Methods for unmanned operation of power data monitoring and emergency response in substations

By generating equipment baselines and reverse evidence in unmanned substation scenarios, and combining Bayesian odds recursion and MaxWalkSAT flip search, the problem of erroneous responses in unmanned substations was solved, achieving more accurate emergency response and data transmission.

CN123073771APending Publication Date: 2026-11-10ZHONGJI (TIANJIN) ENGINEERING CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202611431711.3
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2026-09-15
Publication Date
2026-11-10

AI Technical Summary

Technical Problem

In unmanned substation scenarios, existing technologies cannot effectively distinguish between positive anomaly data and counter-evidence data, leading to frequent false responses. Especially when communication link bandwidth is limited and in the early stages of anomalies, multiple types of data are sent up in a concentrated manner. The edge data received by the master station lacks counter-evidence data structure, making it impossible to accurately determine the cause of the fault.

Method used

By generating a device baseline band at the station and generating a counter-evidence packet from the alarm packet, and using Bayesian odds recursion and MaxWalkSAT flip search to generate a resolution packet, edge data transmission packets of alarm packets and counter-evidence packets are formed, reducing false responses caused by communication anomalies and environmental disturbances.

Benefits of technology

It effectively reduces false responses caused by communication retransmissions, maintenance work, and disturbances at measurement points, reduces the need for the main station to repeatedly retrieve the full monitoring stream, and improves the accuracy and efficiency of emergency response.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN123073771A_ABST
    Figure CN123073771A_ABST
Patent Text Reader

Abstract

This invention discloses a method for power data monitoring and emergency response in unmanned substations, specifically involving the field of power data edge computing. The method involves acquiring monitoring streams, link logs, and work ticket records at the substation, sequentially generating alarm pieces, counter-evidence pieces, post-verification pieces, and resolution pieces. Based on the resolution pieces, it further divides them into direct response pieces, supplementary verification pieces, false alarm retention pieces, and ordinary record pieces before encapsulating them into edge data transmission packets. This invention forms equipment baseline bands by eliminating disturbed equipment windows at the substation, generates counter-evidence pieces based on alarm pieces, and generates resolution pieces through Bayesian odds recursion and MaxWalkSAT flip search. This addresses the problem of false responses easily caused by the lack of counter-evidence data structures in edge data transmission results under unmanned operation scenarios.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This invention relates to the field of power data edge computing technology, and more specifically, to a method for unmanned operation of power data monitoring and emergency response in substations. Background Technology

[0002] In the scenario of unmanned operation of substations, the current, voltage, switch quantity, protection action, equipment temperature, partial discharge signal, video recognition results and environmental quantity in the station are usually collected by the station-end acquisition device and then uploaded to the main station through the edge gateway. The main station will then trigger remote review, dispatching or emergency plan invocation according to over-limit alarm, multi-source consistency or alarm level. The above methods can meet the needs of routine remote monitoring. However, when there are no on-duty personnel at the station, the communication link at the station has bandwidth limitations, and multiple types of data are being centrally uploaded in the early stages of an anomaly, the edge data transmission results received by the main station often only reflect the abnormal data itself, such as temperature rise exceeding limits, partial discharge enhancement, smoke in the image, remote signaling changes, or equipment offline. They lack sensor self-test status, communication retransmission records, changes in adjacent measuring points, maintenance operation periods, sources of environmental disturbances, and historical drift records to eliminate false responses. In this case, the same type of non-fault factors will be amplified by multiple monitoring channels, which manifests as an increase in the number of remote alarms, repeated dispatching of the same equipment, cancellation of anomaly confirmation, and inconsistency between video or temperature alarms and the actual equipment status. This is because the existing uploaded objects mainly organize data around the establishment of anomalies and have not simultaneously formed a counter-evidence data structure at the edge side that can refute the conclusion of anomalies. Therefore, the technical problem to be solved in this application is how to construct positive anomaly data and counter-evidence data together into a data object that can be used for remote emergency response judgment during the edge data transmission process of unmanned substation operation, so as to reduce false responses caused by sensor drift, communication anomalies, environmental disturbances or maintenance interference. Summary of the Invention

[0003] To overcome the aforementioned deficiencies of the prior art, embodiments of the present invention provide a method for unmanned substation power data monitoring and emergency response. This method solves the problems mentioned in the background art by forming an equipment baseline band by removing disturbed equipment windows at the station end, generating a counter-evidence band by reverse lookup of alarm bands, and generating a resolution band by Bayesian odds recursion and MaxWalkSAT flip search.

