Real-time Synchronous Acquisition Method of Multi-channel Partial Discharge Signals Based on Edge Computing

Abstract

This invention relates to the fields of edge computing applications and partial discharge detection technology, specifically a real-time synchronous acquisition method for multi-channel partial discharge signals based on edge computing. This invention integrates multi-channel signal acquisition, time synchronization, feature analysis, and hierarchical storage functions at edge nodes to achieve asynchronous acquisition, unified timestamp marking, and synchronous data fusion of multiple types of partial discharge signals. It also performs local extraction of multi-dimensional feature parameters such as time domain, frequency domain, and energy, as well as intelligent event discrimination. For abnormal signals, the system uses encryption and compression processing for categorized storage and remote uploading; for normal signals, dynamic management is achieved through cyclic buffering. This method significantly improves the accuracy and efficiency of multi-channel signal synchronous acquisition, enhances data security and system adaptability, and is suitable for real-time online monitoring and intelligent diagnosis of partial discharge in power equipment.

Description

Technical Field

[0001] This invention relates to the fields of edge computing applications and partial discharge detection technology, specifically a method for real-time synchronous acquisition of multi-channel partial discharge signals based on edge computing. Background Technology

[0002] During long-term operation, high-voltage power equipment such as transformers and GIS systems are highly susceptible to partial discharge (PD) due to insulation system deterioration and potential defects. PD is an electrical breakdown occurring in a localized area of ​​the insulation, not immediately leading to overall insulation failure. However, persistent partial discharge can seriously threaten equipment safety and shorten its lifespan. Therefore, online, real-time, and high-precision monitoring of partial discharge signals is crucial for ensuring the safe operation of power systems. Existing online partial discharge monitoring solutions often employ multiple types of sensors (such as UHF, ultrasonic, and ground wave sensors) to collect data on various physical quantities, using embedded processors or SOCs for preliminary data analysis and alarm generation. However, their multi-channel signal acquisition typically relies on centralized data synchronization and processing methods, which are limited by bandwidth, latency, and synchronization accuracy, making it difficult to meet the demands for efficient and accurate monitoring in complex environments.

[0003] In recent years, with the development of edge computing technology, moving data processing capabilities to the acquisition terminal has become an important trend in improving the real-time performance and intelligence of monitoring systems. Edge computing can perform synchronous correction, feature extraction, and preliminary discrimination of multi-channel acquired signals locally, significantly reducing data transmission pressure and improving system response speed. However, existing technologies still have certain limitations in high-precision synchronous acquisition and real-time processing of multi-channel signals, especially in areas such as multi-source heterogeneous signal fusion, synchronization accuracy, and anti-interference capabilities, where a unified and efficient solution has not yet been formed. Given these shortcomings, how to achieve high-precision real-time synchronous acquisition of multi-channel partial discharge signals under an edge computing architecture, and improve the synchronization, real-time performance, and data processing efficiency of multi-source heterogeneous signals to meet the high reliability and intelligence requirements of online monitoring of power equipment, is a problem that needs to be solved.

[0004] To address this, a method for real-time synchronous acquisition of multi-channel partial discharge signals based on edge computing is proposed. Summary of the Invention

[0005] The purpose of this invention is to provide a method for real-time synchronous acquisition of multi-channel partial discharge signals based on edge computing, so as to solve the problems mentioned in the background art.

[0006] To achieve the above objectives, the present invention provides the following technical solution:

[0007] A signal acquisition architecture of asynchronous sampling-synchronous fusion is adopted. Each channel asynchronously acquires the partial discharge signal of the electrical equipment under test through an independent high sampling rate analog-to-digital converter. The edge node uses a clock and digital processing chip to add a time stamp to the asynchronous data, reconstructs the timing, synchronizes the partial discharge signals of each channel, compensates for inter-channel delay and jitter, and outputs partial discharge signal data.

[0008] Based on the adaptive synchronization algorithm, local time alignment and timing fusion are performed on multi-channel partial discharge signals to output a multi-channel signal group;

[0009] Multi-parameter feature analysis is performed on the multi-channel signal group. Multi-dimensional feature sequences are generated through digital filtering and sliding window segmentation. Correlation analysis of the multi-channel signal group is used to determine the type and timing of partial discharge events.

[0010] An event-driven data processing and storage mechanism is adopted. Based on real-time feature discrimination, abnormal partial discharge signals are classified, stored and uploaded according to event type. Only key information of normal signals is recorded locally. Signal data is subjected to hierarchical compression and security management.

[0011] Edge nodes periodically optimize synchronization parameters and discrimination thresholds based on historical detection and operational data.

[0012] Preferably, the asynchronous sampling-synchronous fusion signal acquisition architecture includes:

[0013] Each signal acquisition channel is connected to an independent high sampling rate analog-to-digital converter (ADC) for acquiring partial discharge signals from the electrical equipment under test. Edge nodes are configured with field-programmable gate arrays (FPGAs) as digital processing chips, and a high-precision clock is used to add timestamps to the partial discharge signal data acquired from each channel. During synchronous timing reconstruction, the data from each channel is compared using a reference trigger signal, and the partial discharge signal data stream is adjusted based on a digital delay line to compensate for the acquisition start time difference and synchronization errors between channels. During acquisition, the system detects short-term clock drift and uses a phase-locked loop (PLL) circuit or digital filtering algorithm to stabilize the clock. After synchronous reconstruction of the multi-channel partial discharge signal data, a cyclic redundancy check (CRC) method is used to verify the integrity of the time stamp, outputting partial discharge test data with a unified time stamp.

