A partial discharge comprehensive online detection and visualization management system for electrical equipment

Abstract

This invention discloses a comprehensive online detection and visualization management system for partial discharge (PD) in electrical equipment, relating to the field of PD detection technology. The system includes a data acquisition and preprocessing module for acquiring and preprocessing electrical equipment detection data; a drift disturbance assessment module for analyzing link delay drift using timing offset, link load, and node status data; a dynamic correction assessment module for performing event merging correction based on reliability evaluation results; a fusion stability discrimination module for performing cross-channel merging, pulse morphology characterization, cross-modal consistency characterization, and PD defect type output operations based on stability discrimination results; and a visualization management module for enabling link drift tracking and visual comparison of correction effects. This system solves the problems of false alarms and missed alarms caused by cross-channel timing mismatches and unstable defect type discrimination in PD detection of electrical equipment.

Description

Technical Field

[0001] This invention relates to the field of partial discharge detection technology, specifically to a comprehensive online detection and visualization control system for partial discharge in electrical equipment. Background Technology

[0002] During power system operation, partial discharge (PD) in electrical equipment is a key hidden danger that can lead to insulation degradation and ultimately equipment failure. With the continuous expansion of the power grid, the number of high-voltage electrical equipment such as transformers, switchgear, and cables has increased significantly, and their operating status directly affects the reliability of the entire power system. Therefore, effective PD detection of high-voltage electrical equipment and real-time monitoring of its insulation status have become one of the core requirements of power system operation and maintenance.

[0003] For example, the invention patent with announcement number CN117811220B discloses a method and system for online monitoring of partial discharge based on overloaded data volume. The method includes selecting a partial discharge detection module and deploying it as a UDP server; the partial discharge detection module and the monitoring client connecting to the UDP server; requesting access to the partial discharge detection module from the UDP server; and establishing a communication channel between the monitoring client and the partial discharge detection module on the public network based on the communication port information returned by the UDP server. The communication port information includes first port information and second port information. In response to the access request from the monitoring client, the UDP server returns the public IP port of the monitoring client to be communicated as the first port information to the partial discharge detection module, and the public IP port of the partial discharge detection module to be communicated as the second port information to the monitoring client. The monitoring client and the partial discharge detection module communicate in the communication channel, reducing the load on the cloud server and improving the stability of the remote backend.

[0004] For example, the invention patent with announcement number CN120214524B discloses a real-time partial discharge monitoring system based on pulse current, including a signal acquisition module, a preprocessing module, a preliminary diagnosis module, a fault isolation module, and a deep diagnosis module. The real-time acquired signals undergo preprocessing such as data labeling, baseline calibration, and pulse extraction to accurately remove interference and ensure data reliability. Through targeted detection of sensor open circuits and sampling card anomalies, faults can be quickly located. The fault isolation module can automatically switch to a backup sensor when a fault occurs to reduce the impact of the fault. The deep diagnosis module uses multi-dimensional features to construct a fault feature vector and compares it with the sample set to achieve fine classification of faults. The fault early warning module can send sensor and partial discharge fault early warning signals to maintenance personnel in real time. This improves the accuracy and reliability of monitoring, provides strong protection for the safe and stable operation of equipment, and reduces the risk of accidents.

[0005] However, existing detection devices face a technical challenge: partial discharge (PD) detection involves multiple sensor modules, acquisition nodes, forwarding nodes, and aggregation nodes. Signals travel from the acquisition end to the processing end through multiple stages, including acquisition, transmission, aggregation, and processing. Under complex conditions such as electromagnetic interference, link load variations, and fluctuations in interface buffers and forwarding queues, the data streams of each module exhibit non-fixed micro-delay differences along the "acquisition-transmission-aggregation-processing" link, and these delay differences drift over time. Existing devices typically employ fixed compensation parameters or simple timestamp alignment methods, which cannot effectively address this time-varying drift. This leads to cross-channel alignment errors in the arrival time of the same PD event on different modules, resulting in events splicing errors and mismatched related features. This problem is particularly prominent in multimodal correlation analysis, easily causing increased false alarms and false negatives, unstable defect type identification, and fluctuating trend assessments. The problem is even more pronounced under high sampling rates or high-concurrency event conditions.

[0006] Therefore, in order to address the above problems, there is an urgent need for a comprehensive online detection and visual control system for partial discharge of electrical equipment. Summary of the Invention Technical problems to be solved To address the shortcomings of existing technologies, this invention provides a comprehensive online detection and visualization control system for partial discharge of electrical equipment, which solves the problems of false alarms and missed alarms caused by cross-channel timing mismatches and unstable defect type identification in partial discharge detection of electrical equipment.

[0007] Technical solution To achieve the above objectives, the present invention provides the following technical solution: a comprehensive online detection and visualization control system for partial discharge of electrical equipment, comprising: a data acquisition and preprocessing module for acquiring electrical equipment detection data, preprocessing the data, storing it, and constructing a partial discharge detection database; a drift disturbance assessment module for performing link delay drift analysis based on time-series offset, link load, and node status data, and performing drift interval determination and time compensation parameter adjustment operations based on the drift analysis results; a dynamic correction assessment module for evaluating the splicing reliability of events after time compensation based on correction time difference, power frequency phase, and pulse morphology data, and performing event merging correction and compensation-limited event marking operations based on the reliability evaluation results; a fusion stability discrimination module for judging the stability of multi-modal signal fusion by combining correction results, phase correlation, and feature matching data, and performing cross-channel merging, pulse morphology characterization, cross-modal consistency characterization, and partial discharge defect type output operations based on the stability discrimination results; and a visualization management module for calling the partial discharge detection database records to construct time series curves, realizing link drift tracking and visualization comparison of correction effects.

[0008] Further, the specific steps for collecting electrical equipment testing data are as follows: Collecting electrical equipment testing data involves: obtaining UHF signal data by recording the partial discharge pulse waveform, pulse arrival time, and pulse envelope sequence using an UHF antenna; obtaining high-frequency signal data by recording the partial discharge radiation pulse sequence, pulse arrival time, and pulse amplitude sequence using a high-frequency current transformer; obtaining ultrasonic signal data by recording the acoustic emission waveform, acoustic signal arrival time, and sound pressure peak sequence using a piezoelectric ultrasonic sensor; obtaining power frequency phase data by recording the power frequency phase position corresponding to each channel's partial discharge segment using a synchronous sampling unit; obtaining link status data by recording changes in the forwarding queue, interface buffer occupancy, link transmission load, and message flow time records using acquisition nodes, forwarding nodes, and aggregation nodes; and obtaining node status data by recording changes in the local time base, event timestamp recording results, and processing waiting status at each node.

[0009] Further, the specific steps for preprocessing and storing electrical equipment testing data and constructing a partial discharge (PD) detection database are as follows: Perform unified time reference mapping, observation time window segmentation, abnormal pulse removal, noise segment suppression, missing segment completion, and event number association on the electrical equipment testing data to form a set of PD observation segments. Then, normalize the data using the minimum and maximum value method to map it to a unified numerical range. Obtain the cross-channel time difference by calculating the arrival times of multiple channels corresponding to the same PD event; obtain the cross-node timestamp deviation by calculating the timestamps of multiple nodes corresponding to the same PD event; obtain the queue fluctuation amplitude by observing the waiting changes during the forwarding queuing process; and obtain the interface cache occupancy status... The system obtains the interface buffer occupancy rate, the link load rate through the data carrying of the transmission path where the event is located, the clock drift rate through the ratio of the node's local time base to the reference time base, the power frequency phase difference through the power frequency phase position difference of different channels, the pulse envelope similarity through the pulse envelope sequence correspondence of partial discharge segments of different channels, and the cross-modal feature matching difference by calculating the cumulative Euclidean distance of the sampling points corresponding to the amplitude sequence, the cumulative normalized amplitude difference of the corresponding position of the envelope curve, the XOR value of the polarity change code and the absolute value of the power frequency phase position difference, and summing and averaging them. The electrical equipment detection data is stored and a partial discharge detection database is constructed.