[0004] To achieve the above objectives, the present invention provides the following technical solution: a method for unmanned substation power data monitoring and emergency response, comprising: S1. At the station, obtain monitoring stream, link log and work ticket record, form device windows according to device number and time window number, and generate device baseline band after removing disturbed device windows by retransmission count and work occupancy value. Then, extract trigger segments from device windows that exceed the device baseline band and generate alarm pieces. S2. For the device number, time window number and trigger segment of the alarm video, generate changes in adjacent measurement points by adjacent point difference, generate transmission anomalies by retransmission count, generate job occupancy by overlap duration, and merge them to generate counter-evidence video. S3. Generate a fault likelihood sequence based on the trigger segment and the equipment baseline band, generate a counter-evidence placeholder sequence based on changes in adjacent measurement points, transmission anomalies and job occupancy, and perform Bayesian odds recursion on the fault likelihood sequence and the counter-evidence placeholder sequence to generate a posterior piece. S4. Form a node table with alarm pieces, counter-proof pieces and posterior pieces. Establish counter-proof closure edges based on the counter-proof values ​​in the counter-proof pieces. Establish mutually exclusive edges based on the event nodes and counter-proof nodes in the node table. Perform MaxWalkSAT flip search according to the number of collisions, closure difference and flip loss to generate resolution pieces. S5. Based on the event node status, counter-evidence node status, mutual exclusion edge status, and event node weight in the digestion slice, divide it into direct response slice, supplementary verification slice, false alarm retention slice, and ordinary recording slice. Then, encapsulate the division result, alarm slice, counter-evidence slice, and necessary original fragments into an edge data transmission packet and send it to the main station to generate emergency response results.

[0005] In a preferred embodiment, S1 includes: S1-1. In the station-end monitoring stream, the sampled values ​​of the same device are merged by device number, and the sampled values ​​under the same device number are divided by time window number to form device window segments. S1-2. In the link log, count the number of retransmitted packets using the device number and time window number. In the job ticket record, count the overlap between the job ticket effective time and the device window segment using the device number and time window number. Use the number of retransmitted packets and the overlap duration as the retransmission count and job occupation value, respectively, to generate the window segment occupation. S1-3. Form a device baseline band using device windows with zero retransmission count and zero job occupancy value under the same device number, and use device windows with zero retransmission count, zero job occupancy value and exceeding the device baseline band as trigger segments to generate alarm clips containing device number, time window number and trigger segment.

[0006] In a preferred embodiment, S2 includes: S2-1. Read the device number, time window number and trigger segment from the alarm chip. Take the sampled value of the trigger segment and the sampled value of the adjacent measuring point under the same time window number in the monitoring stream. Subtract the sampled value of the trigger segment from the sampled value of the adjacent measuring point to form the adjacent point difference. S2-2. Take the direction of the trigger segment's crossover relative to the equipment baseline as the trigger direction, and compare the positive and negative directions of the adjacent point difference with the trigger direction. When the positive and negative directions of the adjacent point difference are consistent with the trigger direction, a change in the same direction is formed. When the positive and negative directions of the adjacent point difference are opposite to the trigger direction, a change in the opposite direction is formed. Changes in adjacent measuring points are generated by changes in the same direction or in the opposite direction.

[0007] In a preferred embodiment, S2 further includes: S2-3. Count the number of retransmissions within the same time window in the link log along the device number and time window number, and treat the number of retransmissions as a transmission anomaly. Count the overlap between the job ticket effective time and the trigger segment in the job ticket record along the device number and time window number, and treat the overlap duration as job occupancy. S2-4. According to the same device number and the same time window number, merge the changes of adjacent measurement points, transmission anomalies and job occupancy with the trigger segments in the alarm video to generate a counter-evidence video.

[0008] In a preferred embodiment, S3 includes: S3-1. Read the trigger segment and device baseline band in the alarm chip, calculate the cross-boundary distance of the sampled value of the trigger segment relative to the device baseline band according to each time window number in the trigger segment, and concatenate the cross-boundary distances in the order of the time window numbers to generate a fault likelihood sequence. S3-2. Read the changes, transmission anomalies, and job occupancy of adjacent measurement points in the counter-evidence slice. Map the time window number where the reverse change is located, the time window number where the transmission anomaly is located, and the time window number covered by the job occupancy to the same time window number of the fault likelihood sequence, and generate the counter-evidence occupancy sequence.

