Gis electric field data transmission method based on edge computing

Abstract

The application relates to the technical field of gas insulated metal enclosed switchgear online monitoring and edge computing data transmission, in particular to a GIS electric field data transmission method based on edge computing, which comprises the following steps: collecting GIS multiple measuring point electric field waveforms and each edge node clock deviation, time error, link bandwidth, link jitter and packet loss rate; performing transient event detection, effective waveform interception and event absolute time determination to generate a local event package containing a front edge reserved section, a rear edge control point and an event abstract; organizing, cooperatively locking and block sending based on unified event identification; merging event blocks on the center side and performing coarse alignment, front edge fine alignment and rear edge recovery; and generating a parameter update instruction feedback to the edge node; and the application can consider timing positioning accuracy and transmission efficiency.

Description

Technical Field

[0001] This invention relates to the field of online monitoring and edge computing data transmission technology for gas-insulated metal-enclosed switchgear, specifically a GIS electric field data transmission method based on edge computing. Background Technology

[0002] During withstand voltage tests and long-term operation and maintenance, gas-insulated switchgear may exhibit transient electric field signals with a duration less than a preset time threshold. To identify, compare, and locate such signals, it is usually necessary to simultaneously acquire electric field waveforms at multiple measuring points and transmit the acquisition results to the central side for unified analysis. Existing electric field monitoring and transmission methods can complete multi-point waveform acquisition, time stamping, and remote aggregation processing, and are suitable for equipment condition monitoring processes. With the development of demand for multi-point, high-sampling-rate electric field monitoring, electric field data transmission has gradually expanded from single-point, low-frequency data uploading to multi-node concurrent, transient event-driven data collaborative processing. However, when monitoring nanosecond-level transient signals inside gas-insulated switchgear, existing transmission methods are still easily limited by the amount of full waveform data uploaded exceeding the transmission bandwidth limit, insufficient timing consistency between nodes, and the easy loss of key waveforms under link congestion conditions. This leads to difficulties in aligning multi-channel events and an increased probability of missed detection of weak signal measurement points, thereby affecting the efficiency of subsequent fault analysis and location. Summary of the Invention

[0003] The purpose of this invention is to provide a GIS electric field data transmission method based on edge computing, which avoids the limitations of excessive data volume of full waveform upload and the easy loss of key waveforms under link congestion conditions, and is more likely to solve the problem of difficulty in aligning multi-channel events and reduce the probability of missed detection of weak signal measurement points.

[0004] The objective of this invention can be achieved through the following technical solutions: The electric field waveforms of multiple measuring points of the gas-insulated metal-enclosed switchgear are collected, and the clock deviation, timing error, link bandwidth, link jitter and packet loss rate of each edge node are obtained to form the basic node data. Based on the node's basic data and combined with the preset local trigger threshold, candidate transient events are detected, effective waveforms are truncated, and the absolute time of events is determined for each node's waveform. A local event package is generated, which includes at least the leading edge reserved segment, the trailing edge control point, and the event summary. A unified event identifier is generated based on the absolute time of the event and the node identifier of the corresponding edge node, and the local event packets are organized according to the unified event identifier. When a candidate transient event meets the preset collaborative triggering conditions, a locking notification containing a preset locking time window is sent to other edge nodes, so that the receiving nodes can intercept and cache the event within the locking time window according to the reduced local triggering threshold. Then, the local event packets are split into event blocks according to the link status and sent. The central side merges the received event blocks from each node according to a unified event identifier, and performs coarse alignment, fine alignment of the leading edge, and recovery of the trailing edge on the merged result to generate corresponding multi-channel unified event results, missing block information, and recovery quality information. The central side generates parameter update instructions based on the unified event results of multiple channels, missing block information, and recovery quality information, and sends them to the corresponding edge nodes. The edge nodes then update their local trigger thresholds and preset running configuration parameters accordingly.

[0005] In one alternative approach: the original waveforms of multiple points of the gas-insulated metal-enclosed switchgear are acquired by a multi-channel electro-optical field acquisition device, and the time synchronization relationship between the station's master clock and each edge node is established to obtain the clock deviation and time synchronization error of each edge node. At the same time, the link bandwidth, link jitter and packet loss rate are measured to form the basic data record of the nodes.

[0006] In one alternative approach: candidate transient events are detected based on preset waveform characteristic parameters; The effective waveform interval is extracted based on the start point, main peak point, and end point of the candidate transient event; The absolute time of the event is determined based on the event start information and clock offset, and a local event packet is generated. The event summary includes one or more of the event amplitude, duration, and absolute time.

[0007] In one alternative approach: local event packets are mapped to a sending queue indexed by a unified event identifier, and upon detecting a candidate event, a locking notification including the unified event identifier, absolute event time, preset event strength, and preset locking time window is sent to other edge nodes. This causes other edge nodes to lower their local trigger threshold, retain the circular buffer, and generate associated event packets or empty response records within the corresponding time window. Then, the packets are sent in the order of anchor header block, leading edge retention block, event digest block, trailing edge control point block, and integrity status block.

[0008] In one alternative approach: the central side converts the received event blocks into event contexts merged according to a unified event identifier, determines reference nodes based on the acquired event quality parameters of each node, performs coarse alignment based on the absolute time of events reported by each node, performs fine alignment based on the leading edge reserved segment to obtain fine alignment offset, and then combines the trailing edge control points to restore the complete waveform, generating multi-channel unified event results and missing block information.

[0009] In one alternative approach: based on the fine alignment offset determined by the leading edge fine alignment on the center side, as well as the missing block information and recovery quality information, the local trigger threshold, cooperative reduction threshold coefficient, leading edge retention length, trailing edge compression error, block length, and retransmission mode of the anchor head block and leading edge retention block of the edge node are updated.

[0010] In one alternative approach: the background mean, background standard deviation, and first-order difference standard deviation are continuously recorded at each edge node, and the background statistics are used as the basis for subsequent amplitude determination, first-order difference determination, and signal-to-noise ratio calculation.

[0011] In one alternative: a pre-truncation length and a post-truncation length are set for the effective waveform range of the event, wherein the pre-truncation length is determined by the product of the event rising edge parameter and a first preset scaling factor, and the post-truncation length is determined by the product of the event pulse width parameter and a second preset scaling factor. The event rising edge parameter corresponds to the time length during which the waveform rises from the 10% peak value to the 90% peak value; the event pulse width parameter corresponds to the time length during which the waveform amplitude falls back to the preset background standard deviation range from its own peak value; the first preset scaling factor has a value range of [2,5], and the second preset scaling factor has a value range of [1,3].

[0012] In one alternative approach: if no confirmation is received from the central side, the anchor header block and the leading edge reserved block are retransmitted after a timeout; when the missing block information returned by the central side contains the trailing edge control point block number, the corresponding block is retransmitted; after receiving the missing block information returned by the central side, the edge node determines whether to continue retransmission or end the sending process corresponding to the current unified event identifier.