[0010] Furthermore, the specific steps for link delay drift analysis using time offset, link load, and node status data are as follows: Obtain the first... The incident was in the first The following parameters are considered within each observation window: cross-channel time difference, cross-node timestamp deviation, queue fluctuation amplitude, interface buffer occupancy rate, link load rate, and clock drift rate. The absolute values ​​of the cross-channel time difference and cross-node timestamp deviation are calculated and then summed to obtain the time offset input. The square root of the time offset input is then performed to obtain the time offset adjustment term. The queue fluctuation amplitude and interface buffer occupancy rate are summed to obtain the queuing disturbance input. One is added to the queuing disturbance input, and a natural logarithmic operation is performed to obtain the queuing disturbance adjustment term. The time offset adjustment term and the queuing disturbance adjustment term are multiplied to obtain the front-end drift coupling term. The link load rate and clock drift rate are summed and the square root is then performed to obtain the link drift supplement term. Finally, the front-end drift coupling term and the link drift supplement term are summed to obtain the drift disturbance value.

[0011] Furthermore, the specific steps for performing drift interval determination and time compensation parameter adjustment based on the drift analysis results are as follows: By comparing the drift disturbance value and the disturbance threshold in real time, when the drift disturbance value is less than the disturbance threshold, the current partial discharge event is determined to be a controllable drift event, and the current observation time window is marked as a stable tracking time window and archived to the partial discharge detection database; when the drift disturbance value is greater than or equal to the disturbance threshold, it is determined to be a link drift event, and the search range of the partial discharge segment corresponding to the same event number is expanded forward from the current observation time window. A partial discharge segment arrives at the interval and extends backward. The arrival interval of partial discharge segments was adjusted; the update method of time compensation parameters was changed from updating according to the observation time window to updating according to the batch of partial discharge events; the recording method of node status data was changed from recording according to the observation time window to recording synchronously with partial discharge events; the original timestamp recording content was changed from only retaining the acquisition time and the aggregation time to simultaneously retaining the acquisition time, the transmission time, the reception time and the processing time; the content retained in the partial discharge event buffer was changed from only retaining completed splicing events to simultaneously retaining the original partial discharge segments and incomplete splicing events. The corrected cross-channel time difference and the corrected cross-node timestamp deviation were recalculated, and the arrival time difference residual was obtained by the difference between the cross-channel time differences before and after time correction.

[0012] Furthermore, based on the corrected time difference, power frequency phase, and pulse morphology data, the specific steps for evaluating the splicing reliability of the event after time compensation are as follows: obtain the first... The incident was in the first The time difference is calculated for each observation time window, including the corrected cross-channel time difference, corrected cross-node timestamp deviation, power frequency phase difference, pulse envelope similarity, and drift disturbance value. The absolute values ​​of the corrected cross-channel time difference and corrected cross-node timestamp deviation are summed and negative to obtain the time difference suppression input. An exponential operation is performed on the time difference suppression input to obtain the time difference convergence term. The pulse envelope similarity is summed with a constant to obtain the morphological support input. The square root of the morphological support input is performed to obtain the morphological support term. The time difference convergence term and the morphological support term are summed to obtain the feature synthesis term. The absolute value of the power frequency phase difference is processed to obtain the absolute phase difference term. The absolute phase difference term is summed with the drift disturbance value and a constant to obtain the disturbance synthesis term. The feature synthesis term is divided by the disturbance synthesis term to obtain the corrected confidence value.

[0013] Further, the specific steps for performing event merging correction and compensation for limited event marking based on the credibility evaluation results are as follows: By comparing the correction credibility value and the correction threshold in real time, when the correction credibility value is greater than or equal to the correction threshold, the arrival time after time correction is used to participate in the cross-channel merging of the same partial discharge event, maintaining the current event splicing time window, and outputting the corrected fused event to the partial discharge detection database; when the correction credibility value is less than the correction threshold, a candidate merging channel set is constructed based on the screening results of pulse envelope similarity greater than the envelope matching threshold and power frequency phase difference less than the phase difference threshold, and a candidate time queue is obtained by sorting the arrival time difference residuals corresponding to the corrected candidate arrival times from small to large, and a morphological support queue is obtained by sorting them from large to small according to pulse envelope similarity, and the event alignment order is reorganized according to the common channel sorting results in the candidate time queue and the morphological support queue, and the time correction amount of the channel group corresponding to the current event is updated by rolling according to the observation time window, extending the temporary storage time of the current event at the convergence node, and reading the partial discharge pulse waveform segment and power frequency phase position of the associated channel, and re-performing event splicing, phase mapping and feature correspondence; if the same channel group is in continuous If the calibration confidence value within a time window is still less than the calibration threshold, it is marked as a compensation-limited event, and the corresponding partial discharge event of the suspended channel enters the fusion stability discrimination module.

[0014] Furthermore, combining the correction results, phase correlation, and feature matching data, the specific steps for determining the stability of the fused multimodal signals are as follows: Obtain the first... The incident was in the first The following parameters are used to obtain the following data: Corrected confidence value, arrival time difference residual, cross-modal feature matching difference, and drift perturbation value within each observation time window; The phase correlation coefficient is summed with 1 to obtain the correlation enhancement input term; the square root of the correlation enhancement input term is performed to obtain the correlation enhancement term; the corrected confidence value is summed with the correlation enhancement term to obtain the stability support term; the absolute value of the arrival time difference residual is processed to obtain the absolute residual term; the absolute value of the cross-modal feature matching difference is processed to obtain the absolute matching difference term; the absolute matching difference term is incremented by 1 and then subjected to natural logarithm calculation to obtain the matching constraint term; the absolute residual term, the matching constraint term, the drift perturbation value, and 1 are summed to obtain the stability constraint term; the stability support term is divided by the stability constraint term to obtain the fused stable value.

[0015] Furthermore, the specific steps for performing cross-channel merging, pulse morphology characterization, cross-modal consistency characterization, and partial discharge defect type output based on the stability discrimination results are as follows: By comparing the fused stable value with the stability threshold in real time, when the fused stable value is greater than or equal to the stability threshold, it is determined to be a fused stable event. Cross-channel merging of the same partial discharge event is then performed based on the corrected arrival time. The phase segment is determined based on the power frequency phase position. Pulse morphology characterization is formed based on pulse amplitude, pulse width, rise slope, envelope peak value, and polarity change characteristics. Cross-modal consistency characterization is formed based on the number of consistent channels, phase correlation coefficient, and arrival time difference residual. Then, the phase segment results, pulse morphology characterization results, and cross-modal consistency characterization results are compared with those of tip discharge, suspended discharge, surface discharge, and gas discharge. The phase interval, pulse morphology characteristic threshold, and consistency index threshold corresponding to the gap discharge are compared. The number of matching items for each defect type in the three types of characterization results is counted. When all three types of characterization results match the same defect type, the partial discharge defect type is output. When the number of matching items is the largest and unique, the defect type is output and marked as pending verification. When the number of matching items is tied for the largest, the defect type matching the phase segment is output and marked as low confidence. The fused stability value, partial discharge defect type, phase correspondence result, and feature matching result are archived to the partial discharge detection database. When the fused stability value is less than the stability threshold, the current partial discharge event is marked as an unstable event, and the current partial discharge event is paused from entering the partial discharge defect type discrimination and formal trend statistical processing. Only the conclusion of partial discharge existence and risk warning are retained.

[0016] Furthermore, the specific steps for constructing time series curves by calling the partial discharge (PD) detection database records to achieve visualized comparison of link drift tracking and correction effects are as follows: The PD event markers, link drift records, arrival time difference residuals, and stability discrimination results from the PD detection database are called, and multi-layered visualization views are constructed according to device number, observation time window, event number, and channel number. Based on the time series curves, the continuous changes in drift disturbance values, correction confidence values, and fusion stability values ​​are displayed. Based on the event details view, the PD pulse waveform, power frequency phase position, event splicing results, arrival time difference residuals, and cross-modal feature matching differences are displayed. Based on the status comparison view, the changing relationships of cross-channel time difference, cross-node timestamp deviation, interface buffer occupancy rate, link load rate, and clock drift rate in different observation time windows are displayed. Based on the alarm view, the pending verification markers, risk warnings, PD defect types, and trend statistics are displayed, thus realizing PD event location, link drift tracking, time correction effect comparison, fusion result verification, and defect evolution status control.