[0014] Preferably, the multi-channel signal group includes:

[0015] A high-precision timestamp is added to the acquisition time of the partial discharge signal of each channel. Based on the sampling delay and clock deviation parameters between channels, a dynamic adjustment strategy is adopted to perform local time alignment of the acquired signals. Through interpolation and resampling time fusion processing, the partial discharge signals of different channels are synchronized to generate a multi-channel signal group with a unified time base and time consistency.

[0016] Preferably, the multi-parameter feature analysis includes:

[0017] For the synchronized multi-channel signals, a digital filter is used to denoise and correct baseline drift in the original signal, followed by signal segmentation using a sliding window algorithm. In the time-domain feature extraction process, a zero-crossing detection circuit is used to identify the extreme points and amplitude changes of the pulse signal, while pulse counting is performed by a gated shaping circuit. Frequency-domain feature analysis uses a fast Fourier transform module to decompose the signal spectrum, obtaining the energy distribution and dominant frequency information for each frequency band. In the energy feature calculation section, an integrator circuit is used to integrate the signal amplitude within the single pulse width to obtain the single pulse energy and the periodic average energy. In spectral feature extraction, harmonic components are extracted using a harmonic analysis circuit, and the instantaneous frequency is obtained using a Hilbert transform module. Correlation analysis of the multi-channel signals uses a correlation coefficient calculation module to perform point-by-point correlation comparison of the waveforms of each channel signal, and, combined with timestamp information, outputs the discrimination result of the partial discharge event and the corresponding time.

[0018] Preferably, the event-driven data processing and storage mechanism includes:

[0019] For partial discharge signal data identified as abnormal, the acquisition system integrates an encryption circuit module within its local processing unit to perform block encryption on the original signal data. A differentiated compression algorithm module is used to compare the abnormal event data with a set reference signal template, extracting and compressing only signal segments that show significant differences from the template. The compressed partial discharge signal data and encryption key are stored together in a local non-volatile memory chip and uploaded to a remote monitoring platform in groups according to event type via wired or wireless communication interfaces. For partial discharge signal data not identified as abnormal, the system uses a circular cache storage module with a ring storage structure to save only the key information of the signal; expired data is automatically overwritten when the storage space reaches a preset threshold.

[0020] Preferably, the optimized synchronization parameters and discrimination threshold include:

[0021] The built-in statistical analysis module periodically calls the original records of partial discharge signals and abnormal event records in the historical data acquisition and storage unit, and uses a hardware counter to count the frequency of various abnormal events and the corresponding detection results; the synchronization parameters are corrected by weighted averaging of the sampling delay and clock phase offset parameters of the multi-channel signals through digital circuits; the feature discrimination threshold is dynamically adjusted by using a lookup table method based on historical false alarm and missed alarm data to adjust the signal amplitude, pulse width and frequency threshold.

[0022] Preferably, the edge node includes:

[0023] The system includes a local signal acquisition module, a time synchronization module, a feature analysis module, a hierarchical data storage module, and multiple physical communication interface modules; a processor and digital signal processing chip for local asynchronous acquisition, timestamp marking, synchronization fusion, feature parameter extraction, and event discrimination processing of multi-channel partial discharge signals; data is hierarchically stored in local storage units after feature analysis and event discrimination; edge nodes are used for online monitoring of the insulation status of the tested electrical equipment, real-time analysis of partial discharge signals and fault trend prediction, automatic identification of equipment anomalies and insulation degradation, and generation of diagnostic reports and alarm information based on the analysis results, which are remotely transmitted to the monitoring center via the communication interface, supporting intelligent detection and remote management of the operating status of electrical equipment; edge nodes synchronize time and interact with other edge nodes through hardware synchronization signal lines and high-speed communication buses, supporting multi-node collaborative networking; edge nodes are equipped with physically isolated power management modules to allocate independent power to each functional module.

[0024] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0025] 1. In terms of synchronous acquisition of multi-channel partial discharge signals, an asynchronous sampling-synchronous fusion signal acquisition architecture is adopted. Each channel acquires different types of partial discharge signals through an independent high-sampling-rate analog-to-digital converter. Within the edge nodes, high-precision clock marking, digital delay line compensation, and adaptive synchronization algorithms are used to perform local microsecond-level time alignment and timing reconstruction of the multi-channel asynchronous signals. This overcomes the bottlenecks of traditional centralized synchronization methods limited by bandwidth, latency, and wiring complexity. Even in multi-node distributed networking and complex field environments, it can still ensure high-precision synchronization and unified timing of signals from each channel, greatly improving the system's scalability and monitoring reliability.

[0026] 2. In terms of signal processing and intelligent discrimination, based on edge nodes, a multi-parameter feature analysis module is integrated to simultaneously extract time-domain, frequency-domain, energy, and spectral features from synchronized multi-channel signals. Combined with correlation analysis of multi-channel signals, event template matching, and dynamic time warping algorithms, intelligent discrimination of partial discharge anomalies is achieved. This scheme can effectively identify atypical partial discharge signals such as those with weak amplitude and waveform distortion, improving the detection accuracy and robustness of anomalies under complex operating conditions, and significantly outperforming existing schemes that rely on single features or simple discrimination methods.

[0027] 3. Regarding data management and system expansion, an event-driven data processing and hierarchical storage mechanism is implemented at edge nodes. Signal data identified as abnormal in real time undergoes differentiated compression, encrypted storage, and is uploaded in groups according to event type. Normal signals only retain key information and are cached cyclically. This solution effectively reduces local storage requirements and remote transmission bandwidth pressure, while ensuring the integrity and security of abnormal event data, meeting the actual needs of power field sites for efficient data management and information security.