[0013] In one alternative: hierarchical outputs are also generated for the recovery results of each node. The hierarchical outputs include independent triggering results, collaborative triggering results, empty response records, and low-confidence results, and are associated with the outputs of multi-channel unified event results. Among them, the independent trigger result is when only a single node exceeds the trigger threshold, the collaborative trigger result is when multiple nodes complete fine alignment under a unified event identifier, the empty response record is used to characterize the state of a node receiving a lock notification but not detecting a valid waveform, and the low confidence result is when the event confidence is less than the preset confidence threshold. When the event confidence is greater than or equal to the preset confidence threshold, it is determined to be a valid confidence result.

[0014] The beneficial effects of this invention are: 1. This invention establishes a node-based data system integrating sampling status and network status by jointly collecting and recording the original electric field waveforms at multiple measurement points, edge node clock deviations, timing errors, link bandwidth, link jitter, and packet loss rate. This provides a unified basis for subsequent event detection, time compensation, and link flow control, thereby improving the timing consistency during multi-node collaborative processing. By performing transient event detection, effective waveform truncation, absolute event time determination, and local encapsulation of leading edge retention segments, trailing edge control points, and event summaries on waveforms at edge nodes, the amount of uploaded data can be significantly compressed while retaining key leading edge timing information used for time difference positioning, effectively alleviating the network throughput pressure caused by uploading the entire waveform under high sampling rate conditions.

[0015] 2. This invention, through a cross-node organization method based on a unified event identifier and a collaborative locking notification mechanism triggered when conditions are met, enables other nodes to lower their local trigger thresholds, freeze historical circular buffers, and backtrack to scan weak signals within the corresponding time window. This reduces the probability of missed reports at long-distance measurement points and improves the completeness of multi-channel event capture. By hierarchically and block-wise transmitting according to anchor header blocks, leading edge reserved blocks, event digest blocks, trailing edge control point blocks, and integrity status blocks, and by controlling the priority transmission of key blocks, timeout retransmission, and missing block retransmission in conjunction with link status, it can prioritize the arrival of key timing information under link-limited or congested conditions, reducing the impact of missing key waveforms on the central side reconstruction.

[0016] 3. This invention merges event blocks at the central side using a unified event identifier and sequentially performs coarse alignment, fine leading-edge alignment, and trailing-edge recovery. This enables the reconstruction of unified multi-channel event results even with out-of-order arrival of blocks and clock deviations at nodes, improving the alignment accuracy and recovery integrity of waveforms at multiple measurement points. By outputting independent triggering results, collaborative triggering results, empty response records, and low-confidence results for each node's recovery results, and combining these with time confidence level, leading-edge integrity level, signal-to-noise ratio level, and missing block recovery status for hierarchical judgment, subsequent location analysis can distinguish between true no-signal, communication loss, and low-confidence events, improving the reliability of fault analysis results. Attached Figure Description

[0017] Other features, objects, and advantages of the invention will become more apparent from the following detailed description of non-limiting embodiments with reference to the accompanying drawings: Figure 1 This is a flowchart of the edge computing-based GIS electric field data transmission method of the present invention. Detailed Implementation

[0018] 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.

[0019] Please see Figure 1 The GIS electric field data transmission method based on edge computing includes the following steps: S1, collecting electric field waveforms from multiple measuring points in the GIS, and obtaining clock deviation, time synchronization error, link bandwidth, link jitter and packet loss rate of each edge node to form node basic data; S2. Based on the node basic data and combined with the preset local trigger threshold, perform candidate transient event detection, effective waveform truncation and event absolute time determination on the waveform of each node, and generate a local event package. The local event package includes at least the leading edge reserved segment, the trailing edge control point and the event summary. S3. Generate a unified event identifier based on the absolute time of the event and the node identifier of the corresponding edge node, and organize the local event packets according to the unified event identifier; when the candidate transient event meets the preset collaborative triggering conditions, send a locking notification containing a preset locking time window to other edge nodes, so that the receiving node can intercept and cache the event within the locking time window according to the reduced local triggering threshold; then split the local event packets according to the link status to generate event blocks and send them. S4. The central side merges the received event blocks from each node according to the unified event identifier, and performs coarse alignment, fine alignment of the leading edge and recovery of the trailing edge on the merged result to generate the corresponding multi-channel unified event result, missing block information and recovery quality information. S5. The central side generates parameter update instructions based on the unified event results of multiple channels, missing block information and recovery quality information and sends them to the corresponding edge nodes. The edge nodes then update their local trigger thresholds and preset running configuration parameters accordingly. In S1, the original waveforms of multiple measurement points of GIS are accessed through a multi-channel electro-optical field acquisition device, and the time synchronization relationship between the station master clock and each edge node is established to obtain the clock deviation and time synchronization error of each edge node. At the same time, the link bandwidth, link jitter and packet loss rate are measured to form the basic data record of the node. In S2, candidate transient events are detected based on preset waveform characteristic parameters; valid waveform intervals are extracted based on the start point, main peak point, and end point of the candidate transient events; and the absolute time of the event is determined based on the event start point information and clock deviation to generate a local event packet; the event summary includes one or more of the event amplitude, duration, and absolute time of the event.