[0009] In a preferred embodiment, S3 further includes: S3-3. Take the out-of-bounds distance in the fault likelihood sequence that is not occupied by the counter-evidence placeholder sequence as the positive likelihood segment, and take the out-of-bounds distance in the fault likelihood sequence that is occupied by the counter-evidence placeholder sequence as the counter-evidence deduction segment. Perform Bayesian odds recursion on the positive likelihood segment and the counter-evidence deduction segment in the order of time window number to generate the posterior recursion sequence. S3-4. Take the posterior value of the last window number in the posterior recursive sequence as the corrected posterior value, and generate a posterior chip with the device number, window number and corrected posterior value.

[0010] In a preferred embodiment, S4 includes: S4-1. Read alarm chips, counter-proof chips, and post-proof chips. Form event nodes with the same device number and the same time window number. Form neighbor counter-proof nodes, link counter-proof nodes, and job counter-proof nodes with changes in adjacent measurement points, transmission anomalies, and job occupancy in the counter-proof chips, respectively. Use the corrected post-proof value as the weight of the event nodes. When there are event nodes and at least one counter-proof node under the same device number and the same time window number, form a node table. Otherwise, directly form the event node into a dissolution chip. S4-2. The neighboring point counter-proof value is formed according to the number of time windows of reverse change in the neighboring counter-proof node, the link counter-proof value is formed according to the number of retransmissions in the link counter-proof node, and the job counter-proof value is formed according to the overlap duration in the job counter-proof node. When the neighboring point counter-proof value, the link counter-proof value, and the job counter-proof value are all non-zero, a counter-proof closure edge is established between the three counter-proof nodes. Otherwise, only the counter-proof nodes with non-zero values ​​are retained and the establishment of counter-proof closure edges is prohibited.

[0011] In a preferred embodiment, S4 further includes: S4-3. Establish mutual exclusion edges between the event node and the neighboring counter-proof node, the link counter-proof node, and the job counter-proof node, respectively. The number of nodes at both ends of the mutual exclusion edge that are simultaneously retained is used to form the conflict number. The number of nodes at both ends of the counter-proof closure edge that are inconsistent in state is used to form the closure difference. The event node weight of the flipped event node is used to form the flip loss. When the conflict number is not zero, generate a conflict cost sequence in the order of conflict number, closure difference, and flip loss. Otherwise, directly form a resolution piece by the event node state, the counter-proof node state, and the mutual exclusion edge state. S4-4. The MaxWalkSAT conflict resolution algorithm for Markov logic networks is adopted. The state flipping is performed on each event node, neighbor node, link node, and job node, and the conflict cost sequence is recalculated after each flipping. If the current conflict cost sequence is less than the previous conflict cost sequence in the order of conflict number, closure difference, and flipping loss, the current flipping is retained; otherwise, the current flipping is withdrawn. When no flipping is retained in a continuous round of traversal, a resolution piece containing the state of the event node, the state of the node, and the state of the mutually exclusive edge is generated; otherwise, the next round of traversal is entered.

[0012] In a preferred embodiment, S5 includes: S5-1. Read the event node status, proof node status, mutual exclusion edge status and event node weight in the unsolvable piece. When the event node is retained, the nodes at both ends of each mutual exclusion edge are not retained at the same time and the event node weight is not zero, a direct response piece is generated. Otherwise, when the event node is retained and at least one mutual exclusion edge has both ends of the node retained at the same time, a supplementary verification piece is generated. S5-2. When an event node is flipped and at least one counter-evidence node is retained, a false alarm retention slice is generated. Otherwise, the resolution slice that has not generated a direct response slice, a supplementary verification slice, and a false alarm retention slice is generated into a normal record slice, and the partitioning result is generated using the direct response slice, the supplementary verification slice, the false alarm retention slice, and the normal record slice.

[0013] In a preferred embodiment, S5 further includes: S5-3. According to the division results, extract the original sampled value of the trigger segment from the alarm segment, extract the original record of the time window number where the adjacent measurement point changes, transmission anomalies and operation occupancy are located from the counter-evidence segment, and merge the original sampled value of the trigger segment and the original record into the necessary original segment. S5-4. Encapsulate the partitioning results, alarm fragments, counter-evidence fragments, and necessary original fragments into an edge data transmission packet with the same device number and the same time window number, and send the edge data transmission packet to the main station to generate the emergency response result.