[0028] 4. Regarding system reliability and intelligent adaptive capabilities, the edge nodes of this invention employ modular hardware design and physically isolated power management. Different functional modules are independently powered and isolated from each other, achieving highly stable operation of the acquisition, processing, and storage units. Simultaneously, the system integrates an adaptive synchronization parameter and discrimination threshold optimization mechanism based on historical operating data, enabling dynamic adjustment of synchronization algorithm parameters and anomaly detection thresholds according to the field environment and equipment status. This solution not only significantly improves the system's operational reliability and anti-interference capabilities under harsh power environments such as high voltage and strong electromagnetic interference, but also enhances the system's self-learning and self-optimization capabilities. The system can continuously optimize synchronization accuracy and anomaly detection performance, effectively reducing false alarms and missed alarms, and minimizing manual maintenance and debugging workload. Attached Figure Description

[0029] Figure 1 This is a flowchart of the real-time synchronous acquisition method for multi-channel partial discharge signals based on edge computing proposed in this invention;

[0030] Figure 2 This is a block diagram illustrating the principle of the asynchronous sampling-synchronous fusion structure proposed in an embodiment of this invention.

[0031] Figure 3 This is a block diagram of the functional structure of the edge node proposed in an embodiment of the present invention. Detailed Implementation

[0032] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0033] Please see Figures 1 to 3 The present invention provides a real-time synchronous acquisition method for multi-channel partial discharge signals based on edge computing, the technical solution of which is as follows:

[0034] Example 1

[0035] This embodiment uses a 110kV substation GIS equipment as the application object. During operation, problems such as insulation aging and defects may cause partial discharge.

[0036] For multi-channel partial discharge signal acquisition, an asynchronous sampling-synchronous fusion signal acquisition architecture is employed. First, various types of partial discharge sensors, such as UHF, ultrasonic, and ground wave sensors, are installed at each monitoring point. The analog signals output by each type of sensor are asynchronously acquired via independent high-sampling-rate analog-to-digital converters. Different channels have their own independent sampling clocks, and the raw signals are not perfectly aligned on the time axis. All acquired raw data first enters the FPGA chip configured at the edge node. The FPGA integrates a high-precision clock management module, which generates a standard time reference using a temperature-compensated crystal oscillator or an external high-precision timing signal. The FPGA adds a timestamp generated based on this high-precision clock to all input acquisition channel data in real time. This timestamp can achieve microsecond or even higher precision, ensuring that each frame of data accurately records its acquisition time. Next, the FPGA compares the data streams acquired by each channel using the set reference trigger signal. It automatically detects and calculates the sampling start time difference between each channel caused by factors such as hardware delay, cable length, and front-end response characteristics. With the built-in digital delay line, it dynamically adjusts the timing of each acquired data to align them on a unified time axis. In order to overcome the short-term drift of the clock signal that may occur during long-term operation, the system also introduces a phase-locked loop circuit to continuously lock the reference clock and uses digital filtering algorithms when necessary to remove high-frequency noise or periodic interference, ensuring high stability and consistency of the clock signal. A synchronization error self-calibration module based on multi-channel cross-correlation analysis is integrated into the edge node. This module includes an event triggering unit, a cross-correlation calculation unit, and a parameter feedback adjustment unit. The event triggering unit monitors the partial discharge signals acquired by each channel in real time and automatically captures typical event waveforms. The cross-correlation calculation unit performs cross-correlation analysis on the event waveform sequences of different channels to accurately measure the actual time delay between channels. The parameter feedback adjustment unit dynamically adjusts the time-stamp alignment parameter or digital delay compensation amount in the synchronization fusion algorithm based on the correlation results, realizing automatic correction of remaining small synchronization errors. This enables high-precision synchronization of multi-channel acquired data at the millisecond or even microsecond level, significantly improving the channel time consistency and overall synchronization performance of the system under complex operating conditions such as high noise and strong interference. Before outputting the synchronized multi-channel data, the timestamp field of all data frames is checked for integrity and consistency using a cyclic redundancy check method. If timestamp anomalies or loss are detected, the redundancy mechanism can be used to restore or trigger an alarm. Finally, the verified multi-channel signals are output with a unified timestamp, forming a highly consistent synchronized multi-channel dataset that can be used for subsequent analysis.

[0037] This method enables high-precision synchronous fusion of original asynchronous sampling signals during the acquisition of multi-type, multi-sampling-channel partial discharge signals. It significantly improves the consistency of multi-channel signals in the time domain and the accuracy of time stamps, providing reliable foundational data for subsequent applications such as spatiotemporal correlation analysis of partial discharge events, discharge type identification, and fault location. This method effectively lowers the technical barrier of multi-channel synchronization relying on expensive distributed clocks or complex wiring, enhances the flexibility and engineering adaptability of online data fusion at the edge, and ensures the stability and integrity of field monitoring data, providing strong support for the practical promotion of online partial discharge monitoring technology.

