POE-based partial discharge device synchronization and positioning method

Abstract

The invention provides a POE-based partial discharge device synchronization and positioning method, which comprises the following steps: performing unified power supply and data communication on a master node and a plurality of slave nodes through POE networking, and performing high-precision timestamp marking on partial discharge signals acquired by each node based on a PTP protocol and high-frequency clock counting of an FPGA chip to obtain a primary partial discharge signal group; performing time alignment on the primary partial discharge signal group based on the timestamp of the main node to obtain an aligned partial discharge signal group; extracting a partial discharge signal feature group from the aligned partial discharge signal group, and constructing a sensing signal network diagram based on the network structure of the POE network and the partial discharge signal feature group; performing message passing and local topology modeling on the sensing signal network diagram based on a pre-trained graph neural network model to obtain a signal topology network diagram; and performing dependency coding and partial discharge positioning analysis on the signal topology network diagram based on a global attention mechanism to obtain a partial discharge node position.

Description

TECHNICAL FIELD

[0001] The present application relates to partial discharge synchronization technology field, especially to a kind of partial discharge equipment synchronization and positioning method based on POE. BACKGROUND

[0002] Partial discharge as the important early sign of power equipment insulation deterioration and potential fault, widely exists in transformer, switch cabinet, cable and high-voltage electrical equipment, its occurrence position and propagation characteristic are directly related to the safety and reliability of equipment operation, the key technical link for realizing power equipment condition monitoring and fault early warning is that partial discharge signal is collected and accurately positioned by multiple nodes synchronization, with the development of smart grid and digital substation, partial discharge on-line monitoring system based on multi-sensor node collaborative sensing gradually becomes mainstream technical direction.

[0003] However, the existing partial discharge synchronization and positioning method still faces many technical bottlenecks in complex field environment. First, the multi-node synchronization scheme based on traditional GPS or dedicated timing link is strongly dependent on external clock source, which is difficult to work stably in underground cable channel, metal enclosed space or strong electromagnetic interference environment, limiting its application scenarios. Secondly, the existing positioning method based on TDOA and other geometric models usually relies on accurate propagation model and node physical coordinate information, which is sensitive to signal noise, path multipath effect and network topology change, and the positioning accuracy and robustness are difficult to guarantee in actual engineering. In addition, traditional methods usually separate the design of signal processing, network modeling and positioning calculation, lack the joint modeling ability of sensor network topology relationship and global correlation characteristics of partial discharge signal, and it is difficult to fully exploit the fault positioning information contained in multi-node collaborative observation. SUMMARY

[0004] (I) Technical problems solved In view of the defects of the prior art, the present application provides a partial discharge equipment synchronization and positioning method based on POE, which has the advantages of high synchronization accuracy, simple deployment and strong positioning robustness, solves the problem of low accuracy of partial discharge source positioning caused by unstable GPS signal and difficulty in realizing high-precision time synchronization in partial discharge positioning scene.

[0005] (II) Technical scheme To achieve the above object, the present application provides the following technical scheme: The present application provides a partial discharge equipment synchronization and positioning method based on POE, comprising the following steps: POE networking is used to supply power and data communication for the master node and multiple slave nodes, and high-precision time stamp marking is carried out on the partial discharge signals collected by each node based on PTP protocol and high-frequency clock counting of FPGA chip, to obtain a primary partial discharge signal group; aligning the primary partial discharge signal set based on the master node to obtain an aligned partial discharge signal set; extracting a partial discharge signal feature set from the aligned partial discharge signal set, and constructing a sensing signal network graph based on the POE networking network structure and the partial discharge signal feature set; performing message passing and local topology modeling on the sensing signal network graph based on a pre-trained graph neural network model to obtain a signal topology network graph; performing dependency coding and partial discharge positioning analysis on the signal topology network graph based on a global attention mechanism to obtain a partial discharge node position.

[0006] According to one of the preferred embodiments of the present application, the high-precision time stamp marking of the partial discharge signals collected by each node based on the PTP protocol and the FPGA chip high-frequency clock counting obtains a primary partial discharge signal set, which includes: calibrating the local clock of the master node and each slave node based on the PTP protocol to obtain a calibrated node clock set; initializing a high-frequency clock counting module in the FPGA chip of each node, and initializing the counting starting point of each high-frequency clock counting module based on the calibrated node clock set; performing partial discharge signal detection on each node to obtain a partial discharge trigger signal set; acquiring the partial discharge counting value set corresponding to the partial discharge trigger signal set using each high-frequency clock counting module, and generating a partial discharge precision time stamp set based on the partial discharge counting value set and the calibrated node clock set; timestamp marking the partial discharge trigger signal set using the partial discharge precision time stamp set to obtain a primary partial discharge signal set.

[0007] According to another preferred embodiment of the present application, the local clock calibration of the master node and each slave node based on the PTP protocol obtains a calibrated node clock set, which includes: taking the local clock of the master node as the master clock and taking the local clock of each slave node as the slave clock; sending a synchronization message to each slave node using the master node based on the PTP protocol, and recording the first time stamp corresponding to the synchronization message when it is sent using the master clock; receiving the synchronization message using each slave node, and recording the second time stamp set corresponding to the synchronization message when it arrives using the slave clock; generating a follow-up message based on the first time stamp using the master node, and sending the follow-up message to each slave node; sending a delay request message set to the master node using each slave node, and recording the third time stamp set corresponding to the delay request message set when it is sent using each slave clock; receiving, by the master node, the delay request message set, and recording, by the master clock, a fourth timestamp set corresponding to the delay request message set when arriving; generating, by the master node, a delay response message according to the fourth timestamp set, and sending the delay response message to each slave node; calculating, according to the first timestamp, the second timestamp set, the third timestamp set and the fourth timestamp set, a clock deviation set and a network delay set for each slave node; performing time correction for each slave clock based on the clock deviation set and the network delay set, and taking each time-corrected slave clock as a calibrated node clock set.

[0008] According to another preferred embodiment of the present application, the calculating, according to the first timestamp, the second timestamp set, the third timestamp set and the fourth timestamp set, a clock deviation set and a network delay set for each slave node comprises: selecting each slave node as a target slave node, extracting the second timestamp of the target slave node from the second timestamp set, extracting the third timestamp of the target slave node from the third timestamp set, and extracting the fourth timestamp of the target slave node from the fourth timestamp set; constructing a bidirectional timestamp pair between the target slave node and the master node based on the first timestamp, the second timestamp, the third timestamp and the fourth timestamp; calculating a round-trip communication delay between the target slave node and the master node according to the bidirectional timestamp pair; calculating a network one-way delay corresponding to the target slave node according to the round-trip communication delay; calculating a clock deviation of the target slave node relative to the master node based on the network one-way delay and a time difference between the first timestamp and the second timestamp; collecting the clock deviations of all slave nodes into a clock deviation set, and collecting the network one-way delays of all slave nodes into a network delay set.