[0020] This invention is applied to the scenario of multi-point electro-optical field monitoring of gas-insulated switchgear (GIS). During the withstand voltage test or long-term operation and maintenance of GIS, nanosecond-level transient burst signals may occur inside. The system needs to capture multi-channel signals at an ultra-high sampling rate and perform strict time sequence comparison in order to achieve fault location. The edge computing-based GIS electric field data transmission method has an overall architecture that includes a hierarchical processing flow for edge node feature extraction and central side collaborative reconstruction. After the edge node acquires the original waveform and its own state basic data, it performs event detection and truncation on the waveform to generate a local event packet containing the leading edge reserved segment, trailing edge control point and event summary; Cross-node collaborative locking and block transmission are performed through a unified event identifier; after receiving the data, the central side merges the event blocks and performs coarse alignment, fine leading edge alignment, and trailing edge recovery. The central side sends parameter update instructions to the edge nodes based on the recovery quality to complete the status feedback; when the system is in an abnormal operating condition, if the edge node loses the synchronization signal with the master clock for a short time, it will automatically switch to the local free oscillation clock to maintain sampling, and add a degradation mark to the timestamp in the generated event packet to avoid the central side introducing an incorrect absolute time reference. In this embodiment, the data flow inside the edge node does not jump directly from the sampling end to the sending end, but passes through a three-level buffer: a sampling circular buffer, a candidate event temporary storage area, and a queue of events to be sent; the sampling circular buffer continuously writes the latest sampling point in a loop according to a fixed write pointer; When the detection logic determines that a candidate event has occurred, the cache page for the corresponding time range is immediately frozen, and its status is changed from overwriteable to retained. After the start point, peak point and end point of the event are located, the frozen cache page is mapped as a candidate event temporary storage area, and the interception thread extracts the leading edge reserved segment and the trailing edge segment in read-only mode. Once the absolute time of the event is calculated and the unified event identifier is determined, the event will enter the event queue to be sent. Its status will proceed in the order of pending judgment, locked, pending segmentation, sending, pending retransmission, and completed. If an event does not receive relevant feedback from the central side or trigger the local retransmission condition within the preset retention period, the sending control logic will change its status to timeout and release the corresponding cache page. If the number of candidate events exceeds the local retention limit at the same time, events with higher time reliability and greater event intensity will be retained first. Low-priority candidate events will only retain event summaries and no longer retain complete trailing data, so as to avoid the circular cache being continuously filled and affecting the sampling of subsequent new events. In this embodiment, the collaborative triggering condition in S3 adopts an executable two-level judgment rule: the initiating node determines whether the local candidate event meets the independent transmission condition, that is, the deviation of the event peak relative to the current background mean is not less than 3 times the background standard deviation of the local node, and the first-order difference peak near the main peak is not less than 3 times the first-order difference standard deviation. Next, determine whether the collaborative locking conditions are met, namely, the event intensity reaches more than 70% of the independent transmission threshold of this node, the current time synchronization error is not greater than the preset allowable time synchronization error, and the link packet loss rate is not higher than the preset allowable packet loss rate; the above allowable time synchronization error and allowable packet loss rate are calibrated according to the positioning accuracy requirements and network stress test results during the on-site debugging phase, and written into the node parameter table respectively; for GIS positioning scenarios of 220kV and above, the allowable time synchronization error can be calibrated in the range of 5ns to 20ns, and the allowable packet loss rate can be calibrated in the range of 0.1% to 5%; Locking notifications are sent only when both of the above conditions are met simultaneously; if only the independent sending condition is met but the cooperative locking condition is not met, the node only uploads local event packets without triggering other nodes to lower the threshold, thereby avoiding invalid diffusion of cooperative triggering in low-quality clock or poor link conditions. To ensure that the generation of unified event identifiers based on absolute event time and node identifiers has definite implementation rules, this implementation method further stipulates that: if a candidate event is first met by this node in a collaborative locking condition, then this node shall act as the initiating node and generate a unified event identifier in the order of internal date sequence number - absolute event time quantization value - initiating node identifier; Among them, the absolute time quantization value of the event is the integer result of dividing the absolute time of the event by the preset merged time granularity. The preset merged time granularity is preferably 2 to 5 times the upper limit of the allowed time synchronization error. Once other nodes receive a lock notification, they directly use the unified event identifier carried in the lock notification and do not regenerate it themselves. If an event does not trigger a collaborative lock and is sent independently by a single node, that node will still generate a unified event identifier in the above order and use it for single event archiving on the central side. If a node detects multiple candidate events consecutively within a quantized time granularity, it appends a sequence number of 1 to 3 digits to the end of the unified event identifier according to the local detection order to prevent different events from being incorrectly merged. If this node detects a new candidate event again within the time window corresponding to a unified event identifier that is locked or being sent, it first compares the absolute time difference between the new candidate event and the existing unified event identifier. When the time difference is not greater than half of the current locked time window, the new candidate event is merged into the original unified event identifier and the trailing edge range is extended according to the same event context. When the time difference is greater than half of the current locked time window, a new unified event identifier is generated to prevent a continuous discharge from being mistakenly split into multiple independent events. If waveforms are passively collected without evaluating the node's own clock state and network environment, data congestion and timing errors are very likely to occur when multiple nodes upload concurrently due to sudden events. Therefore, this implementation uses a multi-channel electro-optical field acquisition device to access the original waveform of the point and establishes a time synchronization relationship using the station's master clock. In data flow, the time synchronization protocol message is compared with the local hardware clock to generate clock deviation and time synchronization error. The specific calculation rule is as follows: record the timestamp of the synchronization message sent by the master clock. Timestamps received by edge nodes Timestamps of edge nodes sending delayed request messages and the timestamp received by the master clock Clock deviation, expressed by the formula: Time synchronization error, expressed by the formula: Meanwhile, the network probing module periodically sends a fixed-length UDP probe packet sequence to the central side and receives response packets from the central side. It calculates the link bandwidth and link jitter based on the arrival time interval of the response packets and calculates the packet loss rate based on the proportion of unresponding packets. The measured basic node data is continuously written to the node's shared state record area, which can be directly read and called by the subsequent event interception and flow control modules, thus realizing the logical decoupling of data acquisition and state assessment. If the network detection module does not receive a link response in the current period, it uses the smoothed statistical value from the previous period as the current link status to prevent the flow control parameter calculation from being interrupted. The smoothed statistics are updated through discrete weighted calculation. Specifically, each time a valid measurement is obtained, the current measurement is multiplied by a weighting coefficient of 0.125, and then added to the smoothed statistics of the previous period multiplied by a weighting coefficient of 0.875 to form the current smoothed statistics. If no response is received, the update calculation is skipped to maintain the original value. If no link response is received for more than a preset number of consecutive periods, the link bandwidth smoothing statistics will be forcibly downgraded to the preset safety lower limit value corresponding to the congestion state, and the packet loss rate smoothing statistics will be set to 100% to prevent the local sending queue memory from overflowing due to incorrectly maintaining a good historical estimate when the link is physically disconnected. And when the node is powered on and there are no historical statistical values, the conservative initial value written into the node parameter table is used as the smoothed statistical value of the previous period to start the discrete weighted recursive calculation. To avoid the measurement thread and the event processing thread reading inconsistent states at the same time, the node basic data is updated in a periodic submission manner in this embodiment: the time synchronization error, clock deviation and link status are first sampled and calculated within the current period, and then switched to the current valid state at the end of the period. The event detection thread uses only the same currently valid state throughout the entire process of searching for a candidate event from the starting point to writing the event summary, without switching to new period data midway through processing. This avoids the absolute time compensation and link classification basis of the same event coming from two different statistical periods. At nanosecond-level high sampling rates, uploading the entire original waveform would cause network throughput overload, while extracting only peak features would result in the loss of key leading-edge timing for time difference positioning. Therefore, this implementation detects candidate transient events based on waveform feature parameters and extracts effective waveform intervals based on the start point, main peak point, and end point of the event. In the processing flow, discrete sampling points collected in real time enter the feature extraction pipeline, and the truncation action is triggered when the feature parameters exceed the preset judgment threshold. The preset judgment threshold is not a fixed constant, but is determined by the background statistical results of the most recent event-free stable interval of the node: the amplitude judgment threshold is the background mean plus 3 times the background standard deviation, and the first difference judgment threshold is 3 times the first difference standard deviation. When both amplitude determination and first-order difference determination are met, the adjacent intervals before and after the corresponding sampling points are marked as candidate event search intervals. The event start point is defined as the position where the waveform first exceeds the amplitude determination threshold for no less than 3 consecutive sampling points. The main peak point is defined as the sampling point with the largest absolute amplitude in the candidate interval. The end point is defined as the position where the waveform returns to the background mean ± 1 times the background standard deviation after the main peak and maintains no less than 5 consecutive sampling points. Based on this, the extracted effective waveform interval is divided into a leading edge retention segment and a trailing edge segment: the leading edge retention segment consists of the event start point to the main peak point, and a fixed number of original sampling points after the main peak point; The trailing edge segment refers to the waveform interval from the end position of a fixed number of original sampling points after the autonomous peak point to the end point where the waveform returns to the background mean ± 1 times the background standard deviation and continuously maintains no less than 5 sampling points. For the trailing edge segment, a sliding window of a preset length is used. Traverse the sequence of sampling points and extract those that satisfy the conditions in the window. Local maxima that are the maximum values ​​within a window and whose difference from their adjacent minimum values ​​is greater than the first-order difference standard deviation, and those that satisfy the condition within a window. Local minimum points that are the minimum values ​​and whose difference from the adjacent maximum values ​​is greater than the first-order difference standard deviation are used as trailing control points. Each trailing control point consists of the local time offset of the point relative to the event start point and its corresponding amplitude, so as to achieve feature preservation and depth compression of trailing waveform data. The truncation interval is calculated by combining the local hardware timestamp with the clock deviation obtained in the previous steps to generate the absolute time of the event. The calculation order is limited to: first read the original hardware timestamp corresponding to the start of the event, then subtract the current valid clock deviation value to obtain the compensated timestamp, and use the compensated timestamp as the absolute time of the event to write the event digest. The absolute time, along with information such as event amplitude and duration, is encapsulated into an event digest, which, together with the waveform data, constructs a local event package. The preferred order of the fields in the event digest is a unified event identifier placeholder field, event absolute time, event amplitude, duration, time reliability identifier, and integrity identifier, so that the central side can parse it with a fixed offset. If the calculated absolute time of the event exceeds the reasonable span of the current physical time, it indicates that the deviation compensation calculation has malfunctioned. The system then uses the difference between the current valid clock deviation value and the clock deviation value of the previous valid period as the judgment condition for the reasonable span, so as to avoid abnormal time jumps. Once it exceeds this range, it will backtrack to extract the uncompensated original timestamp and add a low-confidence time sequence identifier to the event summary. In the process of candidate event detection, truncation, and encapsulation, a unidirectional control branch is adopted: if the starting point has been determined but the main peak has not yet exceeded the minimum number of samples to be kept, then continue scanning; if the main peak has been determined but the end point condition has not been met, then keep the current cache page frozen and prohibit the background statistics update of this segment; if the end point does not appear within the maximum waiting length, then process it as a truncated event, that is, use the end of the maximum waiting length as a temporary end point to generate an event packet, and add a tail truncation mark to the integrity identifier to ensure that the detection thread will not block subsequent event processing due to waiting for the end point for a long time; If a second higher peak appears within the same frozen cache segment, the main peak point is reset with the higher peak point, and the start position of the end point search is postponed accordingly. If there is already a stable interval between the second peak and the first peak that meets the end point conditions, it is split into two candidate events and entered into the subsequent encapsulation process respectively, thereby avoiding the incorrect merging of adjacent events.