[0017] The present invention has the following beneficial effects: (1) This invention, through comprehensive analysis of timing offset, link load and node status, realizes the adaptive adjustment of link drift interval determination and time compensation parameters, thereby achieving the effect of link delay drift dynamic evaluation and adaptive optimization of compensation parameters, effectively solving the problems of fixed compensation parameters being unable to adapt to non-fixed micro-delay drift and large cross-channel time alignment error in the prior art.

[0018] (2) This invention comprehensively evaluates the credibility of partial discharge event splicing and optimizes the event alignment order based on the correction time difference, power frequency phase and pulse morphology characteristics, thereby achieving the effect of accurate splicing of partial discharge events and quantitative evaluation of correction effect, effectively solving the problems of event splicing error, feature mismatch and inability to effectively determine correction credibility in the prior art.

[0019] (3) This invention, by combining timing correction results, phase correlation characteristics and cross-modal feature matching, determines the stability of multimodal signal fusion and accurately identifies the type of partial discharge defect, thereby achieving the effect of deep fusion of multimodal signals and accurate identification of partial discharge defect type, effectively solving the problems of poor multimodal data fusion, unstable defect identification and high false alarm and false alarm rates in the prior art.

[0020] (4) This invention constructs a multi-layered visualization view through a visualization management module, presenting the entire process of drift, correction and discrimination with time-series curves and status views, thereby achieving the effect of full-process visualization control and real-time status tracking of partial discharge detection, effectively solving the problems of opaque detection process, lack of visualization control, and difficulty in fault location and trend analysis in the prior art.

[0021] Of course, any product implementing this invention does not necessarily need to achieve all of the advantages described above at the same time. Attached Figure Description

[0022] Figure 1 This is a structural diagram of a comprehensive online detection and visualization control system for partial discharge of electrical equipment according to the present invention; Figure 2 This is a comparison chart of the time series trends and key time windows of drift disturbance values ​​of various types of sensing nodes in this invention; Figure 3 This is a schematic diagram of the partial discharge feature maps PRPS and PRPD of the present invention; Figure 4 This is a topology diagram of the link node status of the present invention. Detailed Implementation

[0023] 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.

[0024] Please see Figures 1-4 This invention provides a technical solution: a comprehensive online detection and visualization control system for partial discharge (PD) of electrical equipment, comprising: a data acquisition and preprocessing module for acquiring electrical equipment detection data, preprocessing the data, storing it, and constructing a PD detection database; a drift disturbance assessment module for performing link delay drift analysis based on time-series offset, link load, and node status data, and performing drift interval determination and time compensation parameter adjustment operations based on the drift analysis results; a dynamic correction assessment module for evaluating the splicing reliability of events after time compensation based on correction time difference, power frequency phase, and pulse morphology data, and performing event merging correction and compensation-limited event marking operations based on the reliability evaluation results; a fusion stability discrimination module for judging the stability of multi-modal signal fusion by combining correction results, phase correlation, and feature matching data, and performing cross-channel merging, pulse morphology characterization, cross-modal consistency characterization, and PD defect type output operations based on the stability discrimination results; and a visualization management module for calling the PD detection database records to construct time series curves, realizing link drift tracking and visualization comparison of correction effects.

[0025] Specifically, the steps for collecting electrical equipment testing data are as follows: Data acquisition for electrical equipment testing: Partial discharge pulse waveforms and arrival times are recorded using an UHF antenna with a working bandwidth of 300MHz–3GHz, a sampling rate of 100MSPS, and hardware timestamp synchronization. The pulse envelope sequence is extracted by calculating the instantaneous amplitude using Hilbert transform, thus obtaining UHF signal data. Partial discharge radiated pulse sequences, arrival times, and amplitude sequences are recorded using a high-frequency current transformer with a working bandwidth of 1MHz–50MHz and a sampling rate of 20MSPS. Polarity change segments are extracted using zero-crossing detection and sign-flipping determination, thus obtaining high-frequency signal data. A sampling rate of 5MHz is used within a working frequency range of 20kHz–1MHz. SPS uses piezoelectric ultrasonic sensors to record acoustic emission waveforms, acoustic signal arrival times, and sound pressure peak sequence to acquire ultrasonic signal data. A synchronous sampling unit, using a power frequency synchronous clock as a reference, records the power frequency phase position corresponding to each channel's partial discharge segment to acquire power frequency phase data. Acquisition nodes, forwarding nodes, and aggregation nodes record changes in the forwarding queue, interface buffer usage, link transmission load, and message flow times in real time during data transmission to acquire link status data. Each node records changes in its local time base and event timestamps using a unified reference clock, and the processing wait time is obtained by statistically analyzing the time difference between the message enqueue time and the processing start time to acquire node status data.

[0026] In this implementation scheme, multi-mode partial discharge signals and link and node status data are collaboratively collected by multiple types of sensing units according to preset bandwidth, sampling rate and timestamp recording method. Key feature parameters such as pulse envelope sequence, polarity change segment and processing waiting status are accurately extracted by feasible methods such as Hilbert transform, zero cross-detection, and time difference statistics. At the same time, full-dimensional data acquisition and standardization of power frequency phase, link transmission, node clock and other data are completed. This provides a complete, standardized and reusable raw data and feature foundation for subsequent drift disturbance assessment, dynamic time correction, multi-mode signal fusion discrimination and defect type identification, ensuring the data accuracy and feasibility of the whole process detection and analysis.

[0027] Specifically, the steps for preprocessing and storing electrical equipment testing data and constructing a partial discharge testing database are as follows: A unified time reference mapping is performed on the electrical equipment detection data. The acquisition times of each node are aligned to the same time reference system using GPS synchronization signals. Then, observation time windows are segmented, dividing the continuous data stream into fixed-duration analysis windows based on the power frequency cycle. Next, abnormal pulse removal is performed, identifying and removing pulses with amplitudes exceeding three standard deviations of a set threshold and pulse widths less than the minimum physical pulse width. Noise segment suppression is then performed, using wavelet threshold denoising algorithms to suppress background noise interference. Missing segments are then filled in using linear interpolation or forward padding. Finally, event number association is performed, assigning unique event identifiers to multiple channels of the same partial discharge event, forming a set of partial discharge observation segments. These segments are then normalized using a minimum-maximum method to map to a unified numerical range, achieving dimensionless processing of cross-channel data and eliminating the impact of dimensional differences on subsequent analysis. Cross-channel time is calculated using the arrival times of multiple channels corresponding to the same partial discharge event. The difference is obtained by calculating the cross-node timestamp deviation through the timestamps of multiple nodes corresponding to the same partial discharge event, the queue fluctuation amplitude through the standard deviation of the queue length fluctuation over time during the forwarding queuing process, the interface buffer occupancy rate through the ratio of the current cached data volume to the cache capacity, the link load rate through the ratio of the data volume transmitted per unit time to the link bandwidth, the clock drift rate through the deviation rate between the local clock frequency and the reference clock frequency, the power frequency phase difference through the power frequency phase position difference corresponding to different channels, the pulse envelope similarity through the Pearson correlation coefficient of the pulse envelope sequence of partial discharge segments of different channels, and the cross-modal feature matching difference through the summation and averaging of the Euclidean distance accumulation value of the corresponding sampling points of the amplitude sequence, the normalized amplitude difference accumulation value of the corresponding position of the envelope curve, the XOR value of the polarity change code and the absolute value of the power frequency phase position difference; the electrical equipment detection data is stored and a partial discharge detection database is constructed.