[0009] According to another preferred embodiment of the present application, the time aligning, by the master node, the primary local oscillation signal set to obtain an aligned local oscillation signal set comprises: taking the local oscillation precision timestamp corresponding to the master node in the primary local oscillation signal set as a master timestamp, and taking the local oscillation precision timestamps corresponding to each slave node in the primary local oscillation signal set as a reference timestamp set; The time offset between the reference timestamp group and the master timestamp is calculated to obtain the node time offset group. The node time offset group is then compensated for based on the system delay of each slave node to obtain the compensated time offset group. Based on the compensated time offset group, the reference timestamp group is mapped to a unified time axis with the master node as the reference to obtain a global timestamp group. The global timestamp group is then sorted according to time order to obtain a global timestamp sequence. Based on the global timestamp sequence, the primary partial discharge signal group is subjected to time window truncation and signal rearrangement to obtain an aligned partial discharge signal group.

[0010] According to another preferred embodiment of the present invention, the step of extracting the partial discharge signal feature group from the aligned partial discharge signal group includes: The aligned partial discharge signal group is cleaned and denoised to obtain a denoised partial discharge signal group. Partial discharge pulse detection is performed on the noise-reduced partial discharge signal group to obtain the pulse start signal group; Based on the pulse start signal group, a fixed-length signal segment is extracted to obtain a partial discharge pulse waveform segment group; The pulse arrival time, pulse amplitude, rise and fall time, pulse width, and pulse energy are extracted from the partial discharge pulse waveform segment group to obtain the partial discharge time domain feature group; The partial discharge pulse waveform segment group is transformed in the frequency domain to obtain the partial discharge signal frequency domain group, and the center frequency, mean square frequency and spectral standard deviation are extracted from the partial discharge signal frequency domain to obtain the partial discharge frequency domain feature group. Wavelet transform is performed on the partial discharge pulse waveform segment group, and the wavelet coefficient energy corresponding to the partial discharge pulse waveform segment group after wavelet transform is extracted to obtain the partial discharge time-frequency characteristic group. The partial discharge time-domain feature group, the partial discharge frequency-domain feature group, and the partial discharge time-frequency feature group are fused into a partial discharge multidimensional feature group, and the feature dimensionality of the partial discharge multidimensional feature group is reduced based on principal component analysis to obtain the partial discharge signal feature group.

[0011] According to another preferred embodiment of the present invention, the construction of the sensor signal network diagram based on the network structure of the PoE network and the partial discharge signal characteristic group includes: Each node in the PoE network is aggregated into a network node set as a network node, and the partial discharge signal characteristic group is mapped to the corresponding network node in the network node set to obtain the partial discharge characteristic node set; Based on the physical connection relationship between each node in the PoE network, edges are created between the node sets to obtain the node edge set; Based on the network structure of the PoE network, extract the edge distance set corresponding to the node edge set, and count the switch hop count set corresponding to the node edge set; The feature similarity between each partial discharge signal feature in the partial discharge signal feature group is calculated to obtain the signal similarity matrix; The edge weight set corresponding to the node edge set is calculated based on the edge distance set, the switch hop count set, and the signal similarity matrix. A sensor signal network graph is generated based on the partial discharge feature node set, the node edge set, and the edge weight set.

[0012] According to another preferred embodiment of the present invention, the pre-trained graph neural network model is used to perform message passing and local topology modeling on the sensing signal network graph to obtain a signal topology network graph, including: The partial discharge feature node set in the sensing signal network graph is dimension-mapped based on the pre-trained graph neural network model to obtain the initial node embedding feature set. Select target localized discharge feature nodes one by one from the set of localized discharge feature nodes, and generate messages for the neighbor nodes of the target localized discharge feature nodes based on the initial node embedding feature set and the corresponding edge weight set to obtain the target neighbor message set; The target neighbor message set is aggregated to obtain aggregated neighbor messages; Based on the aggregated neighbor messages, the initial node embedding features of the target partial discharge feature node are gated and updated to obtain standard node embedding features. Then, the initial node embedding feature set is updated to a standard node embedding feature set using all the standard node embedding features. Multi-layer message passing is performed on the standard node embedded feature set to obtain a stacked embedded feature set. Based on the structural dependency relationship of each target localized feature node after multi-layer message passing, local topological encoding is performed on the stacked embedded feature set to obtain a topological embedded feature set. The sensor signal network graph is updated using the topology embedding feature set to obtain a signal topology network graph.

[0013] According to another preferred embodiment of the present invention, the step of performing dependency coding and partial discharge localization analysis on the signal topology network graph based on a global attention mechanism to obtain the location of partial discharge nodes includes: The topology embedding features corresponding to each network node in the signal topology network diagram are sorted by node position and encoded by position to obtain a position encoded feature sequence. The position encoded feature sequence is then linearly mapped to obtain a query vector sequence, a key vector sequence, and a value vector sequence. Based on the query vector sequence and the key vector sequence, the global attention score between any two positional encoding features in the positional encoding feature sequence is calculated, and the global attention score is normalized to obtain the global attention weight matrix. The value vector sequence is weighted and summed based on the global attention weight matrix to obtain a global dependency encoding feature sequence. The global dependency encoded feature sequence is subjected to multi-layer attention stacking encoding to obtain a deep encoded feature sequence, and the deep encoded feature sequence is then subjected to global pooling to obtain localization global features; The location of the partial discharge node is obtained by performing a location regression mapping on the global positioning features.

[0014] To achieve at least one of the above-mentioned objectives, the present invention further provides a PoE-based partial discharge equipment synchronization and positioning system, the system comprising a signal acquisition module, a time synchronization module, a feature extraction module, a topology modeling module, and a positioning analysis module, wherein: The signal acquisition module provides unified power supply and data communication between the master node and multiple slave nodes through PoE networking. Based on the PTP protocol and the high-frequency clock counting of the FPGA chip, it performs high-precision timestamp marking on the partial discharge signals acquired by each node to obtain the primary partial discharge signal group. The time synchronization module performs time alignment on the primary partial discharge signal group based on the master node to obtain an aligned partial discharge signal group. The feature extraction module extracts partial discharge signal feature groups from the aligned partial discharge signal groups and constructs a sensor signal network diagram based on the network structure of the POE network and the partial discharge signal feature groups. The topology modeling module is used to perform message passing and local topology modeling on the sensor signal network graph based on a pre-trained graph neural network model to obtain a signal topology network graph. The localization analysis module performs dependency coding and partial discharge localization analysis on the signal topology network graph based on the global attention mechanism to obtain the location of the partial discharge node.

[0015] The present invention further provides a computer-readable storage medium storing a computer program, which is executed by a processor to implement the above-described method for synchronizing and locating partial discharge devices based on PoE.