[0021] During transmission, the local event packet is split into multiple data blocks, where the anchor header block contains a unified event identifier, an initiating node identifier, and the absolute event time. In a preferred implementation, in S3, local event packets are mapped to a sending queue indexed by a unified event identifier. Upon detecting a candidate event, a locking notification including the unified event identifier, absolute event time, preset event strength, and preset locking time window is sent to other edge nodes. This causes other edge nodes to lower their local trigger threshold, retain the circular buffer, and generate associated event packets or empty response records within the corresponding time window. Then, the packets are sent in the order of anchor header block, leading edge retention block, event digest block, trailing edge control point block, and integrity status block. In S4, the central side converts the received event blocks into event contexts merged according to a unified event identifier, determines the reference node based on the event quality parameters of each node, performs coarse alignment based on the absolute time of the events reported by each node, and performs fine alignment based on the leading edge reserved segment to obtain the fine alignment offset. Then, it combines the trailing edge control point to restore the complete waveform and generate multi-channel unified event results and missing block information. In S5, based on the fine alignment offset determined by the fine alignment of the center side, as well as the missing block information and recovery quality information, the local trigger threshold, cooperative reduction threshold coefficient, leading edge retention length, trailing edge compression error, block length, and retransmission mode of anchor head block and leading edge retention block of the edge node are updated.