[0028] In this implementation scheme, by performing unified time reference mapping, observation time window segmentation, abnormal pulse removal, noise segment suppression, missing segment completion, and event number association, the original multi-source heterogeneous data are integrated into a set of partial discharge observation segments with a unified structure. Minimum-maximum value normalization is used to achieve dimensionless processing, eliminating the impact of dimensional differences between different detection channels on subsequent analysis. Based on this, by extracting feature parameters such as cross-channel time difference, cross-node timestamp deviation, queue fluctuation amplitude, interface buffer occupancy rate, link load rate, clock drift rate, power frequency phase difference, and pulse envelope similarity, and by summing and averaging the cross-modal feature matching difference obtained from the sum of the amplitude sequence Euclidean distance accumulation, the envelope curve normalized amplitude difference accumulation, the polarity change encoding XOR value, and the absolute value of the power frequency phase position difference, a quantitative characterization of the consistency between link status, node status, and cross-modal signals is achieved. This provides a unified data foundation and feature input for subsequent drift disturbance assessment, dynamic correction, and fusion stability discrimination.

[0029] Specifically, the steps for link delay drift analysis using time offset, link load, and node status data are as follows: The following parameters are obtained for the i-th partial discharge event within the t-th observation time window: cross-channel time difference, cross-node timestamp deviation, queue fluctuation amplitude, interface buffer occupancy rate, link load rate, and clock drift rate. These parameters have been normalized and dimensionless using the minimum-maximum method to unify the numerical range and eliminate calculation differences caused by different physical dimensions. The absolute values ​​of the cross-channel time difference and cross-node timestamp deviation are summed to obtain the time offset input. The square root of the time offset input is used to obtain the time offset adjustment term. The queue fluctuation amplitude and interface buffer occupancy rate are summed to obtain the queuing disturbance input. The queuing disturbance input is incremented by one and then subjected to a natural logarithm operation to obtain the queuing disturbance adjustment term. The time offset adjustment term and the queuing disturbance adjustment term are multiplied to obtain the front-end drift coupling term. The link load rate and clock drift rate are summed and the square root is used to obtain the link drift supplement term. The front-end drift coupling term and the link drift supplement term are summed to finally obtain the drift disturbance value calculated based on the normalized dimensionless input.

[0030] The specific formula for calculating the drift disturbance value is as follows: ; In the formula, Indicates the first The incident was in the first The drift disturbance values ​​in each observation time window are used to characterize the degree of impact of cross-channel arrival offset and link fluctuations on event alignment. Indicates the cross-channel time difference, used to reflect cross-channel time alignment error; Indicates the cross-node timestamp deviation, used to reflect the drift of the cross-node time base; This indicates the amplitude of queue fluctuations, used to reflect the amplification of micro-delay differences during the queuing process; This indicates the interface buffer occupancy rate, representing the level of buffer resource usage when an event arrives at the corresponding interface. This indicates the link load rate, representing the transmission load intensity of the link where the event occurs, and is used to reflect the impact of link congestion on transmission delay drift. This represents the clock drift rate, indicating how quickly the node's local time base changes relative to the reference time base.

[0031] Table 1 shows the multi-dimensional timing synchronization performance test data in this embodiment. The first sample has a cross-channel time difference of 22, a cross-node timestamp deviation of 18, a queue fluctuation amplitude of 0.12, an interface buffer occupancy rate of 0.25, a link load rate of 0.30, and a clock drift rate of 0.8, resulting in a calculated drift disturbance value of 1.75. The second sample has a cross-channel time difference of 86, a cross-node timestamp deviation of 74, a queue fluctuation amplitude of 0.35, an interface buffer occupancy rate of 0.48, a link load rate of 0.55, and a clock drift rate of 2.1, resulting in a calculated drift disturbance value of 2.68. The third... The cross-channel time difference of the first sample is 124, the cross-node timestamp deviation is 112, the queue fluctuation amplitude is 0.46, the interface buffer occupancy rate is 0.59, the link load rate is 0.68, and the clock drift rate is 2.7. The calculated drift disturbance value is 3.02. The cross-channel time difference of the fourth sample is 156, the cross-node timestamp deviation is 138, the queue fluctuation amplitude is 0.52, the interface buffer occupancy rate is 0.66, the link load rate is 0.74, and the clock drift rate is 3.2. The calculated drift disturbance value is 3.26.

[0032] Table 1. Multi-dimensional Timing Synchronization Performance Test Data Table like Figure 2 As shown, this is a comparison chart of the time series trends and key time windows of the drift disturbance values ​​of various types of sensor nodes provided in the embodiments of this application. The observation time window refers to the time interval for continuous monitoring of the node drift state; the key observation time windows are four typical monitoring times of 15, 30, 45, and 55 selected. The aggregation node is the core node in the network responsible for data aggregation, the forwarding node is the intermediate node responsible for data relay transmission, and the acquisition node is the front-end raw data acquisition terminal. (Referring to Table 1 and...) Figure 2As can be seen from the comparison of the four sets of data in Table 1, the parameters of sample 1 are all at a low level, with a drift disturbance value of only 1.75. This corresponds to the drift disturbance value of the aggregation node in the figure being below the disturbance threshold throughout, showing an extremely low disturbance state. This indicates that the clock synchronization of the aggregation node is excellent, data transmission and buffer scheduling are stable, and there is no risk of drift exceeding the standard. The parameters of samples 2 and 3 gradually increase, with drift disturbance values ​​rising to 2.68 and 3.02 respectively. This corresponds to the medium disturbance characteristics of the forwarding node in the figure, with occasional instances of exceeding the disturbance threshold, reflecting the accumulation of deviations in the data relay process of the forwarding node. The parameters of sample 4 reach their peak, with a drift disturbance value as high as 3.26. This corresponds to the high disturbance characteristics of the acquisition node in the figure, continuously exceeding the disturbance threshold, which is completely consistent with the high fluctuation of the real scenario of direct data transmission from the field acquisition end. The left figure clearly shows the continuous changing trend of drift disturbance values ​​for the three types of nodes: aggregation, forwarding, and acquisition, with the disturbance intensity increasing sequentially. The right figure intuitively and quantitatively compares the differences in disturbance values ​​of the three types of nodes under different key observation time windows, accurately highlighting the strong correlation between node functional positioning and drift disturbance intensity. It fully verifies the coupling correction effect of drift disturbance values ​​with multiple parameters such as cross-channel time difference, timestamp deviation, queue fluctuation amplitude, and buffer occupancy rate: node type dominates the basic level of disturbance, link load and clock drift rate amplify the disturbance response, and queue fluctuation and buffer occupancy determine the disturbance fluctuation amplitude. This provides intuitive data support and scientific basis for the hierarchical identification and dynamic calibration of clock drift in sensor networks, effectively ensuring the clock stability and data reliability of multi-node collaborative transmission.

[0033] In this implementation scheme, three core influencing factors—time offset, queuing disturbance, and link drift—are decomposed and coupled stepwise to generate time offset adjustment terms, queuing disturbance adjustment terms, front-end drift coupling terms, and link drift supplementary terms. Finally, the drift disturbance value is obtained by weighted summation. This can comprehensively and accurately quantify the degree of combined disturbance of the i-th partial discharge event within the t-th observation time window, under the combined effects of time offset, queue fluctuation, link load change, and node clock drift. The unified dimensionless processing effectively eliminates the calculation bias caused by the difference in the dimensions of different physical parameters, ensuring that the evaluation results are objective and stable. This provides a key, unified, and quantifiable basis for subsequent link drift interval determination, dynamic adjustment of time compensation parameters, and optimization of partial discharge event alignment strategies.