[0014] The technical effects and advantages of this invention are as follows: 1. By encapsulating alarm chips and counter-evidence chips together into edge data transmission packets at the station end, the master station can simultaneously obtain changes in adjacent measurement points, transmission anomalies, and operation occupancy when receiving the trigger segment. Therefore, temperature rise exceeding the limit, partial discharge enhancement, or remote signaling change will not be uploaded as an anomaly, but will be accompanied by counter-evidence that can refute the anomaly during the transmission stage, which relatively reduces false responses caused by communication retransmission, maintenance operations, and measurement point disturbances. 2. By eliminating disturbed equipment windows based on retransmission count and job occupancy value, and then forming the equipment baseline band from the undisturbed equipment windows, the trigger segment source avoids the link retransmission and job ticket coverage period; therefore, the equipment baseline band is not affected by communication disturbances or maintenance processes, relatively reducing the occurrence of alarm segments formed by disturbed data. 3. Changes in adjacent measurement points are determined by the difference between adjacent points, and the number of retransmissions is treated as a transmission anomaly, and the overlapping duration is treated as job occupation and merged into a counter-proof piece, so that the counter-proof data and alarm pieces share the same device number and time window number; therefore, subsequent counter-proof occupancy and conflict resolution have the same time index basis. 4. By concatenating out-of-bounds distances into a fault likelihood sequence and mapping reverse changes, transmission anomalies, and job occupancy into a disproving placeholder sequence, the out-of-bounds distance for each time window number enters either the forward likelihood segment or the disproving deduction segment depending on whether disproving exists. Therefore, the a posteriori chip can reflect the state of the trigger segment after it has been weakened by disproving. 5. By constraining neighboring node counter-proof nodes, link counter-proof nodes, and job counter-proof nodes with counter-proof closure edges, and by constraining event nodes and counter-proof nodes with mutual exclusion edges, MaxWalkSAT flip search resolves node states based on the number of conflicts, closure difference, and flip loss; therefore, the resolution piece can distinguish between different states: event retention, counter-proof retention, and conflict not resolved. 6. By classifying the event node state, counter-evidence node state, mutual exclusion edge state, and event node weight in the fragment, and sending them to the main station along with the necessary original fragments, the direct response, supplementary verification, false alarm retention, and ordinary record processing are separated; thus, the repeated retrieval of the full monitoring stream by the main station is relatively reduced. Attached Figure Description

[0015] Figure 1 This is a flowchart of the method steps of the present invention. Detailed Implementation

[0016] 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.

[0017] Refer to the instruction manual appendix Figure 1 The present invention provides a method for unmanned substation power data monitoring and emergency response, comprising: S1. At the station, obtain monitoring stream, link log and work ticket record, form device windows according to device number and time window number, and generate device baseline band after removing disturbed device windows by retransmission count and work occupancy value. Then, extract trigger segments from device windows that exceed the device baseline band and generate alarm pieces. In this embodiment, S1 is used to first eliminate data disturbances caused by communication retransmission and job occupation at the station end, and then use the undisturbed device window segment to form the device baseline band, so that the subsequent alarm segments are not directly generated by single point over-limit, but generated by the device window segment under the same device number and the same time window number after checking the link log and job ticket records. Among them, the station-side monitoring stream includes at least the device number, measurement point number, sampling time, and sampled value; the link log includes at least the device number, message sending time, message sequence number, and retransmission flag; the job ticket record includes at least the device number, job start time, and job end time; the edge nodes divide the sampling time into time windows according to the same time length, and use the device number and time window number as the common index for subsequent segmentation, occupancy calculation, baseline generation, and trigger segment extraction; Furthermore, in this embodiment, the specific processing is as follows: S1-1. In the station-side monitoring stream, the edge node first reads the device number, sampling time, and sampled value from each sampling record. Sampled values ​​with the same device number are grouped into the same device sampling column. Then, the device sampling column is segmented according to the time window