[0022] When partial discharge occurs at a certain point, if the distance between the measuring point and the point exceeds the preset distance threshold, the signal attenuation exceeds the preset attenuation threshold and does not reach the conventional independent trigger threshold. If each node makes its own independent judgment based solely on its own threshold, it will lead to missed detections in the weak signal channel and destroy the integrity of the multi-channel timing positioning. Therefore, this implementation introduces a horizontal collaborative locking mechanism; the data flow process is as follows: after a node detects a candidate event, it generates a unified event identifier based on the absolute time and node identifier, and maps the local event packet to the sending queue indexed by the identifier; At the same time, a lock notification containing a unified event identifier, absolute event time, event intensity, and lock time window is broadcast to peer edge nodes; Considering that the propagation time difference of partial discharge electromagnetic signals at various measurement points within the GIS is on the order of nanoseconds, while the network communication delay between edge nodes is usually on the order of hundreds of microseconds or even milliseconds, there is a serious physical timing misalignment; in order to ensure the causal consistency of the coordinated triggering, the ring buffer depth of each node is limited to be strictly greater than the sum of the maximum network communication delay and the locking time window. Upon receiving the notification and parsing instructions, the node does not wait for future physical fluctuations. Instead, it uses the absolute time of the event in the notification as a benchmark, combines the local clock deviation to calculate the corresponding historical hardware timestamp, and immediately freezes the circular buffer segment corresponding to the locking time window before and after the historical timestamp to prevent it from being overwritten by new data. The node applies the system's preset cooperative threshold reduction coefficient to the frozen historical buffer segment for backtracking scan. That is, it multiplies the amplitude judgment threshold and first-order differential judgment threshold corresponding to the independent transmission of this node by the cooperative threshold reduction coefficient, which is between 0.5 and 0.8, to obtain the reduced cooperative amplitude threshold and cooperative differential threshold. Based on this, it determines whether the triggering condition is met in the historical buffer segment. If a valid fluctuation is detected during backtracking within the historical buffer segment, the generated associated event packets will be pushed into the network protocol stack and sent according to the link status, in the order of anchor header block, leading edge reserved block, event digest block, trailing edge control point block, and integrity state block. If no valid fluctuations are detected after backtracking, an empty response record is generated, which retains the anchor header information and integrity status identifier used to identify the unified event identifier, but does not send the leading edge reserved block and the trailing edge control point block, thus clearly informing the central side that there is no signal in the channel, rather than that network packet loss has occurred. In this embodiment, to avoid the link status event quality parameters and recovery quality results being merely general descriptions, the rules for their generation and use are further defined as follows: the link status is divided into at least three levels: good, limited, and congested, based on available bandwidth, link jitter, and packet loss rate. When the available bandwidth is greater than or equal to 150% of the bandwidth corresponding to the average event generation rate, the link jitter is less than or equal to the preset jitter threshold, and the packet loss rate is less than or equal to the preset packet loss rate threshold, it is judged as good, and the complete block sequence is sent first. When the bandwidth corresponding to the average event generation rate is less than or equal to 100% of the available bandwidth, which is less than 150% of the bandwidth corresponding to the average event generation rate, and the packet loss rate is less than or equal to the preset packet loss rate threshold, it is judged as restricted, and the anchor header block and the leading edge reserved block are given priority, while the trailing edge control point block is allowed to be delayed. When the available bandwidth is less than 100% of the bandwidth corresponding to the average event generation rate or the packet loss rate is greater than the preset packet loss rate threshold, it is judged as congested, and only the anchor header block, the leading edge reserved block, and the integrity state block are given priority for dequeuing. The preset jitter threshold and preset packet loss rate threshold are calibrated by the 95th percentile value measured by the switching network during the in-station commissioning phase, and written into the same operating parameter table for the central side and edge nodes. The event quality parameters for each node include at least the time reliability level, the front edge integrity level, and the signal-to-noise ratio level. The time reliability level is graded based on the time synchronization error, the front edge integrity level is graded based on whether the front edge retention segment is missing sampling points, and the signal-to-noise ratio level is graded based on the ratio of the event peak value to the background standard deviation. The reference node is determined by comparing the time confidence level, the frontier integrity level, and the signal-to-noise ratio level in the central direction, so as to avoid introducing timing deviations by selecting the reference node based on a single image value. To enable the above-mentioned classification rules to be directly implemented, this implementation method further specifies that the time reliability level is divided into four levels: A, B, C and D. Among them, the time synchronization error is no more than 5ns, which is level A; greater than 5ns and no more than 10ns, which is level B; greater than 10ns and no more than 20ns, which is level C; and greater than 20ns, which is level D. The integrity level of the front edge is divided into three levels: A, B, and C. Level A is when there are no missing sampling points in the front edge retention segment, Level B is when there are no more than 2 missing sampling points in the last 20% of the front edge retention segment, and Level C is when there are no other cases. The signal-to-noise ratio is divided into four levels: A, B, C, and D. The ratio of the event peak value to the background standard deviation is not less than 8 for level A, not less than 5 and less than 8 for level B, not less than 3 and less than 5 for level C, and less than 3 for level D. When comparing reference nodes on the central side, the alphabetical order of the time confidence level is compared first, followed by the frontier integrity level, and then the signal-to-noise ratio level. If all three levels are the same, the node whose absolute time of the event arrives at the central side earliest is taken as the reference node. The edge-side transmission process adopts a deterministic block-level state progression: the block sequence corresponding to each unified event identifier is in one of four states locally: pending first transmission, pending confirmation, pending retransmission, and completed and closed; after the anchor header block is first transmitted, if a resolvable confirmation of the unified event identifier is received from the central side within the timeout window, the transmission of the leading reserved block and the continuation block is allowed to continue. If the anchor block or the leading edge reserved block is not acknowledged within the timeout window, only these two types of critical blocks will enter the pending transmission state, while the trailing edge control point block will remain suspended to avoid continuing to occupy bandwidth before the critical leading edge has been successfully established; if the central side has clearly returned an empty response record that has been included in the context, the sending end will directly skip the leading edge reserved block and the trailing edge control point block for this channel and enter the completed closure state. When multiple unified event identifiers are queued simultaneously, the sending control module dequeues them in the order of priority of critical blocks, priority of earlier events, and priority of events with fewer retransmissions. Only after the anchor header block and the leading edge reserved block of the new event have been dequeued can the trailing edge control block of the old event continue to occupy the remaining bandwidth, thereby suppressing the long-term blocking of the critical leading edge of the new event by the retransmission of the old event that exceeds the data volume threshold. Because edge nodes upload data in chunks via a jittery network, the data blocks received by the central side are usually out of order and accompanied by clock drift between nodes. Direct comparison will lead to systematic timing deviations. The central receiving process writes the scattered data blocks into the corresponding event context buffer according to the unified event identifier; during the merging process, the system reads the event quality parameters reported by each node and selects the node with the highest confidence as the reference node; Perform coarse alignment and translation based on the absolute time of events reported by each node; on the basis of coarse alignment, extract the leading edge retained segment data, perform relevance search, calculate the fine adjustment offset, and complete fine alignment; The complete waveform is restored by interpolation using trailing edge control points. Specifically, the local time offset and amplitude of each trailing edge control point are analyzed. Piecewise linear interpolation is performed between two adjacent control points according to the slope determined by the difference between the two to reconstruct the missing sampling points. The leading edge retained segment is then spliced ​​with the restored continuous trailing edge waveform segment to generate a multi-channel unified event result and missing block information. To ensure that the fine alignment step has a clear input-output mapping, the correlation search is limited to: using the leading edge reserved segment of the reference node as the baseline sequence and the leading edge reserved segment of the node to be aligned as the sliding sequence, sliding by one sampling point within a search range of no more than 25% of the length of the leading edge reserved segment before and after the coarse alignment position, calculating the normalized correlation at each sliding position, and taking the sliding amount with the largest normalized correlation as the fine adjustment offset; To eliminate the interference of background DC bias on similarity, the background mean was subtracted from both the baseline sequence and the sliding sequence before calculation, resulting in the mean-reduced sequence. and The normalized correlation coefficient is calculated using the following formula: in, The sampling point number of the sequence. This is the slip; before performing the above calculations, the system checks the energy term in the denominator in parallel. or If the denominator is zero when the corresponding waveform is in a pure DC state, the system will directly set the sliding position to 0. The value is assigned to 0; If multiple values ​​have the same maximum value, the one with the smallest absolute offset is selected; if the maximum normalized correlation is less than 0.6, fine alignment is abandoned and only coarse alignment results are retained, while a low correlation label is added to the channel. If the leading edge reserved block of a node is permanently lost during transmission, making fine alignment impossible, the center side will only perform waveform splicing based on the coarse alignment result and add a timing degradation mark to the recovery result of the channel. The central side establishes event context lifecycle control for each unified event identifier: when the first anchor header block arrives, an event context is created and a waiting window is started; Within the waiting window, the block arrival order and recoverability level of the channel are updated for each block received; if the anchor block has arrived but the leading edge reserved block has been missing for a long time, the channel is marked as coarse alignment only. If the anchor head block, the front-end reserved block, and the integrity status block have all arrived, the channel enters the decisionable state. When all known participating nodes have entered the decisionable state, or when the waiting window expires and at least one node has entered the decisionable state, the central side immediately triggers this merge and reconstruction, instead of waiting indefinitely for subsequent late blocks. For delayed data blocks that arrive after the waiting window, they are only included in the completion process before the unified event identifier has been archived. Once the unified event identifier has been archived and output, the late block only updates the missing block statistics and does not rewrite the output front-end fine alignment results, so as to ensure that the same event number corresponds to only one deterministic time-series version in the upper-level positioning algorithm. Fixed edge detection and compression parameters cannot adapt to background noise and network fading that change over time. Without a callback mechanism, this may lead to continuous waveform distortion or network congestion. Therefore, after completing the event reconstruction, the central side extracts the fine alignment offset, missing block information and recovery quality results, generates parameter update instructions and sends them to the corresponding edge nodes; After receiving the instruction, the edge node parses and refreshes the running parameter configuration in its local memory, including the event threshold, cooperative threshold reduction coefficient, leading edge retention length, trailing edge compression error, block length, and retransmission mode. These updated parameters will directly affect the detection and flow control logic in the next sampling cycle. Specifically, the parameter update adopts a segmented closed-loop rule: when the absolute value of the fine alignment offset of a node for multiple consecutive events exceeds 20% of the current front edge retention length of the node, the front edge retention length is increased by a step of not less than 10% until the fine alignment offset falls back into the above ratio range. When the missing block information shows that the proportion of missing anchor header blocks or front-end reserved blocks in the most recent statistics window exceeds the preset missing block rate threshold, shorten the block length and switch the retransmission mode of the corresponding block to priority retransmission. Edge nodes locally calculate the maximum absolute amplitude difference between the original trailing edge waveform and the piecewise linear interpolated waveform as the trailing edge recovery error and attach it to the event digest block for uploading. When the central side parses that the trailing edge recovery error is lower than the preset recovery error threshold for multiple consecutive events, the trailing edge compression error is allowed to be moderately relaxed to reduce the transmission load. Since the collaborative threshold reduction coefficient is used to multiply the independent judgment threshold, the smaller its value, the lower the corresponding collaborative threshold. When the discharge position calculated by the central side through the multi-channel time difference exceeds the boundary of the GIS physical equipment, thus determining that the false alarm event rate caused by external interference increases, and the background noise increases synchronously, irrelevant triggering should be suppressed, specifically by increasing the event threshold or increasing the collaborative threshold reduction coefficient. Conversely, when the proportion of empty response records returned by multiple channels within the coordinated time window increases, i.e. the weak signal false alarm rate increases, and the available network bandwidth meets the preset conditions, the capture capability should be improved, specifically by lowering the event threshold or reducing the coordinated threshold reduction coefficient. The aforementioned preset missing block rate threshold, preset recovery error threshold, and consecutive event counting threshold are all calibrated through offline playback tests during system deployment. The missing block rate threshold can be selected from 1% to 10%, the recovery error threshold can be selected from no more than 5% of the peak value of the reference waveform, and the consecutive event counting threshold can be selected from 3 to 10 times to ensure that the parameter adjustment is both responsive and does not oscillate excessively. The update range of each parameter is limited as follows: the adjustment range of the single leading edge retention length is preferably 10% to 25% of the current value; the adjustment range of the single block length is preferably 10% to 20% of the current value; the adjustment range of the single collaborative threshold reduction coefficient is preferably 0.02 to 0.10; the adjustment range of the single event threshold is preferably 5% to 15% of the current threshold value; and the relaxation range of the single trailing edge compression error is preferably 10% to 15% of the current allowable error value. If the same parameter is adjusted in opposite directions in two consecutive updates, the second adjustment will be halved to suppress parameter oscillation; if the edge node does not receive parameter update instructions from the center side for several consecutive times, it will automatically fall back to the system's initial preset safety baseline parameters to prevent long-term configuration failure due to a single abnormal instruction. To ensure that the evaluation results from the central side can be deterministically implemented at the edge, the parameter update in this implementation does not immediately rewrite all the events being processed the moment the instruction arrives. Instead, it adopts an event boundary effectiveness rule: if a node still has a unified event identifier that is in the process of sending or waiting to be retransmitted, the new parameter first enters the pending effectiveness state, and then switches to the current effective parameter as a whole after the batch of events ends. Immediate action is only allowed when a node is idle or only has background statistics tasks remaining. If multiple control rules conflict in the same round of updates, they are executed in the following order: time reliability takes precedence over critical block reachability, critical block reachability takes precedence over trailing edge compression rate, and false alarm suppression takes precedence over false negative compensation. That is, time alignment and reliable arrival of the critical front are given priority. On this basis, trailing edge load and cooperative sensitivity are adjusted to avoid congestion amplification caused by threshold reduction and block lengthening at the same time when the link deteriorates.

