Cable partial discharge online monitoring system and method based on zero sequence current and edge calculation

Abstract

A 10kV cable partial discharge on-line monitoring system and method based on zero sequence current and edge calculation comprises a broadband zero sequence mutual inductor and a grounding down lead, the broadband zero sequence mutual inductor is sleeved on a three-core cable, one end of the grounding down lead is arranged on a sheath of the three-core cable, the other end of the grounding down lead is led out downwards through the high frequency zero sequence mutual inductor to be grounded, and the grounding down lead is connected with the broadband zero sequence mutual inductor. A secondary coil of the broadband zero sequence mutual inductor is connected to an edge computing device through a coaxial cable, and the edge computing device sends a partial discharge diagnosis result to a monitoring platform through a wireless communication module or in a wired mode. The system has the beneficial effects that the sensing path is high in anti-interference capability, the signal fidelity is high, the system has local intelligent diagnosis capability, and the communication and installation cost can be remarkably reduced.

Description

Technical Field

[0001] This invention relates to an apparatus for testing electrical performance, and more particularly to an online monitoring system and method for partial discharge of 10kV cables based on zero-sequence current and edge computing. Background Technology

[0002] Partial discharge (PD) detection in cables is a key technology for assessing insulation health and enabling early warning of faults. With a large number of lines, complex operating conditions, and high costs associated with power outages for maintenance, online, sensitive, and interference-resistant PD monitoring methods are particularly important.

[0003] Correspondingly, online electrical testing methods focus on broadband pulse pickup. A typical approach involves using a high-frequency current transformer (HFCT) to pick up high-frequency pulse currents on the cable sheath or grounding circuit—a method commonly referred to in engineering as grounding measurement. This type of method can capture high-frequency partial discharge pulses and perform trend diagnosis under uninterrupted power supply conditions, making it one of the most common and mature methods for online electrical testing of distribution and transmission cables. Power frequency zero-sequence (leakage current) measurement is also a widely used method in distribution networks. It reflects power frequency grounding imbalance and overall insulation anomalies, falling under the categories of grounding protection and power frequency condition monitoring.

[0004] The mechanism of the ground wire measurement method is as follows: when partial discharge occurs, the rapid charge migration inside the insulation forms a broadband high-frequency pulse current. This pulse propagates along the cable shield, sheath, and grounding loop, and is sensed by sensors such as HFCTs at the grounding down conductor or grounding strip, thus achieving online detection. The advantages of this method are: broadband response capability, the measurement point being at ground potential, high safety, and the sensors mostly being open-type clamps, requiring minimal modification to the cable body, facilitating long-term online monitoring. However, its key limitation stems primarily from its sensitivity to frequency band and channel characteristics. Since the signal acquisition point is located in the grounding loop, the ground wire is often in a region of strong electromagnetic noise convergence. High-frequency interference from inverters, switching operations, power electronic devices, and communication radiation can superimpose with the partial discharge pulse, resulting in a low signal-to-noise ratio and leading to misjudgments or missed detections.

[0005] The basic premise of power frequency zero-sequence measurement is to monitor changes in 50 Hz zero-sequence current to reflect three-phase-to-ground imbalance, increased leakage to ground, or grounding anomalies. It has the advantages of low cost and easy deployment in judging overall insulation degradation and grounding fault trends. However, when used for partial discharge monitoring, it has the following problems: Partial discharge signals are high-frequency transient pulses in the ns–µs range, while the power frequency zero-sequence channel is only effective for zero-sequence current components based on 50 Hz. It is almost impossible to directly respond to partial discharge pulses, and it is completely unable to capture partial discharge transient pulses in the nanosecond to microsecond range. It lacks the basic ability to identify the onset, development, and type of partial discharge. Therefore, it cannot present the onset and development process of partial discharge, nor can it provide the information necessary for diagnosis such as location, PRPD mode, or pulse morphology.