[0016] (III) Beneficial Effects Compared with the prior art, the present invention provides a method and system for synchronizing and locating partial discharge equipment based on PoE, which has the following beneficial effects: This PoE-based partial discharge (PD) device synchronization and positioning method utilizes PoE networking to achieve unified power supply and data communication between the master node and multiple slave nodes. This not only simplifies the field wiring structure and improves the consistency and stability of system deployment, but also provides a reliable network foundation for high-precision time synchronization of multiple nodes. By introducing the PTP protocol to accurately calibrate the local clock of each node, time deviations between nodes are eliminated. Combined with the high-frequency clock counting module inside the FPGA, nanosecond-level fine-grained time stamping of PD trigger events is achieved. By fusing the microsecond-level coarse-grained timestamps provided by PTP with the fine-grained timestamps obtained by FPGA counting, the time stamping accuracy and consistency of PD signals are significantly improved. By using the master node timestamp as a unified reference, event-level time alignment of the primary PD signal group is performed, effectively compensating for residual time errors that are difficult to eliminate by relying solely on PTP synchronization. By calculating the node time offset and introducing system delay compensation, deterministic time deviations introduced by sensor response differences, analog-to-digital conversion delays, and differences in FPGA internal logic paths can be corrected. This ensures that the PD signals collected by each node have a strictly consistent time starting point at the same physical event level, thereby improving the accuracy of PD positioning.

[0017] This PoE-based method for synchronizing and locating partial discharge (PD) devices introduces a pre-trained graph neural network to perform message passing and local topology modeling on a sensor signal network graph that integrates physical connectivity and PD signal characteristics. This enables each node to not only perceive its own PD characteristics but also learn the temporal correlations and structural dependencies of its neighbors. Through edge-weighted message generation and gating update mechanisms, the method effectively suppresses interference from noisy nodes on feature propagation, improving the stability and discriminative power of feature representation. Furthermore, through multi-layer message passing and local topology coding, the method models the propagation path and spatial distribution of PD signals within the network, thereby transforming the originally discrete PD observation data into a more coherent whole. The data is transformed into a signal topology network graph with structural semantics, providing high-quality, structure-aware input features for subsequent partial discharge (PD) localization regression based on a global attention mechanism. This improves the accuracy of PD localization. By performing ultra-bandwidth Bayesian optimization and master-slave hyperparameter updates, the efficiency of high-dimensional Bayesian optimization can be effectively alleviated, significantly reducing the uncertainty of manual parameter tuning and improving the generalization and stability of the joint model in complex PoE topology scenarios. This provides a high-quality model foundation for subsequent message passing, global attention dependency modeling, and fault location accuracy, while reducing model configuration complexity, enabling the model to be configured in PoE networks and improving its applicability.

[0018] This PoE-based partial discharge (PD) device synchronization and localization method utilizes a global self-attention mechanism, enabling any network node in the signal topology network diagram to directly interact with all other network nodes in the network. This breaks the local limitations of information transmission, allowing the system to comprehensively analyze the global response pattern of the entire network. It avoids the mislocalization problem caused by relying solely on local neighborhoods, and avoids the unrealistic assumptions of traditional time difference localization methods such as constant wave speed and line-of-sight propagation, as well as the numerical instability problem of solving complex mathematical equations. It is more robust to complex network structures, time delay jitter, and noise, thereby improving the accuracy of PD localization. Attached Figure Description

[0019] Figure 1 The diagram shown is a flowchart of a partial discharge device synchronization and positioning method based on PoE according to the present invention.

[0020] Figure 2 The diagram shown is a structural diagram of the PoE network in this invention. Detailed Implementation

[0021] The following description is intended to disclose the present invention and enable those skilled in the art to implement it. The preferred embodiments described below are merely examples, and other obvious modifications will occur to those skilled in the art. The basic principles of the invention defined in the following description can be applied to other embodiments, modifications, improvements, equivalents, and other technical solutions that do not depart from the spirit and scope of the invention.

[0022] It is understood that the term "a" should be understood as "at least one" or "one or more", that is, in one embodiment, the number of an element can be one, while in another embodiment, the number of the element can be multiple, and the term "a" should not be understood as a limitation on the number.

[0023] Example 1: Please combine Figure 1 This invention discloses a method for synchronizing and locating partial discharge equipment based on PoE, the method comprising the following steps: PoE networking enables unified power supply and data communication between the master node and multiple slave nodes. Based on the PTP protocol and the high-frequency clock counting of the FPGA chip, the partial discharge signals collected by each node are marked with high precision timestamps to obtain the primary partial discharge signal group.

[0024] In this regard, please combine Figure 2The PoE networking refers to a network system established through Power over Ethernet (PoE), which connects the master node and each slave node via ordinary Ethernet cables, transmitting data while transmitting power, and powering the sensors of the master node and each slave node. Each slave node's corresponding sensor contains an FPGA chip, which stands for Field Programmable Gate Array (FPGA). An FPGA chip is a programmable integrated circuit chip that allows users to configure its internal logic functions through hardware description programming after the chip leaves the factory, realizing custom digital circuits. The partial discharge signal refers to the signal generated by the partial discharge phenomenon, which is a common defect phenomenon in the insulation system of high-voltage electrical equipment. Under the action of a strong electric field, certain local areas in the insulating medium discharge, but have not yet formed a penetrating breakdown channel. This discharge generates a series of physical signals, including electrical pulses, electromagnetic waves, ultrasonic waves, light, and heat, among which the electrical pulse signal is most commonly detected. The unified power supply and data communication between the master node and multiple slave nodes through PoE networking refers to connecting, powering, and communicating between the master node and multiple slave nodes through a PoE switch.

[0025] In detail, the high-frequency clock counting based on the PTP protocol and the FPGA chip performs high-precision timestamp marking on the partial discharge signals collected by each node to obtain a primary partial discharge signal group, including: Based on the PTP protocol, the master node and each slave node are locally clocked to obtain a clock group of calibrated nodes. Initialize the high-frequency clock counting module in the FPGA chip of each node, and initialize the counting start point for each high-frequency clock counting module based on the clock group of the calibration node; Partial discharge signal detection is performed at each node to obtain a partial discharge trigger signal group; The partial discharge count value group corresponding to the partial discharge trigger signal group is obtained by using each high-frequency clock counting module, and a partial discharge accuracy timestamp group is generated based on the partial discharge count value group and the calibration node clock group. The partial discharge trigger signal group is timestamped using the partial discharge precision timestamp group to obtain the primary partial discharge signal group.

