-time sampling moments represents the signal fluctuation and oscillation evolution of the one-dimensional timing signal of the high-frequency voltage channel within that time span, caused by external common-mode interference or internal partial discharge. Therefore, constructing a partial discharge correlation matrix based on these characteristics can integrate the evolution of the partial discharge signal over time with its morphological fluctuation characteristics, eliminating the influence of the absolute amplitude of the partial discharge signal and the time misalignment caused by hardware delay. This allows the partial discharge monitoring gateway to robustly capture the essential morphological characteristics of the partial discharge waveform even in complex industrial electromagnetic interference environments.

[0050] Furthermore, based on the temporal difference characteristics between any two real-time sampling moments within the current monitoring period, and combined with the fluctuation characteristics of the standard voltage value within the time interval between any two real-time sampling moments, a partial discharge correlation matrix is ​​constructed, including: Based on the index of each real-time sampling moment within the current monitoring period, an N×N partial discharge correlation matrix is ​​constructed. The rows and columns of the partial discharge correlation matrix are the indexes of each real-time sampling moment, arranged in ascending numerical order. Here, N is the total number of all real-time sampling moments within the current monitoring period, which is also the maximum value of the index. It should be understood that the partial discharge correlation matrix constructed at this point is a blank matrix.

[0051] For any position in the partial discharge correlation matrix, if the absolute value of the difference between the row index and the column index of that position is equal to 1, it means that the two real-time sampling times corresponding to that position are closely adjacent, and the partial discharge signals between them must be absolutely continuous and directly correlated. In this case, the element at that position is set to 1.

[0052] If the absolute value of the difference between the row index and column index at a given position is greater than the preset maximum visible threshold, it means that the partial discharge signal at that position has undergone multiple independent fluctuations within the time span between the two real-time sampling moments, and the fluctuation process is truncated and unrelated. If the absolute value of the difference is equal to 0, it means that the position corresponds to the same real-time sampling moment and has no fluctuation correlation itself. For either of the above cases, the element at that position is set to 0. The preset maximum visible threshold represents the upper limit of the maximum time span during which a smooth or complete partial discharge signal fluctuation process can remain continuous and directly correlated. Exceeding this time span means that the partial discharge signal has necessarily undergone a turning point or entered the next independent fluctuation stage, and no longer belongs to the same monotonically fluctuating macroscopic fluctuation structure.

[0053] If the absolute value of the difference between the row index and column index at a given position is less than or equal to a preset maximum visible threshold, and the absolute value of the difference is greater than 1, it indicates that the time span between the two real-time sampling moments corresponding to that position is within the maximum time span limit that allows the partial discharge signal fluctuation process to remain continuous and directly correlated. However, there may be one or more real-time sampling moments between these two moments. Whether the partial discharge signal between these two real-time sampling moments belongs to the same continuous and smooth fluctuation process depends on the actual fluctuation of the partial discharge signal between these two real-time sampling moments. Therefore, it is necessary to determine the element value at this position based on the fluctuation characteristics of the standard voltage value within the time interval between the two real-time sampling moments corresponding to the row index and column index at that position.

[0054] In one specific implementation of this invention, the engineering experience range for the preset maximum visible threshold for high-voltage cable partial discharge monitoring is 20 to 50. The specific value of the preset maximum visible threshold depends on the sampling frequency of the analog-to-digital converter (ADC) and the frequency band of the common-mode interference signal in the field: when the ADC sampling frequency is high, or the common-mode interference signal frequency is low, a larger maximum visible threshold (e.g., 35 to 50) is selected. When the ADC sampling frequency is low, or the common-mode interference signal frequency is high, a smaller maximum visible threshold (e.g., 20 to 35) is selected. In this embodiment, the preset maximum visible threshold is set to 30, meaning the maximum span of signal morphology association is an upper limit of a time span including 30 sampling times.

[0055] Further, based on the fluctuation characteristics of the standard voltage value within the time interval between the two real-time sampling times corresponding to the row index and column index of the location, the elements of the location are determined, including: The real-time sampling time corresponding to the minimum value between the row index and column index of the position is taken as the starting time, and the real-time sampling time corresponding to the maximum value between the row index and column index of the position is taken as the ending time. Linear fitting is performed on the standard voltage values ​​at the start and end times, and the result of the linear fitting is used as the global rate of change. Linear fitting is performed on the standard voltage value at any real-time sampling time within the time interval between the start time and the end time, and the result of the linear fitting is used as the local rate of change. If the absolute value of the difference between all local rates of change and the global rate of change in the time interval between the start time and the end time is less than or equal to a preset fluctuation tolerance threshold, then the element at the position is set to 1; otherwise, the element at the position is set to 0.