[0006] The typical approach for online monitoring of 10kV partial discharge is as follows: a grounding lead sensor transmits the signal to a remote centralized acquisition and processing unit via a long signal cable, and then the information is sent to a diagnostic platform via fiber optic cable. The drawbacks of this method are: signal attenuation and distortion over long distances, and the high communication costs and system complexity resulting from uploading massive amounts of raw waveforms, severely restricting large-scale deployment. Sending waveforms from multiple sensors at high speed to the diagnostic platform increases communication costs and reduces practicality. Furthermore, the high cabling costs also increase the difficulty of technology promotion. Summary of the Invention

[0007] The purpose of this invention is to solve the above-mentioned problems in the background technology and provide a 10kV cable partial discharge online monitoring system and method based on zero-sequence current and edge computing, which has strong anti-interference capability of sensing path, high signal fidelity, local intelligent diagnostic capability, and can significantly reduce communication and installation costs.

[0008] The technical solution of this invention is: an online monitoring system for partial discharge of 10kV cables based on zero-sequence current and edge computing, characterized in that it includes a broadband zero-sequence current transformer and a grounding lead-out wire. The broadband zero-sequence current transformer is sleeved on a three-core cable. One end of the grounding lead-out wire is set on the sheath of the three-core cable, and the other end is led down to ground through a high-frequency zero-sequence current transformer. The secondary coil of the broadband zero-sequence current transformer is connected to an edge computing device through a coaxial cable. The edge computing device transmits the diagnostic results of partial discharge to the monitoring platform through a wireless communication module or a wired connection.

[0009] Furthermore, the wideband zero-sequence current transformer adopts a dual-winding composite structure. The first winding is a high-frequency sensing winding with an equivalent inductance parameter of 1-50 μH and a distributed capacitance parameter of 5-30 pF. It is used to respond to the high-frequency transient pulse current signal generated during partial discharge, and the corresponding spectral energy is mainly distributed in the range of 100 kHz-5 MHz. The second winding is a low-frequency energy winding with an equivalent inductance parameter of 1-20 mH, which makes it highly sensitive to discharge pulse current in the frequency band of 10 kHz-300 kHz. It mainly responds to current signals with frequencies in the range of 10 kHz-300 kHz. The two windings are independently distributed on the magnetic core, and their output signals are electrically isolated from each other.

[0010] Furthermore, the edge computing device includes a signal conditioning circuit, an AD sampling circuit, and a Rootspe. The signal conditioning circuit includes a pre-buffer circuit composed of a high-speed operational amplifier and a differential amplifier, used to preprocess the weak differential signal output from the dual windings of the wideband zero-sequence current transformer. The high-speed operational amplifier is connected to the output of the wideband zero-sequence current transformer in a voltage follower manner to achieve high-speed buffering and impedance matching. The differential amplifier amplifies and suppresses common-mode signals from the high-frequency probe winding and the low-frequency energy winding. The signal after conditioning is sent to the AD sampling circuit for synchronous sampling. The Rootspe integrates a multi-core processor, random access memory (RAM), and internal non-volatile memory. The sampled digital signal is written to the RAM via DMA. The internal non-volatile memory is used for storing programs, model parameters, and historical data. The multi-core processor reads the sampled data from the RAM and performs subsequent signal analysis and intelligent diagnostic processing.

[0011] Furthermore, the multi-core processor reads the raw waveform data from the random access memory, performs time-domain and frequency-domain analysis on it, extracts the pulse peak value, pulse width, and spectral features, and combines the pulse peak value, pulse width, and spectral features to form a multi-dimensional feature vector. The multi-dimensional feature vector is input into a pre-trained lightweight machine learning model for inference analysis. During the above processing, the multi-core processor associates the raw waveform segment that triggers the detection condition, the corresponding feature extraction results, and the model inference output results, and stores them in the non-volatile memory in the form of structured data for subsequent querying, trend analysis, or uploading to the upper-level monitoring platform via the communication module.

[0012] Furthermore, a secondary-side protection and overcurrent open-circuit detection circuit is connected to the signal conditioning circuit at the secondary-side output of the wideband zero-sequence current transformer. This secondary-side protection and overcurrent open-circuit detection circuit includes: a current transformer T2; a load and bleeder resistor R4; an input filter network composed of capacitors R2, C1, and C2; a comparator U1 and its reference voltage network composed of resistor R3 and capacitor C3; a reference voltage generation circuit composed of operational amplifier U2A, resistors R6-R8, and capacitor C4; a PWM signal filtering and buffer circuit composed of operational amplifier U2B, resistor R5, and capacitor C5; and an output pull-up resistor R1. Operational amplifier U2A employs a non-inverting amplification structure, and operational amplifier U2B employs a voltage follower structure. The output terminal OC.DET is an open-collector logic signal, facilitating interface with a microcontroller or digital control system.