[0026] The high-frequency clock counting module refers to the high-frequency clock signal generated by the high-stability crystal oscillator inside the FPGA chip. Initializing the counting start point of each high-frequency clock counting module based on the calibration node clock group means re-initializing the counting start point of the high-frequency clock counting module using the local clock of each node in the calibration node clock group. Partial discharge signal detection refers to continuously acquiring analog signals using the partial discharge sensors of each node, and generating a digital pulse when the partial discharge pulse exceeds a threshold using the trigger function of the analog-to-digital converter of the partial discharge sensor, thus obtaining the partial discharge trigger signal group. The partial discharge count value... The group represents the count values ​​corresponding to the high-frequency clock counting modules of each node when the partial discharge arrives. The process of generating a partial discharge accuracy timestamp group based on the partial discharge count value group and the calibration node clock group refers to using the calibration node clock group to collect the coarse-grained timestamps corresponding to each trigger signal, calculating the fine-grained timestamps based on each partial discharge count value and the frequency of the high-frequency clock counting module, adding the coarse-grained timestamps to the fine-grained timestamps to obtain the partial discharge accuracy timestamps, and then aggregating the partial discharge accuracy timestamps of each node into a partial discharge accuracy timestamp group. The coarse-grained timestamps are timestamps with microsecond-level accuracy, and the fine-grained timestamps are timestamps with nanosecond-level accuracy.

[0027] Specifically, the step of performing local clock calibration on the master node and each slave node based on the PTP protocol to obtain a calibrated node clock group includes: The local clock of the master node is used as the master clock, and the local clock of each slave node is used as the slave clock. Based on the PTP protocol, the master node sends synchronization messages to each slave node, and the master clock records the first timestamp corresponding to the synchronization message when it is sent. The synchronization message is received by each slave node, and the second timestamp group corresponding to the arrival of the synchronization message is recorded by the slave clock. The master node generates a follow-up message based on the first timestamp and sends the follow-up message to each slave node; Each slave node sends a delay request message group to the master node, and each slave clock records the third timestamp group corresponding to the delay request message group when it is sent. The master node receives the delay request message group, and the master clock records the fourth timestamp group corresponding to the arrival of the delay request message group. The master node generates a delayed response message based on the fourth timestamp group and sends the delayed response message to each slave node. Based on the first timestamp, the second timestamp group, the third timestamp group, and the fourth timestamp group, a clock skew group and a network latency group are calculated for each slave node; Based on the clock skew group and the network delay group, time correction is performed on each slave clock, and the time-corrected slave clocks are used as the calibration node clock group.

[0028] PTP protocol refers to Precision Time Protocol (PTP) conforming to the IEEE 1588 standard. In the PTP protocol, the synchronization message is the Sync message, the follow message is the Follow_Up message, the delay request message is the Delay_Req message, and the delay response message is the Delay_Resp message.

[0029] Specifically, the step of calculating the clock skew group and network latency group for each slave node based on the first timestamp, the second timestamp group, the third timestamp group, and the fourth timestamp group includes: Select each slave node as a target slave node, extract the second timestamp of the target slave node from the second timestamp group, extract the third timestamp of the target slave node from the third timestamp group, and extract the fourth timestamp of the target slave node from the fourth timestamp group; A bidirectional timestamp pair between the target slave node and the master node is constructed based on the first timestamp, the second timestamp, the third timestamp, and the fourth timestamp; The round-trip communication delay between the target slave node and the master node is calculated based on the bidirectional timestamps. The one-way network delay corresponding to the target slave node is calculated based on the round-trip communication delay. The clock offset of the target slave node relative to the master node is calculated based on the network one-way delay and the time difference between the first timestamp and the second timestamp; All slave node clock skews are aggregated into a clock skew group, and all slave node network one-way delays are aggregated into a network delay group.

[0030] Specifically, the bidirectional timestamp pair refers to the bidirectional timestamp between the second timestamp and the first timestamp, and the bidirectional timestamp between the fourth timestamp and the third timestamp. The round-trip communication delay is the first communication delay obtained by subtracting the first timestamp from the second timestamp and the second communication delay obtained by subtracting the third timestamp from the fourth timestamp. The one-way network delay is the average of the first communication delay and the second communication delay. Calculating the clock deviation of the target slave node relative to the master node based on the one-way network delay and the time difference between the first timestamp and the second timestamp means subtracting the one-way network delay from the time difference between the second timestamp and the first timestamp to obtain the clock deviation.

[0031] By utilizing PoE networking to achieve unified power supply and data communication between the master node and multiple slave nodes, not only is the field wiring structure simplified and the consistency and stability of system deployment improved, but a reliable network foundation is also provided for high-precision time synchronization of multiple nodes. By introducing the PTP protocol to accurately calibrate the local clock of each node, time deviations between nodes are eliminated. Combined with the high-frequency clock counting module inside the FPGA, nanosecond-level fine-grained time stamping of partial discharge trigger events is achieved. By fusing the microsecond-level coarse-grained timestamps provided by PTP with the fine-grained timestamps obtained by FPGA counting, the time stamping accuracy and consistency of the partial discharge signal are significantly improved, providing a highly reliable time reference for subsequent multi-node signal time alignment, sensor signal network graph construction, and spatial positioning analysis based on graph neural networks.

[0032] Based on the master node, the primary partial discharge signal group is time-aligned to obtain an aligned partial discharge signal group.

[0033] Specifically, the step of time-aligning the primary partial discharge signal group based on the master node to obtain an aligned partial discharge signal group includes: The partial discharge precision timestamp corresponding to the master node in the primary partial discharge signal group is used as the master timestamp, and the partial discharge precision timestamp corresponding to each slave node in the primary partial discharge signal group is used as the reference timestamp to obtain a reference timestamp group. The time offset between the reference timestamp group and the master timestamp is calculated to obtain the node time offset group. The node time offset group is then compensated for based on the system delay of each slave node to obtain the compensated time offset group. Based on the compensated time offset group, the reference timestamp group is mapped to a unified time axis with the master node as the reference to obtain a global timestamp group. The global timestamp group is then sorted according to time order to obtain a global timestamp sequence. Based on the global timestamp sequence, the primary partial discharge signal group is subjected to time window truncation and signal rearrangement to obtain an aligned partial discharge signal group.

[0034] Wherein, the time offset refers to the time offset obtained by subtracting the main timestamp from each reference timestamp in the reference timestamp group; the system delay refers to the partial discharge sensor response delay, analog-to-digital conversion delay, and FPGA internal logic delay of each node; the system delay can be obtained based on the hardware configuration parameters of each node; the delay compensation refers to subtracting the system delay of the corresponding node from the corresponding partial discharge precision timestamp to obtain the corresponding compensated time offset; the step of performing time window truncation and signal rearrangement on the primary partial discharge signal group based on the global timestamp sequence to obtain the aligned partial discharge signal group refers to grouping each global timestamp in the global timestamp sequence using a time window of a preset length, and trunculating and rearranging the primary partial discharge signals in each group so that the signals collected by each node belonging to the same partial discharge event are precisely aligned on the time axis.