[0038] Furthermore, in the signal acquisition architecture employing asynchronous sampling-synchronous fusion, multiple acquisition channels independently sample the partial discharge signal. Each acquisition channel adds a timestamp to each set of signal data based on its local high-precision clock. The sampling times of the acquisition channels are not strictly synchronized physically, constituting an asynchronous sampling mode. After the raw signals and timestamps acquired by all channels are uploaded to the edge computing node, the data from each channel are first summarized and analyzed using the timestamps to extract the sampling delay and clock deviation information for each signal segment. Based on these parameters, the edge node adjusts the time axis of the acquired signal data: for channel data with sampling delays, a translation correction is performed according to the timestamp order to arrange them under the same reference time axis; for cases with large clock deviations, time base calibration is performed through timestamp compensation. After calibration, for cases where sampling points in each channel are missing or misaligned under a unified time reference, interpolation methods are used to fill in the missing data. Interpolation methods can include linear interpolation, spline interpolation, etc. For channels with inconsistent sampling rates or differences in data volume due to clock drift, a resampling algorithm is applied to unify the sampling rate and sampling point distribution. After the above local time alignment and timing fusion processing, the signal data from different channels are synchronized to a unified time reference and grouped into multi-channel signal groups according to time sequence, achieving synchronous fusion in data structure.

[0039] By following the above steps, it is possible to ensure that the multi-channel asynchronous sampling data is synchronized and fused with high precision before subsequent analysis. This ensures that the multi-channel signal groups are highly consistent in terms of time reference and signal timing, providing an accurate basis for joint analysis and feature comparison of partial discharge signals. It also improves the reliability and scientific rigor of multi-channel data statistical analysis and abnormal signal detection.

[0040] Furthermore, for the synchronized multi-channel signal group, a digital filter is initially used to process the original signal sequence to remove noise, filtering out power frequency interference and high-frequency noise. The low-frequency component of the filter automatically corrects the signal baseline, eliminating the impact of background drift on feature extraction. The denoised and corrected signal stream is then input into a sliding window segmentation algorithm. The sliding window moves across the signal sequence according to a preset step size, dividing the long-term sequence into multiple continuous signal segments, each of which serves as the basic analysis unit for subsequent feature extraction. In the time-domain analysis stage, the zero-crossing detection algorithm automatically determines the extreme points of each pulse signal, and identifies and statistically analyzes the maximum amplitude and waveform change amplitude of each pulse. Combined with a gated shaping algorithm, pulse counting is performed on the signal segments to achieve accurate quantification of the frequency of partial discharge pulses. After entering the frequency-domain analysis stage, a fast Fourier transform algorithm is used to perform spectral decomposition on the segmented signals to obtain the energy distribution spectrum of different frequency bands and the signal's dominant frequency information. In the energy parameter extraction stage, an integral algorithm is used to integrate the signal amplitude within the width of a single pulse to obtain the single-pulse energy intensity. Simultaneously, the average energy value within each signal cycle is statistically analyzed to reflect the intensity variation of partial discharge events. In spectral feature analysis, a harmonic component extraction algorithm is used to separate the amplitudes of each harmonic order, and the Hilbert transform method is used to extract the instantaneous frequency characteristics of the signal, improving the ability to identify abnormal pulses and complex waveforms. For correlation analysis between multi-channel signals, an event-triggered similarity matching module can be integrated. After synchronizing and denoising multi-channel signal segments, the system monitors the time window of suspected partial discharge events in the signal, automatically extracts the corresponding signal segments from each channel, and extracts multi-dimensional feature parameters from these segments, including pulse amplitude, rise and fall slopes, dominant frequency components, and energy distribution. The system can then perform template matching based on preset event templates, or use dynamic time warping algorithms to perform nonlinear alignment and similarity scoring on signal segments of each channel, identifying synchronous or asynchronous partial discharge events between channels. This method can achieve high-precision discrimination of the synchronization of multi-channel partial discharge events in complex scenarios where multi-channel signals have nonlinear and non-stationary characteristics such as waveform distortion, amplitude variation, and phase drift, improving the ability of collaborative analysis of abnormal events and spatial positioning. By employing a correlation coefficient calculation method, the system performs point-by-point correlation comparison of signal segments from different channels under the same time reference, combining this with timestamp data to output the synchronization discrimination results of partial discharge events and their occurrence times, providing support for spatial positioning and multi-channel collaborative analysis.

[0041] The above-mentioned multi-parameter feature analysis method can comprehensively extract features from partial discharge signals from multiple dimensions such as time domain, frequency domain, energy, spectral features and multi-channel correlation, realize multi-level discrimination and quantitative analysis of complex partial discharge events, improve the accuracy and robustness of partial discharge event identification, and provide diverse and accurate basic data for subsequent fault diagnosis, condition assessment and trend prediction.

[0042] Furthermore, for multi-channel partial discharge signal data that has been synchronized and undergone feature analysis, an event-driven data processing and storage mechanism is adopted. First, within the edge processing unit, based on multi-parameter feature extraction and discrimination algorithms, the acquired signals are automatically classified as abnormal or normal. For signal data identified as abnormal, the locally integrated encryption circuit module is scheduled to perform block-by-block encryption processing on the original signal data using group or streaming encryption technology. Each data block is assigned an independent key during the encryption process, and the key is dynamically generated and managed by the internal security module to ensure that even if data interception occurs, the original signal cannot be recovered, effectively preventing data leakage. After encryption, the data is sent to the differential compression algorithm module for secondary processing. This module uses a reference signal template as a benchmark to compare the abnormal signal with the template point by point or segment by segment. The algorithm dynamically identifies the intervals that are significantly different from the template, and only these intervals are used for data extraction and efficient compression. The compression algorithm can adopt sparse coding based on threshold discrimination, local adaptive dictionary method, or variable-length coding based on event triggering, etc., to further reduce the data volume. The compressed, differentiated data and generated keys are archived and stored locally on a non-volatile chip according to event categories. They are then uploaded to a remote monitoring platform via wired (e.g., Ethernet) or wireless (e.g., 4G / 5G, Wi-Fi) communication interfaces, grouped according to abnormal event type and priority. The system supports multiple data transmission protocols and failure retransmission mechanisms to ensure real-time and reliable arrival of abnormal event data. For normal signal data not identified as abnormal, the system employs a circular caching mechanism. A circular buffer temporarily stores the main statistical parameters and characteristic values ​​of the signal (e.g., pulse count, main frequency, average energy, etc.) in chronological order. The original waveform data is not persistently stored. Once the buffer capacity reaches a set threshold, the oldest data is automatically overwritten by subsequent new data, achieving dynamic updates and space reuse. This circular caching module also supports rapid retrieval and export of key information by time window, facilitating subsequent data traceability and trend analysis. A priority-based data buffering strategy is further introduced. The system automatically and dynamically manages the local data buffer based on the type and severity of partial discharge signal events. When a major or rare anomaly is detected, more storage space and bandwidth resources are allocated to ensure that such data is completely preserved and quickly uploaded to the remote monitoring platform. For common or low-risk events, a low-priority processing approach is adopted, saving only brief statistical parameters or delaying the transmission of raw data. This mechanism enables intelligent scheduling of data transmission and storage for different types of events within limited edge computing and storage resources, improving the response efficiency of critical anomalies while reducing overall system data redundancy and resource consumption.