[0013] A monitoring method for a 10kV cable partial discharge online monitoring system based on zero-sequence current and edge calculation as described above, characterized in that it includes: S1. The system continuously monitors the dual-channel digital signal input to the wideband zero-sequence current transformer in real time, and performs statistical analysis on the background noise level within the sliding time window to obtain the noise characteristic parameters under the current operating conditions. S2. Dynamically adjust the pulse detection threshold according to the noise characteristic parameters so that the detection threshold is adaptively updated as the environmental noise changes; S3. When the amplitude of the input signal exceeds the detection threshold, it is determined to be a suspicious discharge pulse event; S4. Segment and extract suspicious discharge pulse events, perform time-domain and frequency-domain analysis on each segmented discharge pulse event, extract pulse peak value, pulse width and spectral features, and combine the pulse peak value, pulse width and spectral features to form a multi-dimensional feature vector; S5. The extracted multidimensional feature vectors are mapped and input into a lightweight deep learning model for identification and analysis. The output discharge type discrimination result and corresponding confidence level are then calculated, and the number of peak events, peak density, and instantaneous characteristic indicators of amplitude distribution within the time window are statistically analyzed. When the identification result is suspected partial discharge and the confidence level reaches the preset threshold, and the medium- and long-term statistical results formed by the sliding time window indicate a continuous active trend, the system triggers an alarm and records relevant data. Otherwise, the system performs resampling based on the identification result or simply records the event without triggering an alarm.

[0014] Furthermore, the preset threshold is determined by the partial discharge pulse localization pattern learned by the lightweight deep learning model during training, and the confidence level is a probability value output by the lightweight deep learning model through comparison with the preset threshold and linkage with medium- and long-term statistical trends.

[0015] Furthermore, the lightweight deep learning model adopts a hybrid architecture of one-dimensional convolutional neural network and recurrent neural network, and is trained in the cloud or laboratory environment using a large number of known partial discharge and interference samples. Its model parameters are stored in the non-volatile memory and loaded into random access memory for computation when the device is running.

[0016] Furthermore, by analyzing the variation range of peak numbers per unit time, event interval characteristics, and recent alarm history, a partial discharge activity index is constructed to describe the intensity of partial discharge activity in the field. The system continuously monitors this index and adaptively adjusts its operating status based on its dynamic changes; when Below the set low threshold When this occurs, the system enters a low duty cycle sleep mode, reducing the sampling frequency; when Exceeding the upper threshold During this period, the system remains awake and enters a high-density monitoring mode, increasing the sampling frequency. This mechanism significantly reduces the system's average power consumption while ensuring timely detection of abnormal events, thereby supporting long-term, maintenance-free operation of the equipment under typical working conditions.

[0017] Furthermore, the partial discharge activity index is defined as: in, Indicates time The localized activity index; The peak number of events detected per unit of time; The rate of change of the number of peak values ​​over a continuous time window; This refers to the historical alarm weighting factor, used to reflect whether there have been persistent or severe partial discharge events recently; wherein the weighting coefficient... , , The configurable parameter is used to balance the influence of different characteristic quantities on the partial discharge activity index. Its value is determined based on the on-site operating conditions, historical data statistics, and monitoring targets.

[0018] The beneficial effects of this invention are: 1. This invention directly mounts a wideband zero-sequence current transformer onto the outer surface of a three-core cable. Utilizing the characteristic that the vector sum of the three-phase currents is zero, it outputs a near-zero signal under normal operating conditions. In the event of partial discharge, the zero-sequence current of the three-core cable is directly coupled, and its high-frequency zero-sequence component can be captured with high sensitivity. This path naturally suppresses grounding loops and external electromagnetic interference, significantly improving the purity and reliability of the detection.

[0019] 2. To address the characteristics of partial discharge pulses, which possess both nanosecond-level leading edges (MHz band) and millisecond-level energy cores (kHz band), the sensor employs a composite structure of a high-frequency detection winding and a low-frequency energy winding. The two windings are physically isolated and electrically independent on the magnetic core, providing complete and faithful original signals for subsequent multi-dimensional feature fusion analysis, thus achieving lossless extraction of full-band partial discharge features.