[0056] The global rate of change characterizes the theoretical trend of the partial discharge signal as it smoothly transitions from the starting point to the ending point. It serves as a reference baseline between the two points and reflects the expected slope of the partial discharge signal's evolution over this time span. The local rate of change characterizes the actual trend of the partial discharge signal over a specific real-time sampling time span within the time interval between the starting point and the ending point.

[0057] Specifically, this embodiment uses the two-point slope formula to calculate the global rate of change and the local rate of change. Taking the calculation of the global rate of change as an example, the difference between the standard voltage value at the end time and the standard voltage value at the start time is divided by the difference between the index of the end time and the index of the start time, and the result is taken as the global rate of change.

[0058] It should be understood that the method of calculating the rate of change between two points using the two-point slope formula is a well-known technique in this field and will not be elaborated further.

[0059] The preset fluctuation tolerance threshold represents the allowable margin for high-frequency white noise or small random jitter in the hardware of the partial discharge monitoring gateway. If the absolute value of the difference between the local rate of change and the global rate of change in the time interval between the start time and the end time is less than or equal to the preset fluctuation tolerance threshold, it means that the fluctuation of the partial discharge signal is within the allowable margin of white noise during the time span from the start time to the end time. This indicates that the waveform fluctuation between the start time and the end time belongs to the same continuous fluctuation structure that is not truncated by high-frequency abrupt changes, and the two are directly related in terms of morphological evolution. Therefore, the element at this position is set to 1.

[0060] If the absolute value of the difference between any local rate of change and the global rate of change is greater than the preset fluctuation tolerance threshold, it means that the fluctuation of the partial discharge signal exceeds the tolerance margin of white noise during the time span from the start time to the end time. This indicates that a real and independent abnormal high-frequency spike signal has appeared in the evolution of the partial discharge signal. The waveform fluctuation between the start time and the end time does not belong to the same monotonous macroscopic fluctuation structure. Therefore, the element at this position is set to 0.

[0061] In one specific implementation of this invention, the preset fluctuation tolerance threshold is empirically set to a range of 0.05 to 0.15, with the specific value depending on the electromagnetic background noise level of the environment and the sampling accuracy of the analog-to-digital converter (ADC). When the electromagnetic background noise of the environment is high, or the sampling frequency of the ADC is low, the preset fluctuation tolerance threshold is set to a larger value (e.g., 0.10 to 0.15) to prevent morphological connectivity from being interrupted by false white noise. When the electromagnetic shielding of the environment is good, or the sampling frequency of the ADC is high, the preset fluctuation tolerance threshold is set to a smaller value (e.g., 0.05 to 0.10) to prevent weak real partial discharge spikes from being incorrectly included by the threshold. In this embodiment, the preset fluctuation tolerance threshold is set to 0.10.

[0062] Step S204: Based on the time sequence difference characteristics between any two real-time sampling moments within the current monitoring period, and combined with the fluctuation characteristics of the standard current value within the time interval between any two real-time sampling moments, construct the sheath grounding correlation matrix.

[0063] The sheath grounding correlation matrix represents the state matrix of interconnections between sheath grounding current signals at any two different real-time sampling times within the duration of the partial discharge transient pulse in the current monitoring period. Elements in the sheath grounding correlation matrix represent whether there is a direct morphological correlation between the sheath grounding current signals at any two real-time sampling times, and their values ​​are either 0 or 1. A value of 1 indicates that the fluctuations of the sheath grounding current signals at any two real-time sampling times are continuous and directly correlated, belonging to a smooth or complete macroscopic fluctuation structure; a value of 0 indicates that the fluctuations of the sheath grounding current signals at any two real-time sampling times are truncated and uncorrelated, reflecting that the sheath grounding current signal experienced severe high-frequency oscillations or abrupt changes within that time span.

[0064] The timing difference between any two real-time sampling moments represents the time span distance between these two real-time sampling moments; the fluctuation characteristics of the standard current value within the time interval between any two real-time sampling moments represent the signal fluctuation and oscillation evolution law of the one-dimensional timing signal of the low-frequency current channel within that time span due to external common-mode interference or internal real partial discharge. Therefore, constructing a sheath grounding correlation matrix based on the above characteristics can integrate the evolution law and morphological fluctuation characteristics of the sheath grounding current signal over the time span, stripping away the time misalignment effect caused by the absolute amplitude of the sheath grounding current signal and hardware delay, enabling the partial discharge monitoring gateway to robustly capture the essential morphological characteristics of the sheath grounding current waveform even in complex industrial electromagnetic interference environments.