[0043] By employing an event-driven data processing and storage mechanism, this method not only ensures the secure encryption and efficient differentiated compression of abnormal partial discharge signals locally, effectively improving data security and storage utilization, but also enables timely remote reporting of critical abnormal events through event type grouping and priority uploading, guaranteeing the monitoring system's rapid response capability to important electrical faults. Furthermore, the use of critical information caching and automatic overwrite modes for normal signals significantly reduces data redundancy and local storage pressure, facilitating long-term continuous system operation and large-scale multi-point deployment. This provides an efficient, scalable, and low-cost technical foundation for intelligent monitoring and condition assessment of partial discharge in power equipment.

[0044] In the real-time synchronous acquisition process of multi-channel partial discharge signals in edge computing, a dynamic optimization mechanism for synchronization parameters and feature discrimination thresholds is specifically designed. In practice, the local historical data management module is periodically invoked to automatically retrieve previously acquired raw partial discharge signal data and abnormal event identification records. A built-in statistical analysis module, in conjunction with a hardware counter, performs categorized statistical analysis on the frequency of various abnormal events and their corresponding discrimination accuracy. For example, the statistical module can record the number of false alarms, missed alarms, and accurate discriminations within different time windows, and establish associations with historical environmental parameters, equipment operating status, and other information. Regarding the synchronization problem of multi-channel signals, the system periodically analyzes the synchronization errors between timing signals. Based on the statistically derived core parameters such as sampling delay and clock phase offset, a weighted average correction is applied to the sampling timing signals of each channel using digital circuits, dynamically adjusting the synchronization parameters to ensure the timing consistency of the acquired data from each channel. To optimize the feature discrimination threshold, we continuously track historical discrimination errors and false alarms and missed alarms on site. Relying on a lookup table, we automatically and dynamically adjust parameters such as signal amplitude threshold, pulse width threshold, and frequency threshold. The lookup table sets a series of threshold ranges based on changes in the actual acquisition environment and the distribution of historical misjudgments. It automatically selects the optimal threshold based on the latest statistical results and updates the feature discrimination standard in real time to adapt to the complex and ever-changing operating environment on site.

[0045] By employing the aforementioned adaptive optimization method for synchronization parameters and discrimination thresholds, the synchronization accuracy of signals in each channel can be continuously improved based on historical data and event recognition performance at the site. This effectively reduces acquisition timing errors and data offsets between channels, dynamically adjusts feature discrimination thresholds, and makes the identification of partial discharge events more accurate and robust. This method endows long-term, unattended intelligent monitoring of partial discharge in power equipment with powerful self-learning and adaptive capabilities, reduces manual intervention, improves the system's adaptability in complex environments and the accuracy of abnormal event detection, and lays a solid data foundation for subsequent operation and maintenance decisions and fault early warning.

[0046] Furthermore, the synchronous acquisition of multi-channel partial discharge signals based on edge computing clarifies the functional modules and their collaborative processes within the edge node. The edge node includes a local signal acquisition module, an event synchronization module, a feature analysis module, a hierarchical data storage module, multiple physical communication interface modules, and a physically isolated power management module. The signal acquisition module, relying on a multi-channel high-speed analog-to-digital converter, can simultaneously perform high-precision digital processing on the raw analog signals from multiple discharge detection channels. The event synchronization module integrates a high-precision clock source and hardware synchronization signal lines to ensure that all channel acquisition data has a unified and accurate timestamp, facilitating subsequent synchronous fusion and timing consistency correction of multi-channel signals. All acquired data first flows into a high-performance processor or digital signal processing chip, which is responsible for the synchronous fusion of locally asynchronous acquisition data, parameter extraction based on multi-dimensional features such as time-frequency domain and energy, and intelligent event discrimination. The feature analysis module is embedded within the processor or exists as a coprocessor unit. It can quickly analyze the timing characteristics, frequency domain characteristics, energy distribution, and inter-channel correlation of signals. After event identification, the data is routed to the data hierarchical storage module, which classifies and manages abnormal and normal events. Abnormal data is encrypted and compressed by the encryption and compression module to ensure data security and save storage space. Normal data is stored only with necessary feature parameters and statistical information by the circular cache module, and expired data is automatically overwritten using a ring storage structure. All modules are interconnected via a high-speed bus to ensure real-time and efficient data interaction under large data volumes. For remote data transmission, edge nodes integrate multiple wired and wireless communication interfaces to achieve bidirectional data interaction with remote monitoring platforms or the cloud. Multiple nodes can use synchronization signal lines and high-speed buses to achieve time synchronization and data sharing, facilitating network collaborative applications. In addition, each node is equipped with a physically isolated power management module to provide independent power to each functional unit, preventing power interference and fault propagation, and improving the electrical safety and reliability of the overall system.