[0020] 3. The system deploys a deeply optimized lightweight deep learning model on an embedded platform, which can directly perform end-to-end inference on the original waveform and output the probability, confidence level and statistical characteristics of the discharge type. All analysis is completed on the device side, forming an edge intelligence closed loop. It does not rely on the cloud, has a short response latency, and only uploads structured alarm information, which greatly reduces the communication bandwidth requirements and the risk of data leakage. Attached Figure Description

[0021] Figure 1 This is a system overall structure diagram of the present invention; Figure 2This is a schematic diagram of a broadband zero-sequence current transformer. Figure 3 This is a schematic diagram of the magnetic circuit of a broadband zero-sequence current transformer; Figure 4 This is a circuit block diagram of an edge computing device; Figure 5 This is the circuit schematic of the pre-buffer circuit in the signal conditioning circuit; Figure 6 This is the schematic diagram of a differential amplifier in a signal conditioning circuit; Figure 7 This is the principle of secondary side protection and overcurrent open circuit detection circuit; Figure 8 This is a diagram showing the frequency response characteristics of the transformer output voltage under constant current excitation conditions. Figure 9 This is the control flowchart of the present invention. Detailed Implementation

[0022] To more clearly illustrate the technical solution of the present invention, the present invention will be described in detail below with reference to the embodiments and accompanying drawings.

[0023] like Figure 1 As shown, a 10kV cable partial discharge online monitoring system based on zero-sequence current and edge computing includes a broadband zero-sequence current transformer 2 and a grounding lead-down wire 3. The broadband zero-sequence current transformer 2 is sleeved on a three-core cable 1. One end of the grounding lead-down wire 3 is set on the sheath of the three-core cable 1, and the other end is led down to ground through the high-frequency zero-sequence current transformer 2. The secondary coil of the broadband zero-sequence current transformer 2 is connected to an edge computing device 5 through a coaxial cable 4. The edge computing device 5 sends the partial discharge diagnosis results to the monitoring platform 9 through a wireless communication module 7 or a wired cable 8.

[0024] like Figure 2As shown, the wideband zero-sequence current transformer (transformer module) 2 adopts a dual-winding composite structure. The first winding is a high-frequency detection winding 201, which is used to respond to the high-frequency transient pulse current signal generated during partial discharge. Its equivalent inductance parameter is configured as 1-50 μH, and its distributed capacitance parameter is configured as 5-30 pF. The corresponding spectral energy is mainly distributed in the range of 100 kHz-5 MHz, which is used to sensitively detect the rising edge of the discharge pulse and its accompanying high-frequency oscillation component. The second winding is a low-frequency energy winding 202, which is configured as 1-20 mH. It is used to couple the low-frequency zero-sequence current component generated during partial discharge to characterize the main energy characteristics of the discharge pulse. The low-frequency energy winding mainly responds to current signals with frequencies in the range of 10 kHz-300 kHz, which is suitable for quantitatively assessing the severity of partial discharge. The two windings are independently distributed on the magnetic core, and their output signals are electrically isolated from each other, thereby avoiding signal coupling and preventing functional overlap.

[0025] The wideband zero-sequence current transformer 2 is made of ultra-microcrystalline (or nanocrystalline) material.

[0026] In one embodiment of the present invention, such as Figure 3 As shown, the magnetic circuit structure of the open-type broadband zero-sequence current transformer adopts a symmetrical double-C-shaped splicing structure to form a magnetic circuit with high closure.

[0027] like Figure 4 , Figure 5 and Figure 6As shown, the edge computing device 5 includes a signal conditioning circuit, an AD sampling circuit, and a Roots Pen. The AD sampling circuit is model AD9226. The Roots Pen integrates a multi-core processor, random access memory, and non-volatile memory. The signal conditioning circuit includes a pre-buffer circuit and a differential amplifier composed of a high-speed operational amplifier and surrounding components. It is used to preprocess the weak differential signal output from the dual windings (transformer module) of the broadband zero-sequence current transformer. The high-speed operational amplifier is model OPA695IDR and operates in voltage follower mode. The input is from an expansion board. The BNC interface connects to the output of the wideband zero-sequence current transformer, outputting the SIGN signal to a differential amplifier for high-speed buffering and impedance matching. Its bandwidth meets the transmission requirements of the high-frequency detection winding signal within the 100kHz-5MHz range. The differential amplifier, model AD8138ARZ, is connected to the SIGN output of the pre-buffer circuit, outputting differential signals INA and INB to the input of the AD sampling circuit. This amplifies and suppresses common-mode signals from the high-frequency detection winding and the low-frequency energy winding, thereby improving the signal's anti-interference capability. The signal after conditioning is sent to the AD sampling circuit for synchronous sampling. The sampled digital signal is written to a random access memory (RAM) via DMA to achieve high-speed caching of the partial discharge pulse waveform. This non-volatile memory is used for storing programs, model parameters, and historical data. The multi-core processor reads the sampled data from the RAM to perform subsequent signal analysis and intelligent diagnostic processing.