[0065] The sheath grounding correlation matrix is ​​constructed using the exact same method as in step S203 when constructing the partial discharge correlation matrix, and the same preset maximum visibility threshold is used in both methods. It should be understood that the standard voltage values ​​in the above method are replaced with standard current values, and the element values ​​at each position in the sheath grounding correlation matrix are determined using the same method as for determining the element values ​​at each position in the partial discharge correlation matrix.

[0066] It should be noted that both the partial discharge correlation matrix and the sheath grounding correlation matrix are N×N matrices, and each element in the two matrices corresponds one-to-one. Specifically, the element in the i-th row and j-th column of the partial discharge correlation matrix and the element in the i-th row and j-th column of the sheath grounding correlation matrix are used to characterize the signal evolution characteristics within the same time span (i.e., the time interval between the i-th and j-th real-time sampling moments). This complete consistency in row and column coordinates ensures that the waveform morphology correlation between the high-frequency voltage channel and the low-frequency current channel within the same time span can be accurately aligned during subsequent comparisons, laying a reliable data foundation for the quantitative calculation of topological overlap.

[0067] Step S205: Merge the partial discharge correlation matrix and the sheath grounding correlation matrix to obtain the dual-channel topology overlap for the current monitoring period.

[0068] Based on the two-dimensional network topology recorded by the partial discharge correlation matrix and the sheath grounding correlation matrix, a cross-dimensional quantitative comparison of the one-dimensional time-series signals of the high-frequency voltage channel and the low-frequency current channel in terms of waveform macro-fluctuations and transient evolution laws can be made. It can accurately identify the consistency of the morphological correlation of the signal fluctuations of the two channels on the time axis, and abandon the direct dependence on the absolute amplitude and phase delay of the one-dimensional time-series signal. It can keenly and accurately distinguish between external common-mode interference and internal real partial discharge in complex industrial electromagnetic backgrounds.

[0069] Furthermore, by fusing the partial discharge correlation matrix and the sheath grounding correlation matrix, the dual-channel topology overlap for the current monitoring period is obtained, including: The similarity between the partial discharge correlation matrix and the sheath grounding correlation matrix is ​​measured using the Dessian similarity coefficient algorithm, and the result is used as the dual-channel topology overlap.

[0070] Specifically, each element in the partial discharge correlation matrix is ​​multiplied by its corresponding element in the sheath grounding correlation matrix. The results of all the multiplication operations are summed, and twice the sum is used as the numerator. The sum of each element in the partial discharge correlation matrix and the sum of each element in the sheath grounding correlation matrix are calculated. The sum of the sums of the elements in the two matrices is added together and used as the denominator. The ratio of the numerator to the denominator is used as the dual-channel topology overlap for the current monitoring period.

[0071] It should be understood that using the Dessian similarity coefficient algorithm to measure the topological similarity of two binary matrices is a well-known technique in this field, and will not be elaborated upon here.

[0072] Step S30: Initialize and set the adaptive weight vector. Update the adaptive weight vector of the previous real-time sampling time adjacent to each real-time sampling time using the dual-channel topology overlap to obtain the adaptive weight vector of each real-time sampling time. Based on the data characteristics of the sheath grounding current of each real-time sampling time and the historical sampling time, and combined with the adaptive weight vector and partial discharge voltage of each real-time sampling time, determine the actual discharge voltage of each real-time sampling time.

[0073] The adaptive weight vector at each real-time sampling moment represents a set of optimal coupling ratio coefficients used when converting the sheath grounding current at each real-time sampling moment and the common-mode interference component in the partial discharge voltage at each real-time sampling moment from the historical continuous sampling moment.

[0074] Because the operating environment and grounding status of high-voltage cables drift over time, the propagation pattern of external common-mode interference signals into the high-frequency voltage channel is not constant. Therefore, the adaptive weight vector must be updated in real time to ensure it accurately tracks the latest interference coupling environment. Utilizing dual-channel topology overlap to regulate this update process can prevent erroneous learning of internal real discharge pulses while tracking external interference characteristics. When a real partial discharge occurs, the dual-channel topology overlap drops sharply, causing the adaptive weight vector to stop iterating. This completely prevents the underlying algorithm from misinterpreting real discharge pulses as external common-mode interference signals and canceling them out.