[0047] Through the modular design and implementation of the aforementioned edge nodes, high-speed, synchronous, and hierarchical processing of multi-channel partial discharge signals can be achieved locally, greatly improving signal synchronization accuracy and feature extraction efficiency. Furthermore, local intelligent discrimination and hierarchical storage reduce redundant transmission of large-scale data and the pressure on remote storage. Multiple communication interfaces and physical synchronization mechanisms between nodes ensure multi-node collaborative networking capabilities, enabling the system to flexibly expand and adapt to complex, distributed monitoring scenarios. Physically isolated power management further guarantees the long-term stable operation of each module, effectively preventing system-level risks caused by electrical faults. This achieves a highly reliable, highly scalable, and highly adaptive intelligent acquisition and processing method for partial discharge signals, providing a solid technical foundation for the condition monitoring and fault early warning of power equipment.

[0048] This embodiment employs an asynchronous sampling-synchronous fusion signal acquisition process to asynchronously acquire multi-channel partial discharge signals from the electrical equipment under test. Each signal is digitized using an independent high-sampling-rate analog-to-digital converter. At the edge nodes, each acquired channel signal is appended with a high-resolution timestamp using a high-precision clock and digital processing unit, achieving time-series reconstruction and synchronous fusion of the asynchronous data to compensate for inter-channel delays and jitter. After synchronization, an adaptive synchronization algorithm is used to perform local time alignment and time-series data fusion of each channel signal based on the calibrated time information, constructing a complete multi-channel synchronous signal group. Subsequently, multi-parameter feature analysis is performed on the synchronous signal group, including extraction of time-domain, frequency-domain, energy, and spectral features. Simultaneously, the correlation between multi-channel signals is analyzed to further determine the type and occurrence time of partial discharge events. For the acquired data, an event-driven processing and storage process is adopted. Real-time feature discrimination classifies, compresses, and uploads partial discharge signals identified as abnormal according to event type, while normal signals only retain key information in a local buffer, achieving hierarchical compression and data security management. Finally, historical testing data is reviewed periodically, and multi-channel synchronization parameters and feature discrimination thresholds are automatically optimized based on statistical analysis to dynamically adapt to changes in the field operating environment.

[0049] By organically combining the above steps, this embodiment can achieve accurate synchronous acquisition and efficient local processing of multi-channel partial discharge signals, significantly improving the accuracy of signal synchronization and the reliability of event discrimination in asynchronous acquisition scenarios. The event-driven hierarchical data processing method effectively reduces redundant storage and transmission of normal data, improving the efficiency of overall data management and remote interaction. Relying on adaptive optimization of parameters and thresholds, it possesses good on-site adaptability and continuously improving identification accuracy, providing advanced technical means and methodological basis for real-time monitoring and intelligent early warning of partial discharge in electrical equipment.

[0050] Example 2

[0051] This embodiment uses a 110kV substation GIS equipment as the application object. During operation, problems such as insulation aging and defects may cause partial discharge.

[0052] In this embodiment, for the multi-channel partial discharge signals after synchronous fusion, each signal enters the digital filtering process separately. Taking FIR (Finite Impulse Response) and IIR (Infinite Impulse Response) filtering as examples, FIR filters are typically used to suppress fixed-frequency power frequency interference and high-frequency noise with a wide bandwidth. Their filter coefficients can be pre-calculated and fixed in the digital processing module according to the actual sampling rate and interference frequency band on site, using methods such as window function design or least squares. IIR filters are more suitable for achieving high-order, narrow-band filtering requirements, such as suppressing specific harmonic components. Before entering feature extraction, each signal passes through the above-mentioned filter link. The filter parameters (such as cutoff frequency, stopband attenuation, etc.) can be flexibly configured according to the environmental noise test results to achieve the goal of not excessively weakening the effective signal while maximizing the suppression of background noise. After filtering, to prevent baseline drift from affecting the accurate identification of pulse amplitude, the system uses a low-pass filter or moving average algorithm to extract the signal baseline and subtract it from the original signal to achieve automatic baseline correction, making subsequent feature extraction more reliable. Signal segmentation is implemented using a sliding window algorithm. The window length is generally set slightly longer than the duration of a single partial discharge pulse, and the step size can be 1 / 2 to 1 / 10 of the sampling point, depending on the actual pulse density and computational resources. During the window sliding process, each extracted segment is treated as an independent analysis unit for subsequent processing. In terms of time-domain features, the zero-crossing detection algorithm quickly identifies the start and end points of the pulse by judging the positive and negative changes of the signal point by point. Then, extreme point search is applied within each pulse interval to accurately count the maximum amplitude and duration of the pulse. Combined with the pulse counting algorithm, the system can accumulate the number of discharge pulses within each window, providing basic data for equipment status assessment. In the frequency domain analysis, the method uses the FFT algorithm to transform the signal segment within each sliding window from the time domain to the frequency domain, obtaining the amplitude spectrum and energy distribution of each frequency component. By analyzing the dominant frequency component, the main frequency characteristics of the partial discharge signal can be identified, and changes in high-frequency or harmonic components can be detected, which helps in identifying abnormal discharge types. In terms of energy feature extraction, the integral algorithm performs numerical integration on the discrete points of the signal amplitude within the time period of each pulse occurrence to obtain the pulse energy intensity. It also calculates the mean, variance, and other data of all pulse energies within the entire cycle or window to reflect the overall activity and trend of the discharge event. In the spectrum feature analysis stage, harmonic component separation is achieved by extracting the amplitude of the higher-order frequency components of the FFT result to identify the characteristics of higher harmonics and nonlinear interference.Hilbert transform is used to obtain the instantaneous amplitude and frequency of a signal, which can capture short-term abrupt changes and complex waveform characteristics. It is particularly suitable for identifying non-periodic abnormal pulses. Multi-channel correlation analysis uses a unified timestamp as a reference to calculate the correlation coefficient point by point for signal segments of each channel under the same sliding window. It outputs the synchronicity and similarity of signals between channels. If the correlation is high, it indicates that the discharge event occurs synchronously at multiple monitoring points and has spatial distribution characteristics. The correlation analysis results can be used for spatial positioning and source identification, providing data support for subsequent equipment fault analysis.