[0034] Specifically, the steps for determining the drift interval and adjusting the time compensation parameters based on the drift analysis results are as follows: Drift status is determined by comparing the drift disturbance value with a preset disturbance threshold in real time. The disturbance threshold is a quantitative determination of whether the link drift is abnormal. When the drift disturbance value is less than the disturbance threshold, the current partial discharge event is determined to be a controllable drift event, and the current observation time window is marked as a stable tracking time window and archived to the partial discharge detection database. When the drift disturbance value is greater than or equal to the disturbance threshold, it is determined to be a link drift event. Here, f is the forward expansion step size, which takes the value of 2 to 5 partial discharge segment arrival intervals, and b is the backward expansion step size, which takes the value of 3 to 6 partial discharge segment arrival intervals. The search range of partial discharge segments corresponding to the same event number is expanded forward by f partial discharge segment arrival intervals from the current observation time window and backward by b partial discharge segment arrival intervals. The time compensation parameter specifically refers to the time offset Δt used to correct cross-channel timing deviations and the time offset Δt used to correct cross-channel timing deviations. The drift rate α of the corrected node clock drift is adjusted, where the time offset Δt ranges from -10μs to +10μs and the drift rate α ranges from 0ppm to 50ppm. The update method of the time compensation parameter is changed from updating according to the observation time window to updating according to the partial discharge event batch. The node status data recording method is changed from recording according to the observation time window to recording synchronously with the partial discharge event. The original timestamp recording content is changed from only retaining the acquisition time and the aggregation time to simultaneously retaining the acquisition time, transmission time, reception time and processing time. The content retained in the partial discharge event buffer is changed from only retaining the completed splicing event to simultaneously retaining the original partial discharge segment and the incomplete splicing event. The corrected cross-channel time difference and the corrected cross-node timestamp deviation are recalculated, and the arrival time difference residual is obtained by the difference between the cross-channel time difference before and after time correction.

[0035] In this implementation plan, by comparing the drift disturbance value with the preset disturbance threshold in real time, the drift status of partial discharge events is accurately classified and determined. For controllable drift events with a drift disturbance value less than the disturbance threshold, a stable tracking time window is marked and archived to the partial discharge detection database to ensure the traceability of detection data in a stable state. For link drift events with a drift disturbance value that meets the standard, the update method of time compensation parameters, the recording method of node status data, the content of original timestamp records, and the retention rules of the partial discharge event buffer are adjusted simultaneously. By recalculating, the corrected cross-channel time difference and the corrected cross-node timestamp deviation are obtained, and the arrival time difference residual before and after time correction is extracted. This effectively addresses the timing deviation problem caused by link drift, optimizes the accuracy of timing correction, standardizes the processing flow of drift events, ensures the integrity and timing accuracy of detection data, and provides reliable timing data support for subsequent partial discharge event splicing, cross-modal feature characterization, and defect type identification, avoiding interference from link drift on the overall detection accuracy.

[0036] Specifically, based on the corrected time difference, power frequency phase, and pulse morphology data, the specific steps for evaluating the splicing reliability of events after time compensation are as follows: Obtain the corrected cross-channel time difference, corrected cross-node timestamp deviation, power frequency phase difference, pulse envelope similarity, and drift disturbance value of the i-th partial discharge event in the t-th observation time window; sum the absolute values ​​of the corrected cross-channel time difference and the corrected cross-node timestamp deviation and take the negative value to obtain the time difference suppression input term. This input term reflects the remaining time alignment error after time compensation; the more negative the value, the larger the remaining time difference; perform an exponential operation on the time difference suppression input term to obtain the time difference convergence term. Convert the remaining time difference into a convergence factor between 0 and 1 through the nonlinear mapping of the exponential function. When the time difference approaches zero, the convergence term approaches 1; when the time difference increases, the convergence term rapidly decays; sum the pulse envelope similarity with a constant to obtain the morphology support input term. This input term ensures that the basic structure is retained even when the pulse envelope similarity is zero. Support quantity; the square root operation is performed on the morphological support input to obtain the morphological support term. The square root processing compresses the gain change in the high similarity region, so that the morphological support term exhibits asymptotic saturation characteristics when the similarity changes from 0 to 1; the time difference convergence term and the morphological support term are summed to obtain the feature synthesis term, which integrates the positive contributions of time alignment accuracy and waveform morphological consistency; the absolute value processing is performed on the power frequency phase difference to obtain the phase difference absolute term; the phase difference absolute term is summed with the drift disturbance value to obtain the disturbance synthesis term, which integrates the negative interference of phase offset degree and link drift intensity; the feature synthesis term is divided by the disturbance synthesis term to obtain the correction confidence value. This value represents the overall confidence level of the current event after time compensation to enter the subsequent fusion and discrimination process. The higher the value, the smaller the time alignment error, the higher the morphological matching degree, and the weaker the phase offset and link drift interference. Conversely, the lower the value, the more insufficient the compensation effect or the stronger the interference.

[0037] The specific formula for calculating the calibration confidence value is as follows: ; In the formula, Indicates the first The incident was in the first The corrected confidence value in each observation time window is used to characterize whether the current event, after compensation processing, has the basis to enter the subsequent fusion and discrimination process; This indicates the correction of cross-channel time difference, used to reflect the degree of remaining time offset between channels after the time compensation parameter is updated for the current partial discharge event; This indicates the correction of cross-node timestamp deviation, used to reflect the degree of inconsistency between the remaining timestamps of each node after the time compensation parameter is updated for the current partial discharge event; It represents the power frequency phase difference, indicating the degree of offset in the power frequency phase position of the same partial discharge event across multiple channels; Indicates pulse envelope similarity, used to reflect whether cross-modal segments have a basis for homologous splicing; This represents the drift disturbance value, indicating the strength of the disturbance caused by the combined effects of link delay drift and node time drift on the current event.

[0038] In this implementation scheme, compared to traditional methods that rely solely on time difference for event stitching, the system introduces a correction reliability value that integrates four types of features: time difference convergence term, morphological support term, phase suppression term, and drift disturbance term. This enables a comprehensive quantitative assessment of the reliability of compensated events. Traditional methods determine event attribution based solely on a single time difference threshold, which can easily lead to mis-stitching or omissions when link drift causes time difference fluctuations. In contrast, this system effectively compensates for the lack of sufficient time difference information through morphological support provided by pulse envelope similarity, avoids erroneous merging caused by phase mismatch through phase constraints provided by power frequency phase difference, and further suppresses unreliable stitching under high disturbance conditions through link status feedback provided by drift disturbance value. In this embodiment, the event mis-segmentation rate using the single arrival time difference threshold method is approximately 12.7%, while the event mis-segmentation rate using the time difference plus phase dual constraint method is approximately 7.4%. After event merging based on the correction confidence value, the mis-segmentation rate is reduced to 3.1%, which is about 75% lower than the traditional method. This effectively solves the event alignment error problem caused by the non-fixed micro-delay drift of the link in multi-module partial discharge detection.

[0039] Specifically, the steps for performing event merging correction and compensation for restricted event labeling based on the credibility evaluation results are as follows: Confidence is determined by comparing the correction confidence value and the correction threshold in real time. The correction threshold is a preset quantitative value that characterizes whether the event splicing correction effect meets the standard. When the correction confidence value is greater than or equal to the correction threshold, the arrival time after time correction is used to participate in the cross-channel merging of the same partial discharge event, maintaining the current event splicing time window, and the corrected fused event is archived to the partial discharge detection database. When the correction confidence value is less than the correction threshold, a candidate merging channel set is constructed based on the screening results of pulse envelope similarity greater than the envelope matching threshold and power frequency phase difference less than the phase difference threshold. The candidate time queue is obtained by sorting the arrival time difference residuals corresponding to the corrected candidate arrival times from smallest to largest, and the morphological support queue is obtained by sorting the pulse envelope similarity from largest to smallest. A two-level sorting rule is used, with the arrival time difference residual sorting as the primary rule and the pulse envelope similarity sorting as the secondary rule, to sort the two queues. The common channels in the column are integrated and sorted. Based on the integrated sorting results, the event alignment order is reorganized, and the time correction of the channel group corresponding to the current event is updated on a rolling basis according to the observation time window. The temporary storage time of the current event at the convergence node is extended, and the partial discharge pulse waveform segment and power frequency phase position of the associated channel are read. Event splicing, phase mapping and feature correspondence are re-performed. Here, n is the number of consecutive observation time window judgments, which ranges from 3 to 5. If the correction confidence value of the same channel group is still less than the correction threshold in n consecutive time windows, it is marked as a compensation-limited event. The partial discharge event corresponding to the channel is paused and enters the fusion stability discrimination module. The pause time is the current n consecutive observation time window period. When the correction confidence value of the channel group rises to greater than or equal to the correction threshold in subsequent observation time windows, the partial discharge event corresponding to the channel is restored and enters the fusion stability discrimination module.