[0075] Because external common-mode interference undergoes multipath reflection and refraction when propagating within the cable's metallic sheath, the sheath grounding current captured by the partial discharge monitoring gateway at a given real-time sampling moment is essentially the sum of common-mode interference signals arriving along different paths at multiple consecutive historical sampling moments prior to that real-time sampling moment (i.e., time delay and dispersion effects). Therefore, based on the data characteristics of the sheath grounding current at each real-time sampling moment and historical sampling moment, it is possible to fully reconstruct the time lag and waveform distortion caused by multiple reflections of the common-mode interference signal.

[0076] By combining the data characteristics of the sheath grounding current at each real-time sampling moment with those at historical sampling moments, as well as the adaptive weight vector at each real-time sampling moment, the common-mode interference component in the partial discharge voltage at each real-time sampling moment can be accurately calculated. Then, by canceling the partial discharge voltage with the common-mode interference component, the external common-mode interference signal can be filtered out with high precision, and the actual discharge voltage at each real-time sampling moment can be accurately determined.

[0077] It should be noted that the historical sampling time in this embodiment includes both the cached sampling time and the real-time sampling time within the current monitoring period.

[0078] Preferably, in some implementations of the present invention, the actual discharge voltage at each real-time sampling moment is determined based on the data characteristics of the sheath grounding current at each real-time sampling moment and the historical sampling moment, combined with the adaptive weight vector and partial discharge voltage at each real-time sampling moment, including: Take any real-time sampling moment within the current monitoring period as the current real-time sampling moment, and construct the sheath grounding data vector based on the sheath grounding current at each sampling moment within the preset backtracking window of the current real-time sampling moment; The interference estimation voltage is obtained by multiplying the sheath grounding data vector with the adaptive weight vector at the current real-time sampling time. The difference between the partial discharge voltage at the current real-time sampling moment and the estimated interference voltage is taken as the actual discharge voltage at the current real-time sampling moment.

[0079] In this embodiment, any real-time sampling moment within the current monitoring period is taken as the current real-time sampling moment. Taking the current real-time sampling moment as an example, it is used as the last sampling moment of the preset backtracking window. The sheath grounding current at each sampling moment within the preset backtracking window is arranged in chronological order to construct a sheath grounding data vector. The sheath grounding data vector represents the continuous evolution of the sheath grounding current signal formed by multipath propagation and reflection after external common-mode interference enters the high-voltage cable sheath from the observation perspective of the current real-time sampling moment.

[0080] The adaptive weight vector at the current real-time sampling moment represents the latest high-voltage cable interference transfer function model maintained by the adaptive filter deployed inside the partial discharge monitoring gateway through continuous learning and iterative updates up to the current real-time sampling moment. It reflects the physical propagation path characteristics and attenuation and delay laws of external common-mode interference coupling from the low-frequency current channel to the high-frequency voltage channel at the current real-time sampling moment.

[0081] It should be noted that the sheath grounding data vector and the adaptive weight vector at the current real-time sampling moment are in one-to-one correspondence in dimensional length (both are M-dimensional vectors, where M is the total number of sampling moments included in the preset backtracking window), and the elements in the adaptive weight vector physically represent the transfer impedance from the low-frequency current channel to the high-frequency voltage channel, with the unit being ohms.

[0082] Specifically, the process of multiplying the sheath grounding data vector at the current real-time sampling moment with the adaptive weight vector is as follows: each element in the sheath grounding data vector is multiplied sequentially by its corresponding element in the adaptive weight vector, and all product results are summed. The resulting value is used as the interference estimation voltage. The interference estimation voltage represents the background interference voltage that external common-mode interference would theoretically generate in the high-frequency voltage channel without including any internal insulation defect discharge.

[0083] Subtracting the interference-estimated voltage from the partial discharge voltage at the current real-time sampling moment yields the actual discharge voltage at the current real-time sampling moment. This reflects the amplitude of the pure discharge pulse generated independently within the high-voltage cable insulation due to local breakdown or discharge defects, uncontaminated by external environmental noise. It is the ultimate core data for assessing the health status of cable insulation, diagnosing defect types, and providing early warning of breakdown risks.

[0084] In one specific implementation of this invention, the length of the preset backtracking window is consistent with the preset backtracking duration.