number to which the sampling time belongs, so that sampled values ​​with the same device number and falling into the same time window number are grouped into the same device window segment. The device window segment includes at least the device number, time window number, window segment start time, window segment end time, and sampled value within the window. The window segment start time and window segment end time are determined by the time range corresponding to the time window number, and the sampled value within the window is composed of the sampled values ​​within that time range. After this processing, data with different sampling frequencies in the monitoring stream are first converted into device windows organized by device number and time window number. Subsequent retransmission counts, job occupancy values, and device baseline bands are all calculated around the device windows, avoiding direct alarm judgment on discrete sampling points. S1-2. For link logs, edge nodes retrieve message records falling within the same device window time range using the device number and time window number, and count the number of messages corresponding to the retransmission marker, using this number as the retransmission count. For job ticket records, edge nodes read the job start time and job end time under the same device number, calculate the overlap between the job time range (from job start time to job end time) and the window start time to window end time of the device window, and use the duration of the overlap as the job occupancy value; if there is no overlap, the job occupancy value is zero. Subsequently, edge nodes combine the device number, time window number, retransmission count, and job occupancy value into a window occupancy slice. The window occupancy slice is used to mark whether each device window is affected by communication retransmission or job occupancy. A non-zero retransmission count indicates that there is link disturbance in the device window, and a non-zero job occupancy value indicates that there is job interference in the device window. S1-3, Edge nodes read device windows and window occupancy segments under the same device number. Device windows with zero retransmission counts and zero job occupancy are designated as net windows, while those with non-zero retransmission counts or non-zero job occupancy are designated as disturbed device windows. Only net windows are used to form the device baseline band. The device baseline band includes an upper baseline and a lower baseline. The upper baseline is the maximum value of the sampled values ​​within the net window under the same device number, and the lower baseline is the minimum value of the sampled values ​​within the net window under the same device number. For the current window number... Edge nodes can take the net window segment before the current window number to form the upper and lower bounds of the baseline corresponding to the current window number; then, the edge node compares the device window segment with zero retransmission count and zero job occupancy value with the device baseline band; when the sampled value in the device window segment is higher than the upper bound of the baseline or lower than the lower bound of the baseline, the segment containing the sampled value exceeding the device baseline band in the device window segment is extracted as the trigger segment, and an alarm fragment containing the device number, window number and trigger segment is generated; when the device window segment belongs to the disturbed device window segment, it does not participate in the trigger segment extraction; Based on the results of the above steps, the station will not directly use all out-of-bounds sampled values ​​as alarm sources. Instead, it will first use link logs and work ticket records to exclude disturbed equipment windows, and then form equipment baseline bands from undisturbed equipment windows and extract trigger segments. This will give alarm pieces three subsequent traceable fields: equipment number, time window number, and trigger segment, providing a preliminary data foundation for generating counter-evidence pieces, Bayesian odds recursion, and constructing counter-evidence conflict graphs. In practical applications: A substation edge node uses one minute as a time window number to segment the monitoring stream of a main transformer temperature measurement point; if there are two retransmission messages in the link log within the tenth time window, the retransmission count for that time window is not zero and it is not included in the equipment baseline band; if the twelfth time window overlaps with the operation time period of the maintenance work ticket by thirty seconds, the operation occupancy value for that time window is thirty seconds and it is not included in the equipment baseline band; the edge node uses the remaining equipment window segments with zero retransmission count and zero operation occupancy value to form the upper and lower bounds of the baseline; when the sampled value in the fifteenth time window exceeds the equipment baseline band and both the retransmission count and operation occupancy value of that time window are zero, the out-of-bounds sampled segment in the fifteenth time window is extracted as a trigger segment and an alarm segment is formed.