[0035] By using the master node timestamp as a unified reference, event-level time alignment of the primary partial discharge signal group is performed, effectively compensating for residual time errors that are difficult to eliminate by relying solely on PTP synchronization. By calculating the node time offset and introducing system delay compensation, deterministic time deviations introduced by differences in sensor response, analog-to-digital conversion delay, and differences in FPGA internal logic paths can be corrected. This ensures that the partial discharge signals collected by each node have a strictly consistent time starting point at the same physical event level. Through unified time axis mapping and time window clipping, accurate synchronization and alignment of multi-node partial discharge signals are achieved, providing highly consistent and reliable input data for subsequent construction of sensor signal network diagrams and development of graph neural networks and global attention modeling, thereby improving the accuracy of partial discharge localization.

[0036] Partial discharge signal feature groups are extracted from the aligned partial discharge signal groups, and a sensing signal network diagram is constructed based on the network structure of the POE network and the partial discharge signal feature groups.

[0037] Specifically, the step of extracting the partial discharge signal feature group from the aligned partial discharge signal group includes: The aligned partial discharge signal group is cleaned and denoised to obtain a denoised partial discharge signal group. Partial discharge pulse detection is performed on the noise-reduced partial discharge signal group to obtain the pulse start signal group; Based on the pulse start signal group, a fixed-length signal segment is extracted to obtain a partial discharge pulse waveform segment group; The pulse arrival time, pulse amplitude, rise and fall time, pulse width, and pulse energy are extracted from the partial discharge pulse waveform segment group to obtain the partial discharge time domain feature group; The partial discharge pulse waveform segment group is transformed in the frequency domain to obtain the partial discharge signal frequency domain group, and the center frequency, mean square frequency and spectral standard deviation are extracted from the partial discharge signal frequency domain to obtain the partial discharge frequency domain feature group. Wavelet transform is performed on the partial discharge pulse waveform segment group, and the wavelet coefficient energy corresponding to the partial discharge pulse waveform segment group after wavelet transform is extracted to obtain the partial discharge time-frequency characteristic group. The partial discharge time-domain feature group, the partial discharge frequency-domain feature group, and the partial discharge time-frequency feature group are fused into a partial discharge multidimensional feature group, and the feature dimensionality of the partial discharge multidimensional feature group is reduced based on principal component analysis to obtain the partial discharge signal feature group.

[0038] Signal cleaning can be performed using high-pass filters or baseline drift removal, and signal denoising can be performed using Butterworth filters or finite impulse response filters. Partial discharge pulse detection can be performed using a threshold method, i.e., a fixed signal value is set as a signal threshold, and signals exceeding the signal threshold are taken as the pulse start signal. The rise and fall times refer to the signal rise time and the signal fall time. The fusion into a multi-dimensional feature group of partial discharge refers to the feature normalization and feature splicing of the partial discharge time domain features, partial discharge frequency domain features, and partial discharge time-frequency features corresponding to each node to obtain multi-dimensional features of partial discharge.

[0039] Specifically, the construction of the sensor signal network diagram based on the PoE networking network structure and the partial discharge signal characteristic group includes: Each node in the PoE network is aggregated into a network node set as a network node, and the partial discharge signal characteristic group is mapped to the corresponding network node in the network node set to obtain the partial discharge characteristic node set; Based on the physical connection relationship between each node in the PoE network, edges are created between the node sets to obtain the node edge set; Based on the network structure of the PoE network, extract the edge distance set corresponding to the node edge set, and count the switch hop count set corresponding to the node edge set; The feature similarity between each partial discharge signal feature in the partial discharge signal feature group is calculated to obtain the signal similarity matrix; The edge weight set corresponding to the node edge set is calculated based on the edge distance set, the switch hop count set, and the signal similarity matrix. A sensor signal network graph is generated based on the partial discharge feature node set, the node edge set, and the edge weight set.

[0040] Wherein, the edge distances in the edge distance set are the connection distances between the nodes corresponding to the PoE network, the switch hop counts in the switch hop count set refer to the number of Ethernet switches between the two nodes corresponding to each node edge, and the signal similarity matrix can be calculated by the cosine similarity algorithm. The edge weights in the edge weight set include: dividing the switch hop count of the corresponding node edge by a preset hop count scale coefficient to obtain the switch hop count power; dividing the edge distance of the corresponding node edge by a preset distance scale coefficient to obtain the distance power; calculating the sum of the squares of the switch hop count power and the distance power; performing exponential decay calculation on the sum of squares to obtain the decay exponent; and using the product of the signal similarity and the decay exponent as the edge weight.

[0041] By performing multi-domain feature extraction and fusion on the aligned partial discharge signal group, not only is the key information of the partial discharge pulse in time, frequency and time-frequency plane preserved, but also feature redundancy is reduced through principal component analysis, improving the stability of subsequent modeling and analysis. By introducing the network topology information of POE networking, the physical or logical connection relationship of nodes, communication level features and partial discharge signal similarity are integrated into the edge weight calculation, so that the constructed sensor signal network graph has both structural constraints and signal semantic information. It also provides a unified data expression basis for subsequent graph-based anomaly propagation analysis, partial discharge source localization or graph neural network inference.

[0042] The sensor signal network graph is obtained by performing message passing and local topology modeling based on a pre-trained graph neural network model.

[0043] In detail, the pre-trained graph neural network model performs message passing and local topology modeling on the sensing signal network graph to obtain a signal topology network graph, including: The partial discharge feature node set in the sensing signal network graph is dimension-mapped based on the pre-trained graph neural network model to obtain the initial node embedding feature set. Select target localized discharge feature nodes one by one from the set of localized discharge feature nodes, and generate messages for the neighbor nodes of the target localized discharge feature nodes based on the initial node embedding feature set and the corresponding edge weight set to obtain the target neighbor message set; The target neighbor message set is aggregated to obtain aggregated neighbor messages; Based on the aggregated neighbor messages, the initial node embedding features of the target partial discharge feature node are gated and updated to obtain standard node embedding features. Then, the initial node embedding feature set is updated to a standard node embedding feature set using all the standard node embedding features. Multi-layer message passing is performed on the standard node embedded feature set to obtain a stacked embedded feature set. Based on the structural dependency relationship of each target localized feature node after multi-layer message passing, local topological encoding is performed on the stacked embedded feature set to obtain a topological embedded feature set. The sensor signal network graph is updated using the topology embedding feature set to obtain a signal topology network graph.