[0040] This implementation plan focuses on real-time comparison between the correction confidence value and the correction threshold to determine the credibility of partial discharge event correction and implement targeted control. When the correction confidence value meets the standard, the arrival time after correction is used to complete cross-channel merging and archive the fusion events, ensuring that the events with effective correction can stably enter the subsequent process. When the correction confidence value does not meet the standard, a candidate merging channel set is constructed through preset screening conditions. A two-level sorting rule is adopted, with arrival time difference residual as the main factor and pulse envelope similarity as the secondary factor. The common channel sorting results of the candidate time queue and the morphological support queue are integrated to accurately reorganize the event alignment order. At the same time, the time correction amount is updated continuously, the event temporary storage time is extended, and the event splicing and feature mapping are re-completed to ensure the accuracy of time sequence alignment. By defining the number of judgments n in the continuous observation time window, the channel group whose correction confidence value still does not meet the standard after n consecutive time windows is marked as a compensation-limited event and its entry into the fusion stability discrimination module is suspended. This provides high-quality and high-confidence post-correction event data support for subsequent multimodal signal fusion stability discrimination and accurate identification of defect types, ensuring the stability and accuracy of the entire detection process.

[0041] Specifically, the steps for determining the stability of multimodal signal fusion by combining correction results, phase correlation, and feature matching data are as follows: The corrected confidence value, arrival time difference residual, cross-modal feature matching difference, and drift disturbance value of the i-th partial discharge event in the t-th observation time window are obtained. All input parameters are first normalized to eliminate the dimensional differences of different physical quantities, ensuring the accuracy and comparability of subsequent calculation results. The phase sequence of the multi-module partial discharge detection signal within the power frequency cycle is extracted, and the Pearson product-moment correlation function is calculated to obtain the phase correlation coefficient, which ranges from -1 to 1. Then, the phase correlation coefficient is summed with a constant to obtain the correlation enhancement input term. The square root of the correlation enhancement input term is then performed to obtain the correlation enhancement term, thereby strengthening the correlation of phase features. The process involves several steps: First, the time difference is adjusted. Then, the corrected confidence value is summed with the relevant enhancement terms to obtain a stability support term, which reflects the reliability of the partial discharge event after correction. Next, the arrival time difference residual is processed to obtain an absolute residual term, used to quantify the magnitude of the timing deviation. Finally, the cross-modal feature matching difference is processed to obtain a matching difference absolute term, accurately reflecting the degree of deviation in feature matching. A constant is added to the matching difference absolute term, and then a natural logarithm is performed to obtain a matching constraint term, used to balance the influence weight of feature matching. Finally, the residual absolute term, the matching constraint term, and the drift perturbation value are summed, and a constant is added to avoid computational anomalies, resulting in a stability constraint term. The physical meaning of the stability constraint term is as follows: the larger the arrival time difference residual, the larger the cross-modal feature matching difference, and the larger the drift perturbation value, the larger the value of the stability constraint term, representing a lower temporal stability of the current partial discharge event, and vice versa. This stability constraint term allows for a direct assessment of the temporal reliability of the partial discharge event, providing accurate quantitative basis for subsequent event splicing, feature matching, and defect identification, ensuring the stability and accuracy of the entire detection process.

[0042] The specific formula for calculating the fusion stability value is as follows: ; In the formula, Indicates the first The incident was in the first The fused stability value within each observation time window is used to determine whether the current event meets the conditions for entering the formal defect type identification, trend statistics, and discharge intensity archiving process; The value indicates the level of confidence of the current event after timeline rewinding and cross-modal splicing. The phase correlation coefficient represents the degree of synchronization correlation between different channel partial discharge segments in the power frequency phase dimension. This represents the arrival time difference residual, used to reflect the remaining time alignment error after compensation; The cross-modal feature matching difference represents the degree of difference between the compensated fused events in feature dimensions such as amplitude structure, envelope shape, polarity change, and phase position, and is used to reflect the consistency of the cross-modal merging results. This represents the drift disturbance value, indicating the strength of the drift disturbance on the link where the current event occurs.

[0043] This implementation plan clarifies the quantitative standards of core indicators such as time compensation parameters and correlation coefficients, and distinguishes the partial discharge event states under different scenarios by combining scientific judgment logic. It standardizes the specific processes of event splicing and feature matching, ensuring the consistency and accuracy of data from each channel, and providing accurate and reliable quantitative basis for subsequent multimodal signal fusion and partial discharge defect identification. At the same time, through clear parameter ranges and judgment rules, it effectively avoids timing deviations and matching errors, ensuring the standardization and repeatability of the detection process. This lays a solid foundation for subsequent visual control and accurate defect judgment, ensuring that the entire detection process is scientific and controllable, improving the accuracy and practicality of partial discharge detection, and meeting the feasibility requirements of the patented technical solution.

[0044] Specifically, the steps for performing cross-channel merging, pulse morphology characterization, cross-modal consistency characterization, and partial discharge defect type output based on the stability discrimination results are as follows: The fusion status is determined by comparing the fusion stability value with a preset stability threshold in real time. The stability threshold is a preset quantization value that characterizes whether the fusion effect of multimodal partial discharge signals meets the defect discrimination conditions. When the fusion stability value is greater than or equal to the stability threshold, it is determined to be a fusion stability event. Cross-channel merging of the same partial discharge event is performed based on the corrected arrival time. The corresponding phase segment is determined based on the power frequency phase position. The pulse morphology characterization is formed by comprehensively considering the pulse amplitude, pulse width, rise slope, envelope peak value, and polarity change characteristics. The cross-modal consistency characterization is formed by comprehensively considering the number of consistent channels, phase correlation coefficient, and arrival time difference residual. Then, the phase segment results, pulse morphology characterization results, and cross-modal consistency characterization results are compared item by item with the preset phase intervals, pulse morphology characteristic thresholds, and consistency index thresholds corresponding to tip discharge, suspended discharge, surface discharge, and air gap discharge. The number of matching items for each defect type in the three types of characterization results is counted. When all three types of characterization results match the same defect type, the partial discharge defect type is output. Figure 3The diagram shows the partial discharge feature maps PRPS and PRPD of this embodiment, which can intuitively distinguish the phase and amplitude distribution characteristics of different types of partial discharge defects such as tip discharge and floating discharge. When the number of matching items is the largest and unique, the defect type is output and marked as pending verification. When the number of matching items is the largest in a row, the defect type of the phase segment matching is output and marked as low confidence. The fused stable value, partial discharge defect type, phase correspondence result and feature matching result are uniformly archived into the partial discharge detection database to provide data support for subsequent data query, status tracing and trend analysis. When the fused stable value is less than the stable threshold, the current partial discharge event is marked as an unstable event, and the current partial discharge event is suspended from entering the partial discharge defect type discrimination and formal trend statistical processing stage. Only the conclusion of partial discharge existence and the corresponding risk warning information are retained to avoid low reliability data from interfering with the defect identification results and trend statistical accuracy.

[0045] In this implementation scheme, by comparing the fusion stability value with the stability threshold in real time, the fusion effect of multimodal partial discharge signals is graded and judged. For partial discharge events that meet the stability conditions, cross-channel merging, phase segment division, pulse morphology and cross-modal consistency characterization are completed. Four types of typical partial discharge defects are graded and the results are output. At the same time, the relevant discrimination data are archived into the database to provide complete support for subsequent traceability and trend analysis. For unstable events that fail to meet the fusion standards, the defect type discrimination and formal trend statistics are suspended. Only the conclusion of partial discharge existence and risk warning are retained, which effectively avoids low reliability data from interfering with the discrimination accuracy and improves the accuracy, rigor and stability of partial discharge defect identification and system operation.