[0085] In this embodiment of the invention, after obtaining the actual discharge voltage at all real-time sampling moments within the current monitoring period, the actual discharge voltages at all real-time sampling moments are arranged and combined in chronological order to construct an actual voltage sequence. This actual voltage sequence is uploaded to the monitoring and diagnostic platform through the communication interface (such as an Ethernet interface or a fiber optic interface) of the partial discharge monitoring gateway, for use by downstream diagnostic modules for identifying the type and severity of high-voltage cable insulation defects.

[0086] Preferably, such as Figure 3 As shown, in some implementations of this invention, the adaptive weight vector is initialized, and the adaptive weight vector of the previous real-time sampling time adjacent to each real-time sampling time is updated using the dual-channel topology overlap to obtain the adaptive weight vector of each real-time sampling time, including: Step S301: Based on the temporal characteristics of the current monitoring period and combined with the dual-channel topological overlap, determine the dynamic step size control factor for the current monitoring period.

[0087] The dynamic step size control factor represents the learning rate or step size coefficient used by the adaptive filter deployed in the partial discharge monitoring gateway when updating the adaptive weight vector in the current monitoring period. It determines the adaptive filter's tracking sensitivity and weight correction magnitude to the actual discharge voltage at the current real-time sampling moment.

[0088] Based on the timing characteristics of the current monitoring period and the dual-channel topology overlap, the dynamic step size control factor is determined, enabling selective dynamic sleep and deadlock prevention control of the adaptive filter. This ensures rapid convergence to the real steady-state environment background during the initial cold start of the partial discharge monitoring gateway, avoiding weight update deadlock caused by the first captured real partial discharge pulse during the cold start phase. Furthermore, by combining the dual-channel topology overlap, when a real partial discharge is detected that causes a break in the signal waveform characteristics of the high-frequency voltage channel and the low-frequency current channel, the feedback path of the adaptive filter learning from the partial discharge signal as a steady-state background is cut off, ensuring high-precision extraction of real partial discharge features.

[0089] Furthermore, based on the temporal characteristics of the current monitoring period and combined with the dual-channel topological overlap, the dynamic step size control factor for the current monitoring period is determined, including: If the total number of current monitoring cycles and historical monitoring cycles is less than or equal to the preset number of cold start convergence cycles, corresponding to the cold start initialization phase after the partial discharge monitoring gateway is powered on or reset, in order to prevent the actual discharge pulse from being captured at the initial stage of power-on, resulting in zero overlap of the dual-channel topology, which in turn leads to zero step size and permanent deadlock of the adaptive filter, the preset basic step size is used as the dynamic step size control factor of the current monitoring cycle to ensure that the adaptive filter can quickly establish the initial steady-state background reference model.

[0090] If the total number of current and historical monitoring cycles exceeds the preset number of cold-start convergence cycles, corresponding to the normal real-time monitoring phase after the steady-state background benchmark model is established, the product of the preset base step size and the dual-channel topology overlap is used as the dynamic step size control factor for the current monitoring cycle, realizing the selective dynamic dormancy of the adaptive filter. When external common-mode interference occurs, the dual-channel topology overlap approaches 1, and the adaptive filter normally tracks and filters out background noise; when real partial discharge occurs, the dual-channel topology overlap approaches 0, the product operation makes the step size instantly return to zero, and the weight update is immediately frozen to prevent the characteristics of real partial discharge from being mistakenly canceled by the algorithm as interference.

[0091] The preset cold start convergence cycle number represents the total number of partial discharge transient pulse triggers that force the adaptive filter to learn background noise features after the partial discharge monitoring gateway is first powered on or reset. The preset base step size represents the upper limit of the inherent learning rate of the adaptive filter when iterating and updating weights in the face of simple external common-mode interference.

[0092] In this embodiment, the historical monitoring period refers to all monitoring periods that have been completed before the current monitoring period after the partial discharge monitoring gateway is started. The total number of the current monitoring period and the historical monitoring periods is the total number of monitoring periods accumulated by the partial discharge monitoring gateway from the first monitoring period after its initial startup to the current monitoring period.

[0093] In one specific implementation of this invention, the preset number of cold start convergence cycles is set to 50. The preset base step size is set to an empirical value of 0.5.

[0094] Step S302: The adaptive weight vector is initialized to a zero vector. The normalized least mean square algorithm is used to process the sheath grounding data vector of the previous real-time sampling time adjacent to the current real-time sampling time, the adaptive weight vector of the previous real-time sampling time, the actual discharge voltage of the previous real-time sampling time, and the dynamic step size control factor to obtain the adaptive weight vector of the current real-time sampling time.