[0044] The dimensional mapping refers to mapping the set of partial discharge signal features corresponding to the set of partial discharge feature nodes to a unified embedding dimension space through a fully connected layer. The message generation refers to weighting the initial node embedding features of each neighbor node with corresponding edge weights, calculating the attention coefficient between the target partial discharge feature node and each of its neighbor nodes using an attention mechanism, and combining this coefficient with the edge weights to weight the initial node embedding features of the neighbor nodes to obtain the corresponding target neighbor message. Message aggregation can be performed using the max pooling algorithm, and gating updates can be performed using a Gated Recurrent Unit Neural Network (GRU). By using update gates and reset gates to control the degree of update of the target node state by neighbor information, the information flow can be better controlled, and the gradient vanishing or oversmoothing problem in multi-layer message passing can be alleviated. The multi-layer message passing refers to progressively modeling higher-order node neighborhoods and performing multi-layer passing in the order of message generation, message aggregation, gating update, and embedding feature update. The local topological encoding refers to extracting the subgraph formed by the K-hop neighbors of each target partial discharge feature node and encoding the topological structure of the subgraph using a subgraph neural network, and using it as a structural dependency for local topological encoding.

[0045] By introducing a pre-trained graph neural network to perform message passing and local topology modeling on the sensor signal network graph that integrates physical connectivity and partial discharge (PD) signal characteristics, each node can not only perceive its own PD characteristics but also learn the temporal correlation and structural dependency of neighboring nodes. Through edge-weighted message generation and gating update mechanisms, the interference of noisy nodes on feature propagation can be effectively suppressed, improving the stability and discriminativeness of feature representation. Through multi-layer message passing and local topology coding, the propagation path and spatial distribution of PD signals in the network are further modeled, thereby transforming the originally discrete PD observation data into a signal topology network graph with structural semantics. This provides high-quality, structure-aware input features for subsequent PD localization regression based on a global attention mechanism, improving the accuracy of PD localization.

[0046] Specifically, before performing message passing and local topology modeling on the sensing signal network graph based on the pre-trained graph neural network model to obtain the signal topology network graph, the method further includes: Obtain the value range of each hyperparameter in the preset graph neural network model and the preset attention network model to obtain the hyperparameter search space; The hyperparameter search space is sampled using a super-bandwidth Bayesian optimization algorithm to obtain a sampled hyperparameter configuration set. The sampled hyperparameter configuration set is then subjected to a fast performance evaluation to obtain a preliminary performance score set. The sampling hyperparameter configurations corresponding to the primary performance scores with higher values ​​in the primary performance score set are selected as the target hyperparameter configuration group, and multiple local trust domains are constructed based on the target hyperparameter configuration group; Local Bayesian optimization is performed within each local trust domain to obtain an optimized hyperparameter configuration set. Based on the optimized hyperparameter configuration set, the graph neural network model and the attention network model are jointly trained and their precise performance is evaluated to obtain a secondary performance score. The sampling distribution interval of each local trust domain and the ultra-bandwidth Bayesian optimization algorithm are jointly updated based on the secondary performance score until the secondary performance score exceeds the preset performance threshold. Then, the corresponding optimized hyperparameter configuration set is used as the target hyperparameter configuration set. The graph neural network model and the attention network model are configured with parameters using the target hyperparameter configuration set.

[0047] The hyperparameters include the number of layers, hidden dimension, aggregation function, number of neighbor samples, number of attention heads, query, key, and value vector dimensions, dropout rate, learning rate, weight decay, and batch size of the graph neural network model. The hyperparameter search space can be generated by enumeration or categorical methods. The hyperbandwidth Bayesian optimization algorithm (BOHB) can be used for resource allocation through BOHB's HyperBand and for setting the sampling distribution interval through the Bayesian algorithm. Low-fidelity evaluation algorithms can be used for fast performance evaluation based on evaluation dimensions such as validation set localization error, partial discharge node sorting accuracy score, and convergence speed, while high-fidelity evaluation algorithms can be used for accurate performance evaluation. The construction of multiple local trust domains based on the target hyperparameter configuration group refers to establishing local trust domains centered on the target hyperparameter configuration group and with the hyperparameter scale normalization result as the radius. The local Bayesian optimization refers to parameter optimization based on Gaussian process or random feature approximation through the Bayesian optimization selection function.

[0048] By performing ultra-bandwidth Bayesian optimization and master-slave hyperparameter updates, the efficiency of high-dimensional Bayesian optimization can be effectively alleviated, the uncertainty of manual parameter tuning can be significantly reduced, and the generalization and stability of the joint model in complex PoE topology scenarios can be improved. This provides a high-quality model foundation for subsequent message passing, global attention dependency modeling, and fault location accuracy. At the same time, it reduces the complexity of model configuration, enabling the model to be configured in PoE networks and improving its applicability.

[0049] The location of partial discharge nodes is obtained by performing dependency coding and partial discharge localization analysis on the signal topology network graph based on the global attention mechanism.

[0050] The partial discharge node location refers to the location information of the node in the PoE network corresponding to this partial discharge signal that exhibits the partial discharge phenomenon.

[0051] In detail, the process of performing dependency coding and partial discharge localization analysis on the signal topology network graph based on a global attention mechanism to obtain the location of partial discharge nodes includes: The topology embedding features corresponding to each network node in the signal topology network diagram are sorted by node position and encoded by position to obtain a position encoded feature sequence. The position encoded feature sequence is then linearly mapped to obtain a query vector sequence, a key vector sequence, and a value vector sequence. Based on the query vector sequence and the key vector sequence, the global attention score between any two positional encoding features in the positional encoding feature sequence is calculated, and the global attention score is normalized to obtain the global attention weight matrix. The value vector sequence is weighted and summed based on the global attention weight matrix to obtain a global dependency encoding feature sequence. The global dependency encoded feature sequence is subjected to multi-layer attention stacking encoding to obtain a deep encoded feature sequence, and the deep encoded feature sequence is then subjected to global pooling to obtain localization global features; The location of the partial discharge node is obtained by performing a location regression mapping on the global positioning features.

[0052] The node position sorting refers to sorting based on the number of hops between each network node and the master node or the delay order between each network node and the master node. The position encoding refers to sine and cosine position encoding based on the node position order. The linear mapping refers to mapping the position encoding feature sequence to the query space, key space, and value space of the attention mechanism through a fully connected layer, thereby obtaining the corresponding query vector sequence, key vector sequence, and value vector sequence, which can be normalized by the softmax function. The multi-layer attention stacking encoding refers to using multiple attention encoding layers cascaded to perform multiple linear mappings, global attention calculations, attention normalization, weighted summation, residual connections, and layer normalization operations on the global dependency encoding feature sequence. The global pooling method can be global average pooling. The position regression mapping refers to using a regression layer composed of multiple fully connected networks of the attention neural network model to perform position mapping. The position of the local discharge node is the three-dimensional coordinate of the local discharge node in physical space.

[0053] By employing a global self-attention mechanism, any network node in the signal topology network diagram can directly interact with all other network nodes in the network, breaking the local limitations of information transmission. This allows the system to comprehensively analyze the global response pattern of the entire network, avoiding mislocalization problems caused by relying solely on local neighborhoods. It also avoids the unrealistic assumptions of traditional time difference localization methods, such as constant wave speed and line-of-sight propagation, as well as the numerical instability problems of solving complex mathematical equations. Furthermore, it exhibits stronger robustness to complex network structures, time delay jitter, and noise, thereby improving the accuracy of partial discharge localization.