[0046] Specifically, the steps for constructing time series curves from the partial discharge (PD) detection database to visualize and compare link drift tracking and correction effects are as follows: The database contains all detection data, including PD event markers, link drift records, time difference of arrival (TDOA) residuals, and stability judgment results. A multi-layered visualization view is constructed based on device number, observation time window, event number, and channel number. The time series curves display the continuous changes in drift disturbance values, correction confidence values, and fusion stability values. Key indicators for link drift tracking include drift disturbance values, cross-channel time difference, cross-node timestamp deviation, link load rate, and clock drift rate. Key indicators for time correction effect comparison include correction confidence values, TDOA residuals, and fusion stability values. All these indicators are derived from the corresponding stored fields in the PD detection database, providing a clear and intuitive comparison. The system reflects the timing correction and link status change patterns. Based on the event details view, it displays the partial discharge pulse waveform, power frequency phase position, event splicing results, arrival time difference residuals, and cross-modal feature matching differences. Based on the status comparison view, it displays the changes in cross-channel time difference, cross-node timestamp deviation, interface buffer occupancy rate, link load rate, and clock drift rate across different observation time windows. Based on the alarm view, it displays pending review markers, risk warnings, partial discharge defect types, and trend statistics. This enables partial discharge event localization, link drift tracking, time correction effect comparison, fusion result verification, and defect evolution status control. The manual clarifies the one-to-one correspondence between the content displayed in each view and the fields stored in the partial discharge detection database by drawing visualization interface examples and compiling data field mapping tables, providing sufficient data support for the implementation of multi-layer visualization views. Figure 4 This is a topology diagram of the link nodes in this embodiment, showing the acquisition nodes, forwarding nodes, and aggregation nodes in the partial discharge online detection system and their connections in a directed topology format. Circular nodes represent acquisition nodes, square nodes represent forwarding nodes, and diamond nodes represent aggregation nodes. The node fill color gradients from green to red, mapping the clock drift rate; the redder the color, the more severe the drift of the node's local time base. The directed edges gradient from yellow to red, representing the link load rate; the higher the load rate, the redder the color and the thicker the line. The percentage value marked next to the edge is the real-time link load rate. Node labels are offset to avoid overlapping with the graphic, used for quickly locating high-drift nodes and high-load links, supporting drift disturbance assessment and dynamic correction.

[0047] This implementation scheme utilizes the full dataset stored in the partial discharge detection database, including partial discharge event markers, link drift records, arrival time difference residuals, and stability judgment results. A multi-layered visualization view is constructed based on device number, observation time window, event number, and channel number. This clearly defines the core indicators for comparing link drift tracking and time correction effects, as well as the corresponding database data source fields. Supplementary interface examples and data field mapping tables provide ample implementation support. Through different views such as time series curves, event details, status comparisons, and alarms, the scheme presents continuous changes in various core parameters, complete event information, and multiple indicators. The status comparison and alarm-related content intuitively reflect the changing patterns of key processes such as drift disturbance, time correction, and fusion stabilization. It clearly presents the status of partial discharge events and the operation of the link, effectively realizing comprehensive control over partial discharge event location, link drift tracking, time correction effect comparison, fusion result verification, and defect evolution status. It provides convenient and clear visualization support for staff to intuitively grasp the entire detection process, troubleshoot abnormalities, verify judgment results, and dynamically control defect evolution. It significantly improves the operability, traceability, and control efficiency of partial discharge detection, ensuring that the detection process and results are intuitive, controllable, and verifiable.

[0048] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0049] The preferred embodiments of the present invention disclosed above are merely illustrative of the invention. These preferred embodiments do not exhaustively describe all details, nor do they limit the invention to the specific implementations described. Clearly, many modifications and variations can be made based on the content of this specification. This specification selects and specifically describes these embodiments to better explain the principles and practical applications of the invention, thereby enabling those skilled in the art to better understand and utilize the invention. The invention is limited only by the claims and their full scope and equivalents.

Claims

1. A comprehensive online detection and visual control system for partial discharge of electrical equipment, characterized in that, include: The data acquisition and preprocessing module is used to acquire electrical equipment test data, preprocess the electrical equipment test data, store it, and build a partial discharge test database. The drift disturbance assessment module is used to perform link delay drift analysis using time offset, link load and node status data, and to perform drift interval determination and time compensation parameter adjustment operations based on the drift analysis results; The dynamic correction and evaluation module is used to evaluate the credibility of splicing events after time compensation based on correction time difference, power frequency phase and pulse morphology data, and to perform event merging correction and compensation limited event marking operations according to the credibility evaluation results. The fusion stability discrimination module is used to combine the correction results, phase correlation and feature matching data to determine the stability of the multimodal signal after fusion, and to perform cross-channel merging, pulse morphology characterization, cross-modal consistency characterization and partial discharge defect type output operations based on the stability discrimination results; The visualization management module is used to call the partial discharge detection database records to construct time series curves, enabling visualization and comparison of link drift tracking and correction effects.

2. The integrated online detection and visual control system for partial discharge of electrical equipment according to claim 1, characterized in that: The specific steps for collecting electrical equipment test data are as follows: Data acquisition for electrical equipment testing includes: obtaining UHF signal data by recording partial discharge pulse waveforms, pulse arrival times, and pulse envelope sequences using an UHF antenna; obtaining high-frequency signal data by recording partial discharge radiation pulse sequences, pulse arrival times, and pulse amplitude sequences using a high-frequency current transformer; obtaining ultrasonic signal data by recording acoustic emission waveforms, acoustic signal arrival times, and sound pressure peak value sequences using a piezoelectric ultrasonic sensor; obtaining power frequency phase data by recording the power frequency phase position corresponding to each channel's partial discharge segment using a synchronous sampling unit; obtaining link status data by recording changes in the forwarding queue, interface buffer occupancy, link transmission load, and message flow times using acquisition nodes, forwarding nodes, and aggregation nodes; and obtaining node status data by recording changes in the local time base, event timestamp recording results, and processing waiting status at each node.

3. The integrated online detection and visual control system for partial discharge of electrical equipment according to claim 1, characterized in that: The specific steps for preprocessing and storing electrical equipment test data and constructing a partial discharge test database are as follows: The electrical equipment detection data is subjected to unified time reference mapping, observation time window segmentation, abnormal pulse rejection, noise segment suppression, missing segment completion and event number association to form a partial discharge observation segment set, and the minimum and maximum value method is used for normalization processing and mapping to a unified numerical range. Cross-channel time difference is calculated by multi-channel arrival times corresponding to the same partial discharge event; cross-node timestamp deviation is calculated by multi-node timestamp records corresponding to the same partial discharge event; queue fluctuation amplitude is obtained by waiting changes during forwarding queuing; interface buffer occupancy rate is obtained by interface buffer occupancy; link load rate is obtained by data carrying in the transmission path where the event is located; clock drift rate is obtained by the ratio of node local time base to reference time base; power frequency phase difference is obtained by the power frequency phase position difference corresponding to different channels; pulse envelope similarity is obtained by the pulse envelope sequence correspondence of partial discharge segments in different channels; and cross-modal feature matching difference is obtained by calculating the cumulative Euclidean distance of sampling points corresponding to amplitude sequences, the cumulative normalized amplitude difference of corresponding positions of envelope curves, the XOR value of polarity change coding and the absolute value of power frequency phase position difference, and summing and averaging them. The electrical equipment detection data is stored and a partial discharge detection database is constructed.

4. The integrated online detection and visual control system for partial discharge of electrical equipment according to claim 1, characterized in that: The specific steps for link delay drift analysis using time offset, link load, and node status data are as follows: Get the The incident was in the first Cross-channel time difference, cross-node timestamp deviation, queue fluctuation amplitude, interface buffer occupancy rate, link load rate, and clock drift rate within each observation time window; The absolute values ​​of the cross-channel time difference and the cross-node timestamp deviation are calculated separately and then summed to obtain the time offset input. The square root of the time offset input is then performed to obtain the time offset adjustment. The queue fluctuation amplitude is summed with the interface buffer occupancy rate to obtain the queue disturbance input. Add one to the queuing disturbance input term and perform a natural logarithm operation to obtain the queuing disturbance adjustment term; multiply the time offset adjustment term by the queuing disturbance adjustment term to obtain the front-end drift coupling term; The link load rate and clock drift rate are summed and the square root is taken to obtain the link drift supplement term; the front-end drift coupling term and the link drift supplement term are summed to obtain the drift disturbance value.