[0018] S2. For the device number, time window number and trigger segment of the alarm video, generate changes in adjacent measurement points by adjacent point difference, generate transmission anomalies by retransmission count, generate job occupancy by overlap duration, and merge them to generate counter-evidence video. In this embodiment, S2 is used to extract data that can overturn or weaken the alarm judgment after the alarm chip has been formed, around the same device number, the same time window number, and the same trigger segment. This process no longer relies solely on the out-of-bounds result of the trigger segment, but instead forms adjacent measurement point changes, transmission anomalies, and job occupancy from the monitoring stream, link log, and work ticket records, so that the alarm chip can simultaneously reflect the device proximity status, communication link status, and job interference status. Adjacent measurement point changes are used to determine whether the trigger segment is disturbed in the same direction as the surrounding measurement points, transmission anomalies are used to determine whether the time window where the trigger segment is located is affected by message retransmission, and job occupancy is used to determine whether the time window where the trigger segment is located overlaps with the station's work time period. Furthermore, in this embodiment, the specific processing is as follows: S2-1. The edge node reads the device number, time window number, and trigger segment from the alarm chip. Based on the device number, it locates the sampling record of the device to which the trigger segment belongs in the monitoring stream. Then, based on the time window number, it reads the sampling values ​​of adjacent measurement points within the same time window number as the trigger segment. Adjacent measurement points are determined by the station-side measurement point adjacency table, which records the measurement point number, adjacent measurement point number, and the device number to which it belongs. The edge node uses the measurement point number to which the trigger segment belongs as the search field and takes the measurement point directly adjacent to it in the measurement point adjacency table as the adjacent measurement point. Multiple adjacent measurement points exist within the same time window number. When sampling points, the differences between the sampling points and the sampling points of the trigger segment are calculated to obtain multiple neighboring point differences. The sampling point of the trigger segment is the sampling point with the largest cross-boundary distance within the trigger segment. The sampling points of adjacent measuring points are the sampling points within the same time window that are at the same sampling time as the trigger segment. If there are no sampling points of adjacent measuring points at the same sampling time, the sampling point of the adjacent measuring point that is closest to the sampling time of the trigger segment within the same time window is taken. The neighboring point difference is formed by subtracting the sampling point of the trigger segment from the sampling point of the adjacent measuring point, and is used to describe the relative offset direction between the trigger segment and the surrounding measuring points within the same time window. S2-2. The edge node reads the device baseline band corresponding to the trigger segment and forms an upper boundary violation when the sampled value of the trigger segment is higher than the upper boundary of the device baseline band, and a lower boundary violation when the sampled value of the trigger segment is lower than the lower boundary of the device baseline band. The upper or lower boundary violation is used as the trigger direction. Subsequently, the edge node compares the positive and negative directions of the neighbor point difference with the trigger direction: when the trigger direction is an upper boundary violation and the neighbor point difference is positive, or when the trigger direction is a lower boundary violation and the neighbor point difference is negative, a same-direction change is formed; when the trigger direction is an upper boundary violation and the neighbor point difference is negative, or when the trigger direction is a lower boundary violation and the neighbor point difference is positive, a reverse change is formed; when the neighbor point difference is zero, no same-direction change or reverse change is formed. If there are multiple neighbor point differences under the same time window number, the edge node generates the corresponding same-direction change or reverse change respectively, and uses the time window number of the reverse change as the neighbor point occupancy basis for the subsequent counter-evidence occupancy sequence. The change of adjacent measurement points includes at least the device number, time window number, trigger direction, neighbor point difference, and change type, and the change type is same-direction change or reverse change. S2-3. Edge nodes retrieve link logs along the device number and time window number in the alarm chip, count the number of packets with retransmission markers within the same device number and time window number, and use this number of packets as the retransmission count, which is considered a transmission anomaly. For job ticket records, edge nodes read the job start time and job end time under the same device number, and calculate the overlap between the job time period formed by the job start time and job end time and the time range of the time window number where the trigger segment is located. The duration of the overlap is used as the overlap duration, which is used as the job occupancy. If there is no overlap, the overlap duration is zero. In this step, the transmission anomaly reflects whether there is a link retransmission within the time window where the trigger segment is located, and the job occupancy reflects whether the time window where the trigger segment is located is within the effective time of the job ticket. The two serve as the data basis for subsequent link counter-verification nodes and job counter-verification nodes, respectively. S2-4. Edge nodes, using the same device number and time window number, merge adjacent measurement point changes, transmission anomalies, and job occupancy with the trigger segment in the alarm cell to generate a counter-proof cell. The counter-proof cell includes at least the device number, time window number, trigger segment identifier, adjacent measurement point changes, transmission anomalies, and job occupancy. Adjacent measurement point changes are derived from the comparison result between the neighbor difference and the trigger direction. Transmission anomalies are derived from the number of retransmissions within the same time window number. Job occupancy is derived from the overlap duration between the job ticket effective time and the time window number where the trigger segment is located. When there are multiple adjacent measurement point changes in the same alarm cell, the counter-proof cell retains the measurement point number, neighbor difference, and change type corresponding to each adjacent measurement point change. When the transmission anomaly is zero or the job occupancy is zero, the counter-proof cell still retains the corresponding fields and sets them to zero, so that subsequent counter-proof placeholder sequences can directly retrieve values ​​according to the fields. Based on the results of the above steps, S2 expands the alarm segment from a single trigger segment to a counter-evidence segment containing changes in adjacent measurement points, transmission anomalies, and job occupancy. This enables S3 to reserve counter-evidence information according to the time window number, and enables S4 to transform the counter-evidence segment into a neighboring point counter-evidence node, a link counter-evidence node, and a job counter-evidence node. The purpose of this process is that trigger segment out-of-bounds no longer directly enters the response judgment, but first cross-checks with adjacent measurement points, link retransmissions, and job occupancy within the same time window number, thereby providing data basis for the division of false alarm retention segments, supplementary verification segments, and direct response segments. In practical applications: A device forms a trigger segment in the fifteenth time window. The sampled value of the trigger segment is higher than the upper limit of the baseline of the device's baseline band, and the trigger direction is upward boundary crossing. The edge node reads the sampled values ​​of adjacent measurement points under the same time window. If the sampled value of the adjacent measurement point is lower than the sampled value of the trigger segment and forms a negative neighbor difference, a reverse change is generated. At the same time, the link log shows that there are three retransmission messages in this time window, and the job ticket record shows that there is a 20-second overlap of job time in this time window. The edge node merges the reverse change, the number of retransmissions, and the overlap duration with the trigger segment into a counter-evidence.