[0054] Example 2: This invention discloses a PoE-based partial discharge equipment synchronization and positioning system. The system includes a signal acquisition module, a time synchronization module, a feature extraction module, a topology modeling module, and a positioning analysis module, wherein: The signal acquisition module provides unified power supply and data communication between the master node and multiple slave nodes through PoE networking. Based on the PTP protocol and the high-frequency clock counting of the FPGA chip, it performs high-precision timestamp marking on the partial discharge signals acquired by each node to obtain the primary partial discharge signal group. The time synchronization module performs time alignment on the primary partial discharge signal group based on the master node to obtain an aligned partial discharge signal group. The feature extraction module extracts partial discharge signal feature groups from the aligned partial discharge signal groups and constructs a sensor signal network diagram based on the network structure of the POE network and the partial discharge signal feature groups. The topology modeling module is used to perform message passing and local topology modeling on the sensor signal network graph based on a pre-trained graph neural network model to obtain a signal topology network graph. The localization analysis module performs dependency coding and partial discharge localization analysis on the signal topology network graph based on the global attention mechanism to obtain the location of the partial discharge node.

[0055] The processes described above with reference to the flowcharts in the embodiments disclosed in this invention can be implemented as computer software programs. The embodiments disclosed in this invention include a computer program product comprising a computer program carried on a computer-readable medium, the computer program containing program code for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via a communication component, and / or installed from a removable medium. When the computer program is executed by a central processing unit (CPU), it performs the functions defined in the methods of this application. It should be noted that the computer-readable medium described above in this application can be a computer-readable signal medium or a computer-readable storage medium, or any combination of the two. The computer-readable storage medium can be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples of computer-readable storage media may include, but are not limited to: electrical connections having one or more wire segments, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof. In this application, a computer-readable storage medium can be any tangible medium containing or storing a program that can be used by or in connection with an instruction execution system, apparatus, or device. In this application, a computer-readable signal medium may include a data signal propagated in baseband or as part of a carrier wave, carrying computer-readable program code. Such propagated data signals can take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. A computer-readable signal medium can also be any computer-readable medium other than a computer-readable storage medium, which can send, propagate, or transmit a program for use by or in connection with an instruction execution system, apparatus, or device. The program code contained on a computer-readable medium may be transmitted using any suitable medium, including but not limited to: wireless segments, wire segments, optical fibers, RF, etc., or any suitable combination thereof.

[0056] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of the present invention. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, can be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.

[0057] Those skilled in the art should understand that the embodiments of the present invention described above and shown in the accompanying drawings are merely examples and do not limit the present invention. The purpose of the present invention has been fully and effectively achieved. The functions and structural principles of the present invention have been shown and explained in the embodiments. Without departing from the stated principles, the implementation of the present invention may have any variations or modifications.

Claims

1. A method for synchronizing and locating partial discharge equipment based on PoE, characterized in that, The method includes: PoE networking enables unified power supply and data communication between the master node and multiple slave nodes. Based on the PTP protocol and the high-frequency clock counting of the FPGA chip, the partial discharge signals collected by each node are marked with high precision timestamps to obtain the primary partial discharge signal group. Based on the master node, the primary partial discharge signal group is time-aligned to obtain an aligned partial discharge signal group; Partial discharge signal feature groups are extracted from the aligned partial discharge signal groups, and a sensing signal network diagram is constructed based on the network structure of the POE network and the partial discharge signal feature groups. The sensor signal network graph is obtained by performing message passing and local topology modeling based on a pre-trained graph neural network model. The location of partial discharge nodes is obtained by performing dependency coding and partial discharge localization analysis on the signal topology network graph based on the global attention mechanism.

2. The method for synchronizing and locating partial discharge equipment based on PoE according to claim 1, characterized in that, The high-frequency clock counting based on the PTP protocol and FPGA chip performs high-precision timestamp marking on the partial discharge signals collected by each node, resulting in a primary partial discharge signal group, including: Based on the PTP protocol, the master node and each slave node are locally clocked to obtain a clock group of calibrated nodes. Initialize the high-frequency clock counting module in the FPGA chip of each node, and initialize the counting start point for each high-frequency clock counting module based on the clock group of the calibration node; Partial discharge signal detection is performed at each node to obtain a partial discharge trigger signal group; The partial discharge count value group corresponding to the partial discharge trigger signal group is obtained by using each high-frequency clock counting module, and a partial discharge accuracy timestamp group is generated based on the partial discharge count value group and the calibration node clock group. The partial discharge trigger signal group is timestamped using the partial discharge precision timestamp group to obtain the primary partial discharge signal group.

3. The method for synchronizing and locating partial discharge equipment based on PoE according to claim 2, characterized in that, The local clock calibration of the master node and each slave node based on the PTP protocol to obtain a calibrated node clock group includes: The local clock of the master node is used as the master clock, and the local clock of each slave node is used as the slave clock. Based on the PTP protocol, the master node sends synchronization messages to each slave node, and the master clock records the first timestamp corresponding to the synchronization message when it is sent. The synchronization message is received by each slave node, and the second timestamp group corresponding to the arrival of the synchronization message is recorded by the slave clock. The master node generates a follow-up message based on the first timestamp and sends the follow-up message to each slave node; Each slave node sends a delay request message group to the master node, and each slave clock records the third timestamp group corresponding to the delay request message group when it is sent. The master node receives the delay request message group, and the master clock records the fourth timestamp group corresponding to the arrival of the delay request message group. The master node generates a delayed response message based on the fourth timestamp group and sends the delayed response message to each slave node. Based on the first timestamp, the second timestamp group, the third timestamp group, and the fourth timestamp group, a clock skew group and a network latency group are calculated for each slave node; Based on the clock skew group and the network delay group, time correction is performed on each slave clock, and the time-corrected slave clocks are used as the calibration node clock group.

4. The method for synchronizing and locating partial discharge equipment based on PoE according to claim 3, characterized in that, The step of calculating the clock skew group and network latency group for each slave node based on the first timestamp, the second timestamp group, the third timestamp group, and the fourth timestamp group includes: Select each slave node as a target slave node, extract the second timestamp of the target slave node from the second timestamp group, extract the third timestamp of the target slave node from the third timestamp group, and extract the fourth timestamp of the target slave node from the fourth timestamp group; A bidirectional timestamp pair between the target slave node and the master node is constructed based on the first timestamp, the second timestamp, the third timestamp, and the fourth timestamp; The round-trip communication delay between the target slave node and the master node is calculated based on the bidirectional timestamps. The one-way network delay corresponding to the target slave node is calculated based on the round-trip communication delay. The clock offset of the target slave node relative to the master node is calculated based on the network one-way delay and the time difference between the first timestamp and the second timestamp; All slave node clock skews are aggregated into a clock skew group, and all slave node network one-way delays are aggregated into a network delay group.