[0028] The multi-core processor reads raw waveform data from random access memory, performs time-domain and frequency-domain analysis on it, extracts multi-dimensional feature parameters such as pulse peak value, pulse width, and spectral characteristics, and combines these multi-dimensional feature parameters to form a multi-dimensional feature vector. The multi-dimensional feature vector is then input into a pre-trained lightweight machine learning model for inference analysis. During the above processing, the multi-core processor associates the raw waveform segment that triggers the detection condition, the corresponding feature extraction results, and the model inference output results, and stores them in the non-volatile memory in the form of structured data for subsequent querying, trend analysis, or uploading to the upper-level monitoring platform via the communication module.

[0029] like Figure 7As shown, a secondary-side protection and overcurrent open-circuit detection circuit is connected to the signal conditioning circuit at the secondary-side output of the wideband zero-sequence current transformer. The secondary-side protection and overcurrent open-circuit detection circuit includes: a current transformer T2; a load and bleeder resistor R4; an input filter network composed of capacitors R2, C1, and C2; a comparator U1 and its reference voltage network composed of resistor R3 and capacitor C3; a reference voltage generation circuit composed of operational amplifier U2A, resistors R6-R8, and capacitor C4; a PWM signal filtering and buffer circuit composed of operational amplifier U2B, resistor R5, and capacitor C5; and an output pull-up resistor R1. Operational amplifier U2A uses a non-inverting amplification structure, and operational amplifier U2B uses a voltage follower structure. The output terminal OC.DET is an open-collector logic signal, facilitating interface with a microcontroller or digital control system. This transformer secondary side protection and overcurrent open-circuit detection circuit uses current transformer T2 as the signal source. Resistor R4 serves as the secondary side load and bleeder resistor to prevent open circuit and limit induced overvoltage. The sampled signal is then filtered by a low-pass filter network composed of resistor R2 and capacitors C1 and C2 to suppress spikes and noise before being sent to the non-inverting input of comparator U1. Simultaneously, the PWM signal is filtered by resistor R5 and capacitor C5 and buffered by operational amplifier U2B. Operational amplifier U2A, in conjunction with resistors R6, R7, R8 and capacitor C4, generates a stable reference voltage VREF, which is further filtered by resistor R3 and capacitor C3 and sent to the inverting input of comparator U1 to achieve threshold comparison for abnormal current on the secondary side of the transformer. When an overcurrent or abnormal state is detected, the output of comparator U1 is pulled up to +5V by resistor R1 to form an OC.DET logic signal, which facilitates interface with the MCU / control system. This achieves a low-cost, high-reliability solution integrating secondary side protection, filtering and conditioning, threshold judgment, and digital alarm output.

[0030] In this embodiment, the inner diameter of the broadband zero-sequence current transformer core , outer diameter ,high The cross-sectional area of ​​the magnetic core can be calculated from this. This dimension optimizes the effective length of the magnetic circuit while accommodating standard grounding wires and providing sufficient electrical clearance. The ratio of the cross-sectional area to the total number of turns in the secondary coil is used to achieve an optimal balance between sensitivity and volume. This value, verified through simulation, can provide a sufficient and flat voltage output at the key characteristic frequency points of surface discharge.

[0031] The high-frequency detection winding uses approximately 200 turns of fine-diameter enameled wire to reduce distributed capacitance, thereby raising its self-resonant frequency to above 5MHz. This is specifically designed to capture high-frequency oscillations at nanosecond-level leading edges. The low-frequency energy winding uses 600 turns of thick-diameter enameled wire to achieve a higher mutual inductance coefficient. It is primarily used to couple the main current pulses characterizing discharge energy in the 10kHz to 300kHz frequency band. The two windings have independent outputs, enabling targeted signal conditioning at the back end.