[0053] By employing the aforementioned multi-parameter feature extraction and correlation discrimination methods, in-depth multi-channel, full-dimensional feature mining of partial discharge signals can be achieved under complex noise backgrounds and multi-source signal environments. This method not only enhances the extraction and identification capabilities of partial discharge events in a single sampling channel but also incorporates spatial distribution characteristics and signal correlations into the overall discrimination system through multi-channel collaborative analysis, significantly improving the ability to distinguish partial discharge types, occurrence locations, and event synchronicity. Simultaneously, the use of automatic parameter adjustment and multi-step verification mechanisms greatly reduces the errors and insufficient adaptability caused by manual threshold setting. Ultimately, the method provides high-precision, multi-level raw data and criteria support for subsequent fine classification of partial discharge events, abnormal trend early warning, and electrical equipment fault location, significantly improving the application value and intelligence level of partial discharge monitoring methods in practical engineering.

[0054] Example 3

[0055] This embodiment uses a 110kV substation GIS equipment as the application object. During operation, problems such as insulation aging and defects may cause partial discharge.

[0056] In this embodiment, the various functional modules of the edge node work collaboratively to complete the acquisition, processing, and transmission of multi-channel partial discharge signals. The signal acquisition module uses a multi-channel ADC to perform parallel sampling of various discharge sensor signals. The acquired raw data is first timestamped by a hardware synchronization module driven by a high-precision clock, ensuring that the acquisition time of each signal can be accurately determined during subsequent data processing. Time synchronization relies on the local clock and can also dynamically correct the local time reference with the help of external time signals according to the needs of multi-node collaboration, improving the synchronization accuracy when networking across nodes. The acquired multi-channel data will be synchronized and fused on the local high-performance processor / DSP. Specifically, it uses algorithms such as delay compensation, interpolation, and resampling to map the signals of each channel to a unified reference time axis, achieving data structure-level synchronization. During the feature analysis phase, edge nodes employ a modular algorithm library to automatically extract time-domain, frequency-domain, energy, and correlation features suitable for partial discharge signal characteristics. Some features, such as main frequency, harmonics, and pulse count, are accelerated using hardware co-processing units to avoid processing bottlenecks caused by large data volumes. The event discrimination module analyzes feature results in real time based on multi-parameter thresholds and adaptive decision algorithms to automatically determine whether an event is an abnormal partial discharge event. For data identified as abnormal, the system calls locally integrated encryption and compression algorithms to process the original waveform data in blocks, prioritizing the storage and uploading of abnormal segments that differ significantly from typical templates, saving storage space and improving data security. Normal data only stores statistical features in a circular buffer, and the space is dynamically managed using an automatic overwrite strategy. All data transfer links utilize a high-speed bus to achieve low-latency interconnection, improving the real-time processing capabilities within the node. The node is also equipped with multiple communication interfaces, supporting automatic switching between wired / wireless / networked modes to achieve flexible data sharing and remote interaction. In terms of power management, physical isolation, independent allocation, and fault self-switching mechanisms are adopted to ensure stable power supply to each functional module and prevent the spread of single-point faults.

[0057] Through the integration and implementation of the above, edge nodes can achieve highly automated synchronous acquisition of multi-channel partial discharge signals, feature parameter extraction, intelligent event discrimination, and hierarchical and weighted data management. This method improves the accuracy of time synchronization and data fusion for multi-channel signals. Furthermore, by leveraging localized intelligent analysis and hierarchical storage, it significantly reduces data redundancy and remote transmission pressure, optimizing resource utilization and response speed of the monitoring system. Multiple communication interfaces and physical synchronization mechanisms work together, allowing for flexible networking among multiple nodes to adapt to the actual needs of distributed, multi-source monitoring. Physically isolated power management ensures long-term stable operation of nodes, reducing external electrical interference and the risk of cascading failures. Overall, this method lays a solid technical foundation for the intelligent online acquisition and status diagnosis of partial discharge signals from large-scale, distributed power equipment, improving the system's reliability, scalability, and engineering adaptability, and providing a new means for intelligent operation and maintenance and fault early warning of power systems.