[0095] When the partial discharge monitoring gateway is first started, it has not yet acquired any environmental noise patterns. Therefore, the adaptive weight vector is initialized to zero, serving as the adaptive weight vector for the first real-time sampling moment of the first monitoring cycle, representing the starting point for the unbiased learning of the adaptive filter. During normal iteration, the normalized least mean square algorithm is used to process the sheath grounding data vector of the previous real-time sampling moment, the adaptive weight vector of the previous real-time sampling moment, the actual discharge voltage of the previous real-time sampling moment, and the dynamic step size control factor of the corresponding monitoring cycle to obtain the adaptive weight vector for each real-time sampling moment.

[0096] Taking any real-time sampling moment within the current monitoring period as the current real-time sampling moment, and using the current real-time sampling moment as an example, the adaptive weight vector of the previous real-time sampling moment adjacent to the current real-time sampling moment provides the historical reference state. The actual discharge voltage serves as the residual error, providing the deviation magnitude and direction of the weight correction. The sheath grounding data vector serves as the reference input, and together with the actual discharge voltage, determines the gradient compensation direction of the weights. The dynamic step size control factor determines the effective proportion of the gradient compensation amount superimposed on the historical weights. The combination of these four factors enables the adaptive filter to calculate the adaptive weight vector required for the current real-time sampling moment based on the calculation error and input state of the previous real-time sampling moment, along the gradient descent direction.

[0097] Specifically, the formula for calculating the adaptive weight vector for the current real-time sampling moment, based on the data from the previous real-time sampling moment adjacent to the current real-time sampling moment, is as follows: in, The adaptive weight vector representing the current real-time sampling moment; The adaptive weight vector representing the previous real-time sampling time adjacent to the current real-time sampling time; The dynamic step size control factor represents the current monitoring cycle; M represents the total number of data in the sheath grounding data vector of the previous real-time sampling time adjacent to the current real-time sampling time. The sheath grounding data vector represents the previous real-time sampling time adjacent to the current real-time sampling time; This represents the preset zero-bias constant (e.g., 10). -6 ); This represents the actual discharge voltage of the previous real-time sampling time adjacent to the current real-time sampling time.

[0098] It should be understood that the specific iterative formula and calculation derivation process of using the Normalized Least Mean Square (NLMS) algorithm for adaptive weight vector update are well-known techniques in this field, and will only be briefly introduced here without further details.

[0099] In this embodiment of the invention, during the initial cold start of the partial discharge monitoring gateway, the adaptive weight vector at the first real-time sampling moment of the first monitoring cycle is a zero vector. Following the method for calculating the actual discharge voltage in this embodiment, the sheath grounding data vector and the actual discharge voltage at this first real-time sampling moment are calculated and stored. Further, following the method in step S302, the adaptive weight vector at the first real-time sampling moment is updated using the dual-channel topology overlap of the first monitoring cycle to obtain the adaptive weight vector at the second real-time sampling moment. Following the method for calculating the actual discharge voltage in this embodiment, the sheath grounding data vector and the actual discharge voltage at the second real-time sampling moment are calculated and stored. Subsequent iterations are performed using the above iterative calculation method, updating the adaptive weight vector based on the adaptive weight vector of the previous real-time sampling moment adjacent to each real-time sampling moment to obtain the adaptive weight vector for each real-time sampling moment.

[0100] Specifically, for two adjacent monitoring periods, taking the current monitoring period as an example, for the first real-time sampling moment of the current monitoring period, the adaptive weight vector of the last real-time sampling moment of the previous monitoring period adjacent to the current monitoring period is updated using the dual-channel topological overlap of the current monitoring period, so as to obtain the adaptive weight vector of the first real-time sampling moment of the current monitoring period.

[0101] In summary, this invention analyzes the dual-channel topology overlap of the current monitoring period based on the temporal difference characteristics between any two real-time sampling moments within the current monitoring period, as well as the fluctuation characteristics of partial discharge voltage and sheath grounding current within the time interval between these two real-time sampling moments, combined with the partial discharge voltage and sheath grounding current of the buffered sampling moments in the current monitoring period. An adaptive weight vector is initialized, and the adaptive weight vector of the previous real-time sampling moment adjacent to each real-time sampling moment is updated using the dual-channel topology overlap of the current monitoring period, resulting in an adaptive weight vector for each real-time sampling moment. Based on the data characteristics of the sheath grounding current at each real-time sampling moment and historical sampling moments, combined with the adaptive weight vector and partial discharge voltage of each real-time sampling moment, the actual discharge voltage at each real-time sampling moment is determined.