[0023] As a preferred implementation, in S1, the background mean, background standard deviation and first-order difference standard deviation are continuously recorded at each edge node, and the background statistical results are used as the basis for subsequent amplitude determination, first-order difference determination and signal-to-noise ratio calculation. In S2, a pre-truncation length and a post-truncation length are also set for the effective waveform range of the event. The pre-truncation length is determined by the product of the event rising edge parameter and the first preset scaling factor, and the post-truncation length is determined by the product of the event pulse width parameter and the second preset scaling factor. The event rising edge parameter corresponds to the time length for the waveform to rise from the 10% peak value to the 90% peak value; the event pulse width parameter corresponds to the time length for the waveform amplitude to fall back to the preset background standard deviation range from its own peak point; the first preset scaling factor has a value range of [2,5], and the second preset scaling factor has a value range of [1,3]. In S3, if no confirmation is received from the central side, the anchor head block and the leading edge reserved block are retransmitted after a timeout; when the missing block information returned by the central side contains the trailing edge control point block number, the corresponding block is retransmitted; after receiving the missing block information returned by the central side, the edge node determines whether to continue retransmission or end the sending process corresponding to the current unified event identifier. In S4, a hierarchical output is generated for the recovery results of each node. The hierarchical output includes independent triggering results, collaborative triggering results, empty response records, and low-confidence results, and is associated with the output of multi-channel unified event results. Among them, the independent triggering result is when only a single node exceeds the triggering threshold, the collaborative triggering result is when multiple nodes complete fine alignment under a unified event identifier, the empty response record is used to characterize the state of a node receiving a lock notification but not detecting a valid waveform, and the low-confidence result is when the event confidence is less than the preset confidence threshold. When the event confidence is greater than or equal to the preset confidence threshold, it is determined to be a valid and reliable result.