[0019] S3. Generate a fault likelihood sequence based on the trigger segment and the equipment baseline band, generate a counter-evidence placeholder sequence based on changes in adjacent measurement points, transmission anomalies and job occupancy, and perform Bayesian odds recursion on the fault likelihood sequence and the counter-evidence placeholder sequence to generate a posterior piece. In this embodiment, S3 is used to transform the trigger segment in the alarm chip from a single out-of-bounds result into a fault likelihood sequence arranged by time window number, and to transform the adjacent measurement point changes, transmission anomalies, and job occupancy in the counter-proof chip into a counter-proof occupancy sequence on the same time window number, and then recursively calculate the corrected posterior value according to the time window number; the core of this process is that the out-of-bounds distance in the trigger segment that is not counter-proof occupied is used to enhance fault judgment, and the out-of-bounds distance that is counter-proof occupied is used to weaken fault judgment, so that the posterior chip bears both the out-of-bounds strength of the trigger segment and the reverse constraint of the counter-proof chip on the same time window number; Furthermore, in this embodiment, the specific processing is as follows: S3-1, The edge node reads the device number, time window number, trigger segment, and device baseline band from the alarm chip, and calculates the boundary distance item by item according to each time window number within the trigger segment. For each time window number, if the sampled value of the trigger segment is higher than the upper limit of the baseline of the device baseline band, the difference between the sampled value of the trigger segment and the upper limit of the baseline is used as the boundary distance. If the sampled value of the trigger segment is lower than the lower limit of the baseline of the device baseline band, the difference between the lower limit of the baseline and the sampled value of the trigger segment is used as the boundary distance. The trigger segment only contains device windows that exceed the device baseline band, so device windows that have not exceeded the limit are not included in the boundary distance calculation. The edge node then concatenates the boundary distances in the order of the time window numbers to generate a fault likelihood sequence. Each item in the fault likelihood sequence contains the device number, time window number, and boundary distance, which are used to represent the boundary strength of the trigger segment at each time window number. S3-2. Edge nodes read adjacent measurement point changes, transmission anomalies, and job occupancy from the counter-evidence slice and map these three types of counter-evidence information to the same time window number in the fault likelihood sequence. Specifically, when an adjacent measurement point change is a reverse change, the neighboring point occupancy of the time window number where the reverse change occurs is recorded as one; otherwise, the neighboring point occupancy is recorded as zero. When the number of retransmissions in a transmission anomaly is not zero, the link occupancy of the time window number where the transmission anomaly occurs is recorded as one; otherwise, the link occupancy is recorded as zero. When the overlap duration in a job occupancy is not zero, the job occupancy of the time window number covered by the job occupancy is recorded as one; otherwise, the job occupancy is recorded as zero. Edge nodes use the neighboring point occupancy, link occupancy, and job occupancy under the same time window number to form counter-evidence occupancy items, and concatenate each counter-evidence occupancy item in sequence according to the time window number to generate a counter-evidence occupancy sequence. The counter-evidence occupancy sequence and the fault likelihood sequence use the same device number and time window number, so that subsequent recursion can process the out-of-bounds distance and counter-evidence occupancy items at the same time position. S3-3. The edge node aligns the fault likelihood sequence and the counter-evidence occupancy sequence item by item according to the time window number. When the neighbor occupancy, link occupancy, and job occupancy of a certain time window number are all zero, the out-of-bounds distance corresponding to that time window number is taken as the positive likelihood segment. When any one of the neighbor occupancy, link occupancy, and job occupancy of a certain time window number is one, the out-of-bounds distance corresponding to that time window number is taken as the counter-evidence deduction segment. Subsequently, the edge node performs Bayesian odds recursion in the order of time window numbers: starting with the recursion odds of the first time window number, the positive likelihood segment multiplies the recursion odds of the previous time window number by the result of "one plus the out-of-bounds distance", and the counter-evidence deduction segment... The recursive odds of the previous time window number are divided by the result of "one plus the number of counter-proof occupancy slots", where the number of counter-proof occupancy slots is the sum of the occupancy slots of neighboring points, links, and jobs under the same time window number. If there are reverse changes, transmission anomalies, and job occupations simultaneously under the same time window number, the number of counter-proof occupancy slots is three, corresponding to a stronger deduction effect. If there is only one type of counter-proof occupancy slot, the number of counter-proof occupancy slots is one. The edge node converts the recursive odds of each time window number into a posterior value and concatenates them according to the time window number to generate a posterior recursive sequence. The posterior recursive sequence is used to record the degree of fault preservation of the trigger segment after the time-by-time counter-proof deduction. S3-4: Edge nodes read the posterior value of the last window number in the posterior recursive sequence and use it as the corrected posterior value. The corrected posterior value is combined with the device number, the window number corresponding to the trigger segment, and the alarm piece identifier to form a posterior piece. The posterior piece includes at least the device number, window number, corrected posterior value, and posterior recursive sequence index. The corrected posterior value is used to form the event node weight in subsequent S4, and the posterior recursive sequence index is used to backtrack the impact of the positive likelihood segment and the disproving deduction segment in each window number on the corrected posterior value. If all window numbers in the trigger segment are occupied by the disproving placeholder sequence, the corrected posterior value is retained according to the recursive result. If there is no positive likelihood segment in the recursive process, the corrected posterior value participates in the subsequent disproving conflict graph processing as a low-weight event node, instead of being directly used as the response conclusion. Based on the results of the above steps, S3 puts the boundary crossing strength of the trigger segment and the reverse constraint of the counter-evidence piece into the same time window sequence, so that the posterior piece is not directly generated by a single boundary crossing, but is obtained by the fault likelihood sequence and the counter-evidence placeholder sequence together. In this way, when S4 forms the event node weights, it can directly use the corrected posterior value and continue to combine the counter-evidence node, the counter-evidence closure edge and the mutual exclusion edge to resolve the conflict. In practical applications: The trigger segment of a certain device covers three time windows: the fifteenth, sixteenth, and seventeenth. All three are higher than the upper boundary of the baseline of the device's baseline band. The edge nodes calculate the out-of-bounds distance for each of the three time windows and form a fault likelihood sequence. The counter-evidence slice shows that there is a reverse change between adjacent measurement points in the fifteenth time window, the retransmission count is not zero in the sixteenth time window, and there is no counter-evidence occupancy in the seventeenth time window. Therefore, the out-of-bounds distances of the fifteenth and sixteenth time windows enter the counter-evidence deduction segment, and the out-of-bounds distance of the seventeenth time window enters the positive likelihood segment. The edge nodes complete the Bayesian odds recursion in the order of the fifteenth to the seventeenth time windows, and use the posterior value of the seventeenth time window as the corrected posterior value to generate the posterior slice.