Claims

1. A high-precision partial discharge monitoring method for high-voltage cables with signal denoising, characterized in that, The method includes: The partial discharge voltage and sheath grounding current are obtained at each sampling moment within the current monitoring period; the sampling moments include buffered sampling moments and real-time sampling moments; Based on the temporal difference characteristics between any two real-time sampling moments within the current monitoring period, as well as the fluctuation characteristics of partial discharge voltage and sheath grounding current within the time interval between the two real-time sampling moments, and combined with the partial discharge voltage and sheath grounding current at the buffer sampling moment, the dual-channel topology overlap degree of the current monitoring period is analyzed. An adaptive weight vector is initialized and set. The adaptive weight vector of the previous real-time sampling time adjacent to each real-time sampling time is updated using the dual-channel topology overlap to obtain the adaptive weight vector of each real-time sampling time. Based on the data characteristics of the sheath grounding current of each real-time sampling time and the historical sampling time, combined with the adaptive weight vector and partial discharge voltage of each real-time sampling time, the actual discharge voltage of each real-time sampling time is determined.

2. The high-precision partial discharge monitoring method for high-voltage cables with signal denoising according to claim 1, characterized in that, The method involves analyzing the dual-channel topology overlap of the current monitoring period based on the temporal difference characteristics between any two real-time sampling moments within the current monitoring period, as well as the fluctuation characteristics of the partial discharge voltage and sheath grounding current within the time interval between the two real-time sampling moments, combined with the partial discharge voltage and sheath grounding current at the buffer sampling moment. This includes: Based on the distribution range of partial discharge voltage at each buffer sampling time within the current monitoring cycle, the partial discharge voltage at each real-time sampling time is calculated to obtain the standard voltage value at each real-time sampling time. Based on the distribution range of sheath grounding current at the current monitoring cycle and the buffer sampling time, the sheath grounding current at each real-time sampling time is calculated to obtain the standard current value at each real-time sampling time. Based on the temporal difference characteristics between any two real-time sampling moments within the current monitoring period, and combined with the fluctuation characteristics of the standard voltage value within the time interval between any two real-time sampling moments, a partial discharge correlation matrix is ​​constructed. Based on the time sequence difference characteristics between any two real-time sampling moments within the current monitoring period, and combined with the fluctuation characteristics of the standard current value within the time interval between any two real-time sampling moments, a sheath grounding correlation matrix is ​​constructed. By integrating the partial discharge correlation matrix and the sheath grounding correlation matrix, the dual-channel topology overlap for the current monitoring period is obtained.

3. The high-precision partial discharge monitoring method for high-voltage cables with signal denoising according to claim 2, characterized in that, The step of calculating the partial discharge voltage at each real-time sampling moment based on the distribution range of the partial discharge voltage at the buffered sampling moment within the current monitoring period, and obtaining the standard voltage value at each real-time sampling moment, includes: The maximum value of the partial discharge voltage at all buffer sampling times within the current monitoring period is taken as the maximum voltage value. The minimum value of the partial discharge voltage at all buffer sampling times within the current monitoring period is taken as the minimum voltage value; The partial discharge voltage at each real-time sampling moment is normalized using the maximum and minimum voltage values ​​to obtain the standard voltage value at each real-time sampling moment.

4. The high-precision partial discharge monitoring method for high-voltage cables with signal denoising according to claim 2, characterized in that, The partial discharge correlation matrix is ​​constructed based on the temporal difference characteristics between any two real-time sampling moments within the current monitoring period, combined with the fluctuation characteristics of the standard voltage value within the time interval between any two real-time sampling moments. This includes: The real-time sampling time includes a sequence index; based on the sequence index of each real-time sampling time within the current monitoring period, a partial discharge correlation matrix is ​​constructed; the rows and columns of the partial discharge correlation matrix are both sequence indices of each real-time sampling time arranged in ascending numerical order. For any position in the partial discharge correlation matrix, if the absolute value of the difference between the row index and the column index of the position is equal to 1, then the element of the position is set to 1. If the absolute value of the difference between the row index and column index of the position is greater than the preset maximum visible threshold, or if the absolute value of the difference is equal to 0, then the element at the position is set to 0. If the absolute value of the difference between the row index and column index of the location is greater than 1, and the absolute value of the difference is less than or equal to the preset maximum visible threshold, then the element of the location is determined based on the fluctuation characteristics of the standard voltage value within the time interval between the two real-time sampling times corresponding to the row index and column index of the location.