5. The method for synchronizing and locating partial discharge equipment based on PoE according to claim 1, characterized in that, The step of time-aligning the primary partial discharge signal group based on the master node to obtain an aligned partial discharge signal group includes: The partial discharge precision timestamp corresponding to the master node in the primary partial discharge signal group is used as the master timestamp, and the partial discharge precision timestamp corresponding to each slave node in the primary partial discharge signal group is used as the reference timestamp to obtain a reference timestamp group. The time offset between the reference timestamp group and the master timestamp is calculated to obtain the node time offset group. The node time offset group is then compensated for based on the system delay of each slave node to obtain the compensated time offset group. Based on the compensated time offset group, the reference timestamp group is mapped to a unified time axis with the master node as the reference to obtain a global timestamp group. The global timestamp group is then sorted according to time order to obtain a global timestamp sequence. Based on the global timestamp sequence, the primary partial discharge signal group is subjected to time window truncation and signal rearrangement to obtain an aligned partial discharge signal group.

6. The method for synchronizing and locating partial discharge equipment based on PoE according to claim 1, characterized in that, Extracting the partial discharge signal feature group from the aligned partial discharge signal group includes: The aligned partial discharge signal group is cleaned and denoised to obtain a denoised partial discharge signal group. Partial discharge pulse detection is performed on the noise-reduced partial discharge signal group to obtain the pulse start signal group; Based on the pulse start signal group, a fixed-length signal segment is extracted to obtain a partial discharge pulse waveform segment group; The pulse arrival time, pulse amplitude, rise and fall time, pulse width, and pulse energy are extracted from the partial discharge pulse waveform segment group to obtain the partial discharge time domain feature group; The partial discharge pulse waveform segment group is transformed in the frequency domain to obtain the partial discharge signal frequency domain group, and the center frequency, mean square frequency and spectral standard deviation are extracted from the partial discharge signal frequency domain to obtain the partial discharge frequency domain feature group. Wavelet transform is performed on the partial discharge pulse waveform segment group, and the wavelet coefficient energy corresponding to the partial discharge pulse waveform segment group after wavelet transform is extracted to obtain the partial discharge time-frequency characteristic group. The partial discharge time-domain feature group, the partial discharge frequency-domain feature group, and the partial discharge time-frequency feature group are fused into a partial discharge multidimensional feature group, and the feature dimensionality of the partial discharge multidimensional feature group is reduced based on principal component analysis to obtain the partial discharge signal feature group.

7. The method for synchronizing and locating partial discharge equipment based on PoE according to claim 1, characterized in that, The sensor signal network diagram constructed based on the PoE networking network structure and the partial discharge signal characteristic group includes: Each node in the PoE network is aggregated into a network node set as a network node, and the partial discharge signal characteristic group is mapped to the corresponding network node in the network node set to obtain the partial discharge characteristic node set; Based on the physical connection relationship between each node in the PoE network, edges are created between the node sets to obtain the node edge set; Based on the network structure of the PoE network, extract the edge distance set corresponding to the node edge set, and count the switch hop count set corresponding to the node edge set; The feature similarity between each partial discharge signal feature in the partial discharge signal feature group is calculated to obtain the signal similarity matrix; The edge weight set corresponding to the node edge set is calculated based on the edge distance set, the switch hop count set, and the signal similarity matrix. A sensor signal network graph is generated based on the partial discharge feature node set, the node edge set, and the edge weight set.

8. A method for synchronizing and locating partial discharge equipment based on PoE according to claim 7, characterized in that, The pre-trained graph neural network model performs message passing and local topology modeling on the sensor signal network graph to obtain a signal topology network graph, including: The partial discharge feature node set in the sensing signal network graph is dimension-mapped based on the pre-trained graph neural network model to obtain the initial node embedding feature set. Select target localized discharge feature nodes one by one from the set of localized discharge feature nodes, and generate messages for the neighbor nodes of the target localized discharge feature nodes based on the initial node embedding feature set and the corresponding edge weight set to obtain the target neighbor message set; The target neighbor message set is aggregated to obtain aggregated neighbor messages; Based on the aggregated neighbor messages, the initial node embedding features of the target partial discharge feature node are gated and updated to obtain standard node embedding features. Then, the initial node embedding feature set is updated to a standard node embedding feature set using all the standard node embedding features. Multi-layer message passing is performed on the standard node embedded feature set to obtain a stacked embedded feature set. Based on the structural dependency relationship of each target localized feature node after multi-layer message passing, local topological encoding is performed on the stacked embedded feature set to obtain a topological embedded feature set. The sensor signal network graph is updated using the topology embedding feature set to obtain a signal topology network graph.

9. A method for synchronizing and locating partial discharge equipment based on PoE according to claim 8, characterized in that, The process of performing dependency coding and partial discharge (PD) localization analysis on the signal topology network graph based on a global attention mechanism to obtain the PD node locations includes: The topology embedding features corresponding to each network node in the signal topology network diagram are sorted by node position and encoded by position to obtain a position encoded feature sequence. The position encoded feature sequence is then linearly mapped to obtain a query vector sequence, a key vector sequence, and a value vector sequence. Based on the query vector sequence and the key vector sequence, the global attention score between any two positional encoding features in the positional encoding feature sequence is calculated, and the global attention score is normalized to obtain the global attention weight matrix. The value vector sequence is weighted and summed based on the global attention weight matrix to obtain a global dependency encoding feature sequence. The global dependency encoded feature sequence is subjected to multi-layer attention stacking encoding to obtain a deep encoded feature sequence, and the deep encoded feature sequence is then subjected to global pooling to obtain localization global features; The location of the partial discharge node is obtained by performing a location regression mapping on the global positioning features.

10. A PoE-based partial discharge equipment synchronization and positioning system, characterized in that, The system includes a signal acquisition module, a time synchronization module, a feature extraction module, a topology modeling module, and a positioning analysis module, wherein: The signal acquisition module provides unified power supply and data communication between the master node and multiple slave nodes through PoE networking. Based on the PTP protocol and the high-frequency clock counting of the FPGA chip, it performs high-precision timestamp marking on the partial discharge signals acquired by each node to obtain the primary partial discharge signal group. The time synchronization module performs time alignment on the primary partial discharge signal group based on the master node to obtain an aligned partial discharge signal group. The feature extraction module extracts partial discharge signal feature groups from the aligned partial discharge signal groups and constructs a sensor signal network diagram based on the network structure of the POE network and the partial discharge signal feature groups. The topology modeling module is used to perform message passing and local topology modeling on the sensor signal network graph based on a pre-trained graph neural network model to obtain a signal topology network graph. The localization analysis module performs dependency coding and partial discharge localization analysis on the signal topology network graph based on the global attention mechanism to obtain the location of the partial discharge node.

Images