[0024] If a fixed absolute value is used as the criterion for judging waveform changes, frequent false alarms or missed alarms are likely to occur when the electromagnetic background noise at the GIS site drifts with the operating status of the equipment. Therefore, in this implementation, during the data acquisition phase, the edge nodes continuously calculate the background mean, background standard deviation, and first-order difference standard deviation through a sliding window within a stable interval where no events are triggered. The calculated background statistics are periodically updated in the register of the threshold calculation unit as a dynamic benchmark for subsequent determination of waveform amplitude deviation, first-order differential abrupt change, and signal-to-noise ratio calculation. If the background standard deviation exceeds the reasonable range in a single statistical period due to strong external interference, the system will block the update and maintain the statistical record of the previous effective period to prevent abnormal changes in the threshold baseline. The preferred stable interval is the interval in which no event has been triggered and the absolute value of the first difference of multiple consecutive sampling points is lower than the current value of the standard deviation of the first difference; the sliding window length is selected according to the sampling rate and the rate of change of the field noise, so that a single window covers the background noise change within at least one complete power frequency cycle; The reasonable range is determined by the equipment no-load calibration results, preferably limited to 0.5 to 2 times the background standard deviation of the previous effective period. Values ​​exceeding this range are considered abnormal statistics and are discarded. To avoid weak fluctuations at the end of an event or within a locked time window being incorrectly included in the background samples, this implementation method stipulates that once a sampling segment enters the candidate event search interval, a locked time window, or a backtracking scan freeze interval, the sampling segment will not participate in the background statistics update before being released. New background statistics are only allowed to be generated when the stationary interval condition is met continuously and the number of effective samples reaches more than 80% of the target length of the current sliding window; if the number of effective samples in the current period is insufficient, the background statistics of the previous effective period are maintained and the current period is recorded as a state of insufficient samples. When the insufficient sample state occurs more than a preset number of times, the system only increases the time span of the background window by a preset step size, instead of directly lowering the event judgment threshold, so as to avoid the threshold dropping suddenly due to the short-term lack of quiet samples. Different types of transient events have significantly different rise edge steepness and duration. Fixed-length waveform truncation may result in long pulses being truncated or short pulses generating data redundancy. The feature extraction module sets dynamic pre-truncation and post-truncation lengths for the effective waveform range of the event; the system calculates the time length for the waveform to rise from 10% peak value to 90% peak value as the rising edge parameter and maps it to the pre-truncation length; at the same time, it determines the post-truncation length based on the pulse width parameter of the event amplitude fallback. The defined dynamic range will control the memory read pointer to accurately move the required data from the circular cache; if the calculated truncation length exceeds the system's preset maximum cache capacity limit, then the truncation will be forced to be performed according to the maximum capacity limit, prioritizing the integrity of the leading edge data. To ensure that the truncation rules under different event types can be directly implemented, the pre-truncation length is preferably 2 to 5 times the rising edge parameter, and not less than 8 sampling points corresponding to the minimum clock resolution of the system; the post-truncation length is preferably 1 to 3 times the event pulse width parameter. When the ring cache capacity is insufficient to simultaneously satisfy both the pre-truncation length and the post-truncation length, the entire pre-truncation length is retained first, and the post-truncation length is compressed according to the remaining capacity. To avoid limiting the length determination based on parameters to a functional description, this implementation specifies that when the rising edge parameter is no more than 2 sampling points, the pre-truncation length is fixed at 8 sampling points. When the rising edge parameter is greater than 2 sampling points and not greater than 5 sampling points, the pre-truncation length is 3 times the rising edge parameter; when the rising edge parameter is greater than 5 sampling points, the pre-truncation length is 5 times the rising edge parameter, but not exceeding half of the maximum cache capacity. Correspondingly, when the event pulse width parameter is no more than 20 sampling points, the post-truncation length is 1 times the event pulse width parameter; when the event pulse width parameter is greater than 20 sampling points but no more than 100 sampling points, the post-truncation length is 2 times the event pulse width parameter. When the event pulse width parameter is greater than 100 sampling points, the post-truncation length is 3 times the event pulse width parameter, but the total length after truncation must not exceed the current available capacity of the ring buffer. The above multiplier rules are stored in the node parameter table. Edge nodes only need to follow the table lookup results and do not require manual determination at runtime. When the network packet loss rate exceeds the preset packet loss rate threshold, retransmitting all data blocks indiscriminately will increase the network data volume to exceed the preset bandwidth limit; if retransmission is completely abandoned, the central side will lose critical time difference comparison information. Therefore, the edge node's transmission control module enables a timeout retransmission mechanism for the anchor header block and the leading edge reserved block, and automatically triggers retransmission when no confirmation is received from the central side. For trailing control point blocks, retransmission is performed based on the missing block information returned by the central side. After receiving the missing block information, the edge node evaluates whether to continue retransmission based on the current link status. If the available bandwidth of the link is lower than the maintenance threshold, the node actively terminates the sending process corresponding to the current unified event identifier and clears the local cache. The waiting time for timeout retransmission is determined by adding twice the delay jitter to the average round-trip delay in the most recent statistical period, and is limited to a range of no less than 1 event transmission period and no more than 3 event transmission periods; the maximum number of retransmissions for the same anchor block and leading edge reserved block is preferably 2 to 5 times. The threshold is maintained at 1.2 times the average bit rate of the current block to be retransmitted, so as to ensure that the retransmission will not crowd out the basic transmission resources of subsequent new events; The edge node determines whether to continue retransmission or end the current transmission process in the following order: first check whether the missing block information contains an anchor header block or a front-end reserved block. If it does, and the cumulative retransmission count of the block has not reached the upper limit, then retransmission is continued first. If only trailing control point blocks are included, then check whether the available bandwidth of the current link is not lower than the maintenance threshold and whether the number of new events in the current sending queue is less than the preset queuing limit. If both conditions are met, continue to retransmit; otherwise, end the sending process corresponding to the current unified event identifier. The preset queuing limit is preferably set at 50% to 80% of the local sending queue capacity of the edge node to prevent the retransmission of old events from occupying the resources for sending new events for a long time. Timeout retransmission is not unconditionally repeated periodically, but adopts an acknowledgment-driven state inversion rule: when the anchor header block is sent, if an acknowledgment is received from the central side that the corresponding event context has been established, the block immediately changes to the acknowledged state and no longer participates in the subsequent retransmission count; when the front edge reserved block is sent, if the central side returns an acknowledgment that fine alignment can be performed, the block changes to the acknowledged state. If only a missing block information is received without corresponding block confirmation, the block remains in the pending retransmission state until the retransmission limit is reached or the current unified event identifier is locally closed; if a critical block under a unified event identifier has been confirmed but the subsequent control point block continues to be missing, the node will only continue to retransmit when the link is in a good or limited state; once the link enters a congested state, the subsequent retransmission of the unified event identifier will be terminated immediately, the missing block record will be retained and the transmission resources will be released. When outputting the reconstructed waveform, the central side classifies and identifies the signals according to the trigger source and confidence level, and outputs them to the subsequent positioning algorithm; the central side generates hierarchical outputs for the recovery results of each node based on the status flags reported by each node and the alignment errors in the reconstruction process. The tiered output specifically includes: independent triggering results where only a single node exceeds the triggering threshold, collaborative triggering results where multiple nodes complete fine alignment under a unified event identifier, empty response records where a lock notification is received but no valid waveform is detected, and low-confidence results where the event confidence level is lower than a preset confidence level threshold. These hierarchical labels are stored in a shared database as metadata and are associated with the unified event results from multiple channels. This allows the upper-layer positioning algorithm to perform differentiated analysis based on the label weights. If the recovery results of all nodes under a certain event are determined to be low-confidence, the system will intercept the output of that event to the positioning algorithm and trigger the self-check process of the underlying sensors or synchronization links. The event credibility threshold is determined by a combination of item-by-item veto and comprehensive judgment: when any node has a time-series degradation mark, a permanent missing leading edge reserved segment, or a fine alignment offset exceeding 50% of the leading edge reserved length of that node, it is directly judged as low credibility; If the aforementioned rejection conditions are not triggered, a comprehensive classification is then performed based on the time reliability level, frontier integrity level, signal-to-noise ratio level, and block recovery integrity level. At least three of these criteria must be at the normal level for a result to be considered non-low reliability. This rule can prevent situations where the overall score is acceptable but the key frontier is severely misaligned and is still misjudged as usable for localization. To ensure the reproducibility of the comprehensive classification, the completeness level of missing block recovery is further defined as follows: no missing blocks and no interpolation recovery section is the normal level; there are trailing control point blocks that are retransmitted but are eventually recovered is the acceptable level; there are incomplete trailing recovery sections and the level is downgraded. The criteria for calculating the normal level are as follows: when the time confidence level reaches A or B, the frontier integrity level reaches A, the signal-to-noise ratio level reaches A or B, and the block recovery integrity level is normal, it is counted as one normal level. If there are fewer than three normal levels, a low confidence result is output. The central side adopts a first-rejection-then-classification order control for the hierarchical output: first check whether the low confidence rejection condition is hit, if it is hit, directly output the low confidence result and stop its participation in collaborative localization; If the rejection condition is not met, then check whether there are two or more nodes that have completed fine alignment under the same unified event identifier. If so, output the collaborative triggering result. If only one node forms a valid reconstruction and the remaining nodes have empty response records, do not participate, or fail to reconstruct, then the independent trigger result is output; if a node receives a lock notification but only returns an empty response record, then the node is marked as an empty response record separately and output alongside the results of other nodes under the same unified event identifier; the judgment level of low confidence results remains unchanged, and the empty response record is output as a state of no waveform detected.