5. The integrated online detection and visualization control system for partial discharge of electrical equipment according to claim 1, characterized in that: The specific steps for performing drift interval determination and time compensation parameter adjustment based on drift analysis results are as follows: By comparing the drift disturbance value and the disturbance threshold in real time, when the drift disturbance value is less than the disturbance threshold, the current partial discharge event is determined to be a controllable drift event, and the current observation time window is marked as a stable tracking time window and archived to the partial discharge detection database. When the drift disturbance value is greater than or equal to the disturbance threshold, it is determined to be a link drift event, and the search range of the partial discharge segment corresponding to the same event number is expanded backward from the current observation time window. A partial discharge segment arrives at the interval and extends backward. The arrival interval of partial discharge segments was adjusted; the update method of time compensation parameters was changed from updating according to the observation time window to updating according to the batch of partial discharge events; the recording method of node status data was changed from recording according to the observation time window to recording synchronously with partial discharge events; the original timestamp recording content was changed from only retaining the acquisition time and the aggregation time to simultaneously retaining the acquisition time, the transmission time, the reception time and the processing time; the content retained in the partial discharge event buffer was changed from only retaining completed splicing events to simultaneously retaining the original partial discharge segments and incomplete splicing events. The corrected cross-channel time difference and the corrected cross-node timestamp deviation were recalculated, and the arrival time difference residual was obtained by the difference between the cross-channel time differences before and after time correction.

6. The integrated online detection and visual control system for partial discharge of electrical equipment according to claim 1, characterized in that: The specific steps for evaluating the splicing reliability of events after time compensation based on corrected time difference, power frequency phase, and pulse morphology data are as follows: Get the The incident was in the first Corrected cross-channel time difference, corrected cross-node timestamp deviation, power frequency phase difference, pulse envelope similarity, and drift disturbance value in each observation time window; The absolute values ​​of the corrected cross-channel time difference and the corrected cross-node timestamp deviation are summed and negative to obtain the time difference suppression input term; the time difference suppression input term is then subjected to exponential operation to obtain the time difference convergence term. The pulse envelope similarity is summed with a constant to obtain the morphological support input term; The morphological support term is obtained by taking the square root of the morphological support input term; the feature synthesis term is obtained by summing the time difference convergence term and the morphological support term. The absolute value of the power frequency phase difference is processed to obtain the absolute phase difference term; the absolute phase difference term is summed with the drift disturbance value to obtain the disturbance synthesis term; the characteristic synthesis term is divided by the disturbance synthesis term to obtain the correction confidence value.

7. The integrated online detection and visual control system for partial discharge of electrical equipment according to claim 1, characterized in that: The specific steps for performing event merging correction and compensation for restricted event marking based on the credibility evaluation results are as follows: By comparing the correction confidence value and the correction threshold in real time, when the correction confidence value is greater than or equal to the correction threshold, the arrival time after time correction is used to participate in the cross-channel merging of the same partial discharge event, the current event splicing time window is maintained, and the corrected fused event is output and archived to the partial discharge detection database. When the correction confidence value is less than the correction threshold, a candidate merged channel set is constructed based on the screening results of pulse envelope similarity greater than the envelope matching threshold and power frequency phase difference less than the phase difference threshold. The candidate time queue is obtained by sorting the arrival time difference residuals corresponding to the corrected candidate arrival times from small to large. The morphological support queue is obtained by sorting the pulse envelope similarity from large to small. The event alignment order is reorganized according to the common channel sorting results in the candidate time queue and the morphological support queue. The time correction amount of the channel group corresponding to the current event is updated by rolling according to the observation time window. The temporary storage time of the current event at the convergence node is extended. The partial discharge pulse waveform segment and power frequency phase position of the associated channel are read, and the event splicing, phase mapping and feature correspondence are re-performed. If the same channel group is consecutive If the calibration confidence value within a time window is still less than the calibration threshold, it is marked as a compensation-limited event, and the corresponding partial discharge event of the suspended channel enters the fusion stability discrimination module.

8. The integrated online detection and visual control system for partial discharge of electrical equipment according to claim 1, characterized in that: The specific steps for determining the stability of multimodal signal fusion by combining correction results, phase correlation, and feature matching data are as follows: Get the The incident was in the first Corrected confidence value, arrival time difference residual, cross-modal feature matching difference, and drift perturbation value in each observation time window; Summing the phase correlation coefficient with 1 yields the correlation enhancement input term; The relevant enhancement terms are obtained by taking the square root of the relevant enhancement input terms. The corrected confidence value is summed with the relevant enhancement term to obtain the stable support term; the absolute value of the arrival time difference residual is processed to obtain the residual absolute term. The absolute value of the cross-modal feature matching difference is processed to obtain the absolute term of the matching difference; the absolute term of the matching difference is incremented by one and then the natural logarithm is performed to obtain the matching constraint term; the absolute term of the residual, the matching constraint term, the drift perturbation value and one are summed to obtain the stability constraint term; the stability support term is divided by the stability constraint term to obtain the fusion stability value.

9. The integrated online detection and visual control system for partial discharge of electrical equipment according to claim 1, characterized in that: The specific steps for performing cross-channel merging, pulse morphology characterization, cross-modal consistency characterization, and partial discharge defect type output based on the stability discrimination result are as follows: By comparing the fusion stability value with the stability threshold in real time, when the fusion stability value is greater than or equal to the stability threshold, it is determined to be a fusion stable event. Cross-channel merging of the same partial discharge event is then performed based on the corrected arrival time. The phase segment is determined based on the power frequency phase position. A pulse morphology characterization is formed based on pulse amplitude, pulse width, rise edge slope, envelope peak value, and polarity change characteristics. A cross-modal consistency characterization is formed based on the number of consistent channels, phase correlation coefficient, and arrival time difference residual. Finally, the phase segment results, pulse morphology characterization results, and cross-modal consistency characterization results are compared with those of tip discharge and suspended discharge. The phase intervals, pulse morphology characteristic thresholds, and consistency index thresholds corresponding to floating discharge, surface discharge, and air gap discharge are compared. The number of matching items for each defect type in the three types of characterization results is counted. When all three types of characterization results match the same defect type, the partial discharge defect type is output. When the number of matching items is the largest and unique, the defect type is output and marked as pending verification. When the number of matching items is tied for the largest, the defect type matching the phase segment is output and marked as low confidence. The fused stable value, partial discharge defect type, phase correspondence result, and feature matching result are archived into the partial discharge detection database. When the fusion stability value is less than the stability threshold, the current partial discharge event is marked as an unstable event, and the current partial discharge event is suspended from entering the partial discharge defect type identification and formal trend statistical processing. Only the partial discharge existence conclusion and risk warning are retained.

10. A comprehensive online detection and visualization control system for partial discharge of electrical equipment according to claim 1, characterized in that: The specific steps for constructing time-series curves by calling the partial discharge detection database records to achieve visualized comparison of link drift tracking and correction effects are as follows: The system utilizes partial discharge (PD) event markers, link drift records, time difference of arrival (TDOA) residuals, and stability assessment results from the PD detection database. It constructs a multi-layered visualization view based on device number, observation time window, event number, and channel number. The system displays continuous changes in drift disturbance values, correction confidence values, and fusion stability values ​​using time-series curves. The event details view displays PD pulse waveforms, power frequency phase positions, event splicing results, TDOA residuals, and cross-modal feature matching differences. The status comparison view shows the relationship between cross-channel time difference, cross-node timestamp deviation, interface buffer occupancy rate, link load rate, and clock drift rate across different observation time windows. The alarm view displays pending verification markers, risk warnings, PD defect types, and trend statistics. This enables PD event localization, link drift tracking, time correction effect comparison, fusion result verification, and defect evolution status control.

Images