[0032] Testing and verification of key performance parameters: The core performance of the sensor is characterized by measured curves, such as... Figure 8 The sensor's transmission impedance frequency response characteristic diagram shows that the effective response frequency band of the low-frequency energy channel has an upper limit of approximately 119kHz and a lower cutoff frequency of approximately 7kHz, which can effectively cover the main frequency range of the surface discharge signal and meet the wideband response requirements. This bandwidth range not only covers the frequency range where energy is most concentrated during surface discharge, but also has a certain degree of frequency domain redundancy, which is beneficial for subsequent spectral feature extraction and abnormal signal identification.

[0033] Transmission impedance It is a key parameter for measuring the efficiency of a sensor in converting primary-side current into secondary-side voltage, defined as The measured transmission impedance curve is in The sensor maintains high stability within its core frequency band. Within this band, the sensor's transfer function for different frequency components remains consistent, and the output signal accurately reflects the spectral proportions of the input current, providing a reliable data foundation for subsequent intelligent diagnostic algorithms based on spectral characteristics.

[0034] A method for online monitoring of partial discharge in 10kV cables based on zero-sequence current and edge computing, such as... Figure 9 As shown, it includes: S1. The system continuously monitors the dual-channel digital signal input to the wideband zero-sequence current transformer in real time, and performs statistical analysis on the background noise level within the sliding time window to obtain the noise characteristic parameters under the current operating conditions. S2. Dynamically adjust the pulse detection threshold according to the noise characteristic parameters so that the detection threshold is adaptively updated as the environmental noise changes; S3. When the amplitude of the input signal exceeds the detection threshold, it is determined to be a suspicious discharge pulse event; S4. Segment and extract suspicious discharge pulse events. Perform multi-dimensional feature information extraction on each segmented discharge pulse event, including time domain analysis and frequency domain analysis, extract multi-dimensional feature parameters including pulse peak value, pulse width distribution and spectral features, and combine the multi-dimensional feature parameters to form a multi-dimensional feature vector. S5. The extracted multidimensional feature vectors are mapped and input into a lightweight deep learning model for identification and analysis. The discharge type discrimination result and corresponding confidence level are output, and the instantaneous characteristic indicators of peak event quantity, peak density and amplitude distribution within the time window are statistically analyzed. When the identification result is suspected partial discharge and the confidence level reaches the preset threshold, and the medium- and long-term statistical results formed by the sliding time window indicate a continuous active trend, the system triggers an alarm and records relevant data. Otherwise, the system performs resampling judgment based on the identification result or only records the event without triggering an alarm.

[0035] Furthermore, the preset threshold is determined by the partial discharge pulse localization pattern learned by the lightweight deep learning model during training, and the confidence level is a probability value output by the lightweight deep learning model through comparison with the preset threshold and linkage with medium- and long-term statistical trends.

[0036] Furthermore, the lightweight deep learning model employs a hybrid architecture of one-dimensional convolutional neural networks and recurrent neural networks. It is trained using a large number of known partial discharge and interference samples in a cloud or laboratory environment. The model parameters are stored in the non-volatile memory and loaded into random access memory for computation during device operation. Furthermore, focusing on real-time statistics of partial discharge pulse peaks, a partial discharge activity index is constructed to describe the intensity of on-site partial discharge activity by analyzing the variation amplitude of peak numbers per unit time, event interval characteristics, and recent alarm history. The edge processing device continuously monitors this index and adaptively adjusts the system's operating status based on its dynamic changes; when Below the set low threshold When this occurs, the system enters a low duty cycle sleep mode, reducing the sampling frequency; when Exceeding the upper threshold During this period, the system remains awake and enters a high-density monitoring mode, increasing the sampling frequency. This mechanism significantly reduces the system's average power consumption while ensuring timely detection of abnormal events, thereby supporting long-term, maintenance-free operation of the equipment under typical working conditions.

[0037] Furthermore, the partial discharge activity index is defined as: in, Indicates time The localized activity index; The peak number of events detected per unit of time; The rate of change of the number of peak values ​​over a continuous time window; This refers to the historical alarm weighting factor, used to reflect whether there have been persistent or severe partial discharge events recently; wherein the weighting coefficient... , , The configurable parameter is used to balance the influence of different characteristic quantities on the partial discharge activity index. Its value is determined based on the on-site operating conditions, historical data statistics, and monitoring targets.