[0025] The embodiments described above do not constitute a limitation on the scope of protection of this technical solution. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the above embodiments should be included within the scope of protection of this technical solution.

Claims

1. A GIS electric field data transmission method based on edge computing, characterized in that, Includes the following steps: S1. Collect electric field waveforms from multiple measurement points in the GIS, and obtain clock deviation, timing error, link bandwidth, link jitter and packet loss rate of each edge node to form basic node data; S2. Based on the node basic data and combined with the preset local trigger threshold, perform candidate transient event detection, effective waveform truncation and event absolute time determination on the waveform of each node to generate a local event package. The local event package includes at least a leading edge reserved segment, a trailing edge control point and an event summary. S3. Generate a unified event identifier based on the absolute time of the event and the node identifier of the corresponding edge node, and organize the local event package according to the unified event identifier; When the candidate transient event meets the preset collaborative triggering conditions, a locking notification containing a preset locking time window is sent to other edge nodes, so that the receiving node intercepts and caches the event within the locking time window according to the reduced local triggering threshold; then, the local event packet is split into event blocks according to the link status and sent. S4. The central side merges the event blocks received from each node according to the unified event identifier, and performs coarse alignment, fine alignment of the leading edge and recovery of the trailing edge on the merging result to generate the corresponding multi-channel unified event result, missing block information and recovery quality information. S5. The central side generates a parameter update instruction based on the multi-channel unified event result, the missing block information and the recovery quality information and sends it to the corresponding edge node. The edge node then updates its local trigger threshold and preset running configuration parameters accordingly.

2. The GIS electric field data transmission method based on edge computing according to claim 1, characterized in that, In S1, the original waveforms of multiple measurement points of the GIS are accessed through a multi-channel electro-optical field acquisition device, and the time synchronization relationship between the station master clock and each edge node is established to obtain the clock deviation and time synchronization error of each edge node. At the same time, the link bandwidth, link jitter and packet loss rate are measured to form the basic data record of the node.

3. The GIS electric field data transmission method based on edge computing according to claim 1, characterized in that, In S2, candidate transient events are detected based on preset waveform characteristic parameters; The effective waveform interval is extracted based on the start point, main peak point, and end point of the candidate transient event; The absolute time of the event is determined based on the event start-up information and clock offset, and a local event packet is generated. The event summary includes one or more of the event magnitude, duration, and absolute event time.

4. The GIS electric field data transmission method based on edge computing according to claim 1, characterized in that, In S3, local event packets are mapped to a sending queue indexed by a unified event identifier. After a candidate event is detected, a locking notification including the unified event identifier, the absolute time of the event, the preset event strength, and the preset locking time window is sent to other edge nodes. This causes other edge nodes to lower their local trigger threshold, retain the circular buffer, and generate associated event packets or empty response records within the corresponding time window. Then send them in the order of anchor header block, leading edge reserved block, event digest block, trailing edge control point block, and integrity status block.

5. The GIS electric field data transmission method based on edge computing according to claim 1, characterized in that, In S4, the central side converts the received event blocks into event contexts merged according to a unified event identifier, determines reference nodes based on the event quality parameters of each node, performs coarse alignment based on the absolute time of events reported by each node, and performs fine alignment based on the leading edge reserved segment to obtain fine alignment offset. Then, it combines the trailing edge control points to restore the complete waveform and generate multi-channel unified event results and missing block information.

6. The GIS electric field data transmission method based on edge computing according to claim 5, characterized in that, In S5, based on the fine alignment offset determined by the fine alignment of the center side, as well as the missing block information and recovery quality information, the local trigger threshold, cooperative reduction threshold coefficient, leading edge retention length, trailing edge compression error, block length, and retransmission mode of anchor head block and leading edge retention block of the edge node are updated.

7. The GIS electric field data transmission method based on edge computing according to claim 2, characterized in that, In S1, the background mean, background standard deviation and first-order difference standard deviation are continuously recorded at each edge node, and the background statistical results are used as the basis for subsequent amplitude determination, first-order difference determination and signal-to-noise ratio calculation.

8. The GIS electric field data transmission method based on edge computing according to claim 3, characterized in that, In S2, a pre-truncation length and a post-truncation length are also set for the effective waveform range of the event. The pre-truncation length is determined by the product of the event rising edge parameter and the first preset scaling factor, and the post-truncation length is determined by the product of the event pulse width parameter and the second preset scaling factor. The event rising edge parameter corresponds to the time length during which the waveform rises from the 10% peak value to the 90% peak value; the event pulse width parameter corresponds to the time length during which the waveform amplitude falls back to the preset background standard deviation range from the autonomous peak point; the first preset scaling factor has a value range of [2,5], and the second preset scaling factor has a value range of [1,3].

9. The GIS electric field data transmission method based on edge computing according to claim 4, characterized in that, In S3, if no confirmation is received from the central side, the anchor head block and the leading edge reserved block are retransmitted after a timeout. When the missing block information returned by the central side includes the trailing edge control point block number, the corresponding block is retransmitted. After receiving the missing block information returned by the central side, the edge node determines whether to continue retransmission or terminate the transmission process corresponding to the current unified event identifier.

10. The GIS electric field data transmission method based on edge computing according to claim 5, characterized in that, In S4, a hierarchical output is also generated for the recovery results of each node. The hierarchical output includes independent triggering results, collaborative triggering results, empty response records and low confidence results, and is associated with the output of multi-channel unified event results. The independent triggering result is when only a single node exceeds the triggering threshold; the collaborative triggering result is when multiple nodes complete fine alignment under a unified event identifier; the empty response record is used to characterize the state where a node receives a lock notification but does not detect a valid waveform; the low confidence result is when the event confidence is less than a preset confidence threshold; and when the event confidence is greater than or equal to the preset confidence threshold, it is determined to be a valid confidence result.

Images