[0058] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for real-time synchronous acquisition of multi-channel partial discharge signals based on edge computing, characterized in that, include: A signal acquisition architecture of asynchronous sampling-synchronous fusion is adopted. Each channel asynchronously acquires the partial discharge signal of the electrical equipment under test through an independent high sampling rate analog-to-digital converter. The edge node uses a clock and digital processing chip to add a time stamp to the asynchronous data, reconstructs the timing, synchronizes the partial discharge signals of each channel, compensates for inter-channel delay and jitter, and outputs partial discharge signal data. Based on the adaptive synchronization algorithm, local time alignment and timing fusion are performed on multi-channel partial discharge signals to output a multi-channel signal group; Multi-parameter feature analysis is performed on multi-channel signal groups. Multi-dimensional feature sequences are generated through digital filtering and sliding window segmentation. Correlation analysis of multi-channel signal groups is used to determine the type and timing of partial discharge events. For synchronized multi-channel signals, digital filters are used to denoise and correct baseline drift, followed by signal segmentation using a sliding window algorithm. In the time-domain feature extraction process, a zero-crossing detection circuit is used to identify the extreme points and amplitude changes of pulse signals, while pulse counting is accomplished by a gated shaping circuit. Frequency domain feature analysis uses a fast Fourier transform module to perform spectral decomposition on the signal to obtain energy distribution and dominant frequency information for each frequency band; In the energy characteristic calculation section, the signal amplitude within the single pulse width is integrated using an integrator circuit to obtain the single pulse energy and the periodic average energy. In the spectral feature extraction, each harmonic component is extracted through the harmonic analysis circuit, and the instantaneous frequency is obtained by combining the Hilbert transform module; the correlation analysis of multi-channel signals adopts the correlation coefficient calculation module to perform point-by-point correlation comparison of the waveforms of each channel signal, and outputs the discrimination result of partial discharge event and the corresponding time by combining the timestamp information. An event-driven data processing and storage mechanism is adopted. Based on real-time feature discrimination, abnormal partial discharge signals are classified, stored, and uploaded according to event type. Only key information of normal signals is recorded locally. Signal data is subject to hierarchical compression and security management. For partial discharge signal data identified as abnormal, the acquisition system integrates an encryption circuit module within the local processing unit to perform block encryption on the original signal data. A differential compression algorithm module is used to compare the abnormal event data with a set reference signal template, extracting and compressing only signal segments that are significantly different from the template. The compressed partial discharge signal data and encryption key are stored together in a local non-volatile memory chip and uploaded to a remote monitoring platform in groups according to event type via wired or wireless communication interfaces. For partial discharge signal data that is not identified as abnormal, the system adopts a circular cache storage module. Using a ring storage structure, only key information of the signal is saved, and expired data is automatically overwritten when the storage space reaches a preset threshold. The edge nodes periodically optimize synchronization parameters and discrimination thresholds based on historical detection and operation data. The built-in statistical analysis module periodically calls the original records of partial discharge signals and abnormal event records in the historical data collection and storage unit, and uses hardware counters to count the frequency of various abnormal events and the corresponding detection results. The synchronization parameters are corrected by weighted averaging of the sampling delay and clock phase offset parameters of the multi-channel signals through digital circuits; The feature discrimination threshold is dynamically adjusted based on historical false alarm and missed alarm data, using a lookup table method to adjust the signal amplitude, pulse width, and frequency threshold.

2. The method for real-time synchronous acquisition of multi-channel partial discharge signals based on edge computing according to claim 1, characterized in that, The asynchronous sampling-synchronous fusion signal acquisition architecture includes: Each signal acquisition channel is connected to an independent high sampling rate analog-to-digital converter (ADC) for acquiring partial discharge signals from the electrical equipment under test. Edge nodes are configured with field-programmable gate arrays (FPGAs) as digital processing chips, and a high-precision clock is used to add timestamps to the partial discharge signal data acquired from each channel. During synchronous timing reconstruction, the data from each channel is compared using a reference trigger signal, and the partial discharge signal data stream is adjusted based on a digital delay line to compensate for the acquisition start time difference and synchronization errors between channels. During acquisition, the system detects short-term clock drift and uses a phase-locked loop (PLL) circuit or digital filtering algorithm to stabilize the clock. After synchronous reconstruction of the multi-channel partial discharge signal data, a cyclic redundancy check (CRC) method is used to verify the integrity of the time stamp, outputting partial discharge test data with a unified time stamp.

3. The method for real-time synchronous acquisition of multi-channel partial discharge signals based on edge computing according to claim 1, characterized in that, The multi-channel signal group includes: A high-precision timestamp is added to the acquisition time of the partial discharge signal of each channel. Based on the sampling delay and clock deviation parameters between channels, a dynamic adjustment strategy is adopted to perform local time alignment of the acquired signals. Through interpolation and resampling time fusion processing, the partial discharge signals of different channels are synchronized to generate a multi-channel signal group with a unified time base and time consistency.

4. The method for real-time synchronous acquisition of multi-channel partial discharge signals based on edge computing according to claim 1, characterized in that, The edge nodes include: The system includes a local signal acquisition module, a time synchronization module, a feature analysis module, a hierarchical data storage module, and multiple physical communication interface modules; a processor and digital signal processing chip for local asynchronous acquisition, timestamp marking, synchronization fusion, feature parameter extraction, and event discrimination processing of multi-channel partial discharge signals; data is hierarchically stored in local storage units after feature analysis and event discrimination; edge nodes are used for online monitoring of the insulation status of the tested electrical equipment, real-time analysis of partial discharge signals and fault trend prediction, automatic identification of equipment anomalies and insulation degradation, and generation of diagnostic reports and alarm information based on the analysis results, which are remotely transmitted to the monitoring center via the communication interface, supporting intelligent detection and remote management of the operating status of electrical equipment; edge nodes synchronize time and interact with other edge nodes through hardware synchronization signal lines and high-speed communication buses, supporting multi-node collaborative networking; edge nodes are equipped with physically isolated power management modules to allocate independent power to each functional module.

Images