5. The high-precision partial discharge monitoring method for high-voltage cables with signal denoising according to claim 4, characterized in that, The step of determining the element at the position based on the fluctuation characteristics of the standard voltage value within the time interval between the two real-time sampling times corresponding to the row and column indices of the position includes: The real-time sampling time corresponding to the minimum value between the row index and column index of the position is taken as the starting time, and the real-time sampling time corresponding to the maximum value between the row index and column index of the position is taken as the ending time. Linear fitting is performed on the standard voltage values ​​at the start and end times, and the result of the linear fitting is used as the global rate of change. Linear fitting is performed on the standard voltage value at any real-time sampling time within the time interval between the start time and the end time, and the result of the linear fitting is used as the local rate of change. If the absolute value of the difference between all local rates of change and the global rate of change in the time interval between the start time and the end time is less than or equal to a preset fluctuation tolerance threshold, then the element at the position is set to 1; otherwise, the element at the position is set to 0.

6. The high-precision partial discharge monitoring method for high-voltage cables with signal denoising according to claim 2, characterized in that, The fusion of the partial discharge correlation matrix and the sheath grounding correlation matrix to obtain the dual-channel topology overlap for the current monitoring period includes: The similarity between the partial discharge correlation matrix and the sheath grounding correlation matrix is ​​measured using the Dessian similarity coefficient algorithm, and the result is used as the dual-channel topology overlap.

7. The high-precision partial discharge monitoring method for high-voltage cables with signal denoising according to claim 1, characterized in that, The step of determining the actual discharge voltage at each real-time sampling moment based on the data characteristics of the sheath grounding current at each real-time sampling moment and historical sampling moments, combined with the adaptive weight vector and partial discharge voltage at each real-time sampling moment, includes: Take any real-time sampling moment within the current monitoring period as the current real-time sampling moment, and construct the sheath grounding data vector based on the sheath grounding current at each sampling moment within the preset backtracking window of the current real-time sampling moment; The interference estimation voltage is obtained by multiplying the sheath grounding data vector with the adaptive weight vector at the current real-time sampling time. The difference between the partial discharge voltage at the current real-time sampling moment and the estimated interference voltage is taken as the actual discharge voltage at the current real-time sampling moment.

8. The high-precision partial discharge monitoring method for high-voltage cables with signal denoising according to claim 7, characterized in that, The initialization setting of the adaptive weight vector involves updating the adaptive weight vector of the previous real-time sampling time adjacent to each real-time sampling time using the dual-channel topology overlap, to obtain the adaptive weight vector for each real-time sampling time, including: Based on the temporal characteristics of the current monitoring period and combined with the dual-channel topological overlap, the dynamic step size control factor of the current monitoring period is determined. The adaptive weight vector is initialized to a zero vector. The normalized least mean square algorithm is used to process the sheath grounding data vector of the previous real-time sampling time adjacent to the current real-time sampling time, the adaptive weight vector of the previous real-time sampling time, the actual discharge voltage of the previous real-time sampling time, and the dynamic step size control factor to obtain the adaptive weight vector of the current real-time sampling time.

9. A high-precision partial discharge monitoring method for high-voltage cables with signal denoising according to claim 8, characterized in that, The determination of the dynamic step size control factor for the current monitoring period based on the time-series characteristics of the current monitoring period and the dual-channel topological overlap includes: If the total number of current monitoring cycles and historical monitoring cycles is less than or equal to the preset number of cold start convergence cycles, then the preset base step size is used as the dynamic step size control factor for the current monitoring cycle; if the total number of current monitoring cycles and historical monitoring cycles is greater than the preset number of cold start convergence cycles, then the product of the preset base step size and the dual-channel topology overlap is used as the dynamic step size control factor for the current monitoring cycle.

10. A high-precision partial discharge monitoring method for high-voltage cables with signal denoising according to claim 1, characterized in that, The method for obtaining the monitoring period is as follows: Any sampling time is taken as a candidate time. If the partial discharge voltage of the candidate time is greater than the preset hardware trigger threshold, then the candidate time is taken as the transient event trigger time. The start time of the cycle is obtained by subtracting the trigger time of the transient event from the preset backtracking time; the sampling time within the time interval between the start time of the cycle and the trigger time of the transient event is used as the cache sampling time. The transient event trigger time is added to the preset duration to obtain the cycle end time; the sampling time within the time interval between the transient event trigger time and the cycle end time is used as the real-time sampling time. The time interval between the start time and the end time of the cycle is taken as the monitoring cycle.

Images