[0038] During the verification process, a broadband zero-sequence current transformer was installed outside a three-core cable, simultaneously coupling the zero-sequence current components of the three-phase conductors, and connected to an edge computing device via a coaxial cable. The system continuously samples the acquired zero-sequence current signal according to the described monitoring method, and performs signal conditioning, feature extraction, and intelligent analysis processing at the device end. Verification results show that the system described in this invention can achieve stable monitoring of cable zero-sequence discharge pulses under uninterrupted power conditions. Furthermore, the edge computing-based monitoring method can identify and judge discharge events at the device end, reducing the need for raw data transmission and enabling online monitoring and trend analysis of partial discharge states. The accuracy reached 90%, the recall rate reached 83%, the average precision reached 98.2%, and the inference time was 0.49 seconds, demonstrating significant performance.

[0039] The above are merely specific embodiments of the present invention and are not intended to limit the present invention. For those skilled in the art, the present invention can have various modifications and variations. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A 10kV cable partial discharge online monitoring system based on zero-sequence current and edge computing, characterized in that, The device includes a broadband zero-sequence current transformer and a grounding lead. The broadband zero-sequence current transformer is sleeved on a three-core cable. One end of the grounding lead is set on the sheath of the three-core cable, and the other end is led down to ground through a high-frequency zero-sequence current transformer. The secondary coil of the broadband zero-sequence current transformer is connected to an edge computing device through a coaxial cable. The edge computing device transmits the diagnostic results of partial discharge to the monitoring platform through a wireless communication module or a wired connection.

2. The 10kV cable partial discharge online monitoring system based on zero-sequence current and edge computing as described in claim 1, characterized in that, The wideband zero-sequence current transformer adopts a dual-winding composite structure. The first winding is a high-frequency sensing winding with an equivalent inductance parameter of 1-50μH and a distributed capacitance parameter of 5-30pF. It is used to respond to the high-frequency transient pulse current signal generated during partial discharge, and the corresponding spectral energy is mainly distributed in the range of 100kHz-5MHz. The second winding is a low-frequency energy winding with an equivalent inductance parameter of 1-20mH. It mainly responds to current signals with frequencies in the range of 10kHz-300kHz. The two windings are independently distributed on the magnetic core, and their output signals are electrically isolated from each other.

3. The 10kV cable partial discharge online monitoring system based on zero-sequence current and edge computing according to claim 1, characterized in that, The edge computing device includes a signal conditioning circuit, an AD sampling circuit, and a Rootspe. The signal conditioning circuit includes a pre-buffer circuit composed of a high-speed operational amplifier and a differential amplifier, used to preprocess the weak differential signal output from the dual windings of a wideband zero-sequence current transformer. The high-speed operational amplifier is connected to the output terminal of the wideband zero-sequence current transformer in a voltage follower manner to achieve high-speed buffering and impedance matching. The differential amplifier is used to amplify and suppress common-mode signals from the high-frequency probe winding and the low-frequency energy winding. The signal after conditioning is sent to the AD sampling circuit for synchronous sampling. The Rootspe integrates a multi-core processor, random access memory (RAM), and internal non-volatile memory. The sampled digital signal is written to the RAM via DMA. The internal non-volatile memory is used to store programs, model parameters, and historical data. The multi-core processor reads the sampled data from the RAM and performs subsequent signal analysis and intelligent diagnostic processing.

4. The 10kV cable partial discharge online monitoring system based on zero-sequence current and edge computing according to claim 3, characterized in that, The multi-core processor reads the raw waveform data from the random access memory, performs time-domain and frequency-domain analysis on it, extracts the pulse peak value, pulse width and spectral features, and combines the pulse peak value, pulse width and spectral features to form a multi-dimensional feature vector; The multidimensional feature vectors are input into a pre-trained lightweight machine learning model for inference analysis. During the above processing, the multi-core processor associates the original waveform segments that trigger the detection conditions, the corresponding feature extraction results, and the model inference output results, and stores them in the non-volatile memory in the form of structured data for subsequent querying, trend analysis, or uploading to the upper monitoring platform via the communication module.

5. The 10kV cable partial discharge online monitoring system based on zero-sequence current and edge computing according to claim 3, characterized in that, A secondary-side protection and overcurrent open-circuit detection circuit is connected to the signal conditioning circuit at the secondary-side output of the wideband zero-sequence current transformer. The secondary-side protection and overcurrent open-circuit detection circuit includes: a current transformer T2; a load and a bleeder resistor R4; an input filter network composed of capacitors R2 and C1 and C2; a comparator U1 and its reference voltage network composed of resistor R3 and capacitor C3; a reference voltage generation circuit composed of operational amplifier U2A, resistors R6 to R8 and capacitor C4; a PWM signal filtering and buffer circuit composed of operational amplifier U2B, resistor R5 and capacitor C5; and an output pull-up resistor R1. The operational amplifier U2A adopts a non-inverting amplification structure, and the operational amplifier U2B adopts a voltage follower structure. The output terminal OC.DET is an open-collector logic signal, which is convenient for interfacing with a microcontroller or digital control system.

6. A monitoring method for a 10kV cable partial discharge online monitoring system based on zero-sequence current and edge calculation as described in any one of claims 1-5, characterized in that, include: S1. The system continuously monitors the dual-channel digital signal input to the wideband zero-sequence current transformer in real time, and performs statistical analysis on the background noise level within the sliding time window to obtain the noise characteristic parameters under the current operating conditions. S2. Dynamically adjust the pulse detection threshold according to the noise characteristic parameters so that the detection threshold is adaptively updated as the environmental noise changes; S3. When the amplitude of the input signal exceeds the detection threshold, it is determined to be a suspicious discharge pulse event; S4. Segment and extract suspicious discharge pulse events, perform multi-time domain analysis and frequency domain analysis on each segmented discharge pulse event, extract pulse peak value, pulse width and spectral features, and combine the pulse peak value, pulse width and spectral features to form a multi-dimensional feature vector; S5. The extracted multidimensional feature vectors are mapped and input into a lightweight deep learning model for identification and analysis. The discharge type discrimination result and corresponding confidence level are output, and the number of peak events, peak density and amplitude distribution instantaneous characteristic indicators within the time window are statistically analyzed. When the identification result is suspected partial discharge and the confidence level reaches the preset threshold, and the medium and long-term statistical results formed by the sliding time window indicate a continuous active trend, the system triggers an alarm and records relevant data. Otherwise, resampling will be performed based on the identification results, or only an event will be recorded without triggering an alarm.

7. The online monitoring method for partial discharge of 10kV cables based on zero-sequence current and edge calculation according to claim 6, characterized in that, In step S5, the preset threshold is determined by the partial discharge pulse localization pattern learned by the lightweight deep learning model during training, and the confidence level is a probability value output by the lightweight deep learning model through comparison with the preset threshold and linkage with the medium- and long-term statistical trends.

8. The online monitoring method for partial discharge of 10kV cables based on zero-sequence current and edge calculation according to claim 6, characterized in that, The lightweight deep learning model adopts a hybrid architecture of one-dimensional convolutional neural network and recurrent neural network. It is trained in the cloud or laboratory environment using a large number of known partial discharge and interference samples. Its model parameters are stored in the non-volatile memory and loaded into random access memory for computation when the device is running.

9. The monitoring method of the 10kV cable partial discharge online monitoring system based on zero-sequence current and edge calculation as described in claim 6, characterized in that, In step S5, by analyzing the variation of peak quantity per unit time, event interval characteristics, and recent alarm history, a partial discharge activity index is constructed to describe the intensity of partial discharge activity on site. The system continuously monitors this index and adaptively adjusts its operating status based on its dynamic changes; when Below the set low threshold When this occurs, the system enters a low duty cycle sleep mode, reducing the sampling frequency; when Exceeding the upper threshold At this time, the system remains awake and enters a high-density monitoring mode to increase the sampling frequency.

10. The online monitoring method for partial discharge of 10kV cables based on zero-sequence current and edge computing according to claim 9, characterized in that, The partial discharge activity index is defined as follows: in, Indicates time The localized activity index; The peak number of events detected per unit of time; The rate of change of the number of peak values ​​over a continuous time window; This refers to the historical alarm weighting factor, used to reflect whether there have been persistent or severe partial discharge events recently; wherein the weighting coefficient... , , The configurable parameter is used to balance the influence of different characteristic quantities on the partial discharge activity index. Its value is determined based on the on-site operating conditions, historical data statistics, and monitoring targets.

Images