[0020] S4. Form a node table with alarm pieces, counter-proof pieces and posterior pieces. Establish counter-proof closure edges based on the counter-proof values ​​in the counter-proof pieces. Establish mutually exclusive edges based on the event nodes and counter-proof nodes in the node table. Perform MaxWalkSAT flip search according to the number of collisions, closure difference and flip loss to generate resolution pieces. In this embodiment, S4 is used to transform the alarm piece, the counter-proof piece, and the posterior piece into a solvable graph structure, so that the station does not directly output the response result based on the corrected posterior value, but first puts the event node and the counter-proof node into the same counter-proof conflict graph for state resolution; the event node represents the fault event corresponding to the alarm piece, and the neighbor counter-proof node, the link counter-proof node, and the job counter-proof node represent the counter-proof effect of the reverse change of adjacent measurement points, communication retransmission, and job occupation on the fault event, respectively; the weight of the event node comes from the corrected posterior value in the posterior piece and is used to measure the loss caused by the flip event node in the flip search; Furthermore, in this embodiment, the specific processing is as follows: S4-1. The edge node reads the device number, time window number, and trigger segment from the alarm segment, reads the adjacent measurement point changes, transmission anomalies, and job occupancy from the counter-proof segment, and reads the corrected posterior value from the posterior segment. For the same device number and the same time window number, the edge node generates the trigger segment corresponding to the alarm segment as an event node, generates the adjacent measurement point changes in the counter-proof segment as a neighbor counter-proof node, generates the transmission anomaly as a link counter-proof node, and generates the job occupancy as a job counter-proof node, and configures the corrected posterior value as the event node weight. Both the event node and the counter-proof node are set to either reserved or flipped, with the initial state being reserved. Reserved indicates that the data conclusion corresponding to the node enters the current resolution process, while flipped indicates that the data conclusion corresponding to the node is excluded. If there are both an event node and at least one counter-proof node under the same device number and the same time window number, the edge node combines the event node, counter-proof node, and event node weight to form a node table. If there is no counter-proof node, the event node state is set to reserved, the counter-proof node state is set to empty, the mutual exclusion edge state is set to empty, and a resolution segment is directly formed. S4-2. The edge node reads the neighboring node, link node, and job node from the node table and calculates the corresponding counter-proof values: the neighboring node counter-proof value is the number of time windows covered by the reverse change in the neighboring node; the link node counter-proof value is the number of retransmissions within the corresponding time window number of the link node; and the job node counter-proof value is the overlap duration within the corresponding time window number of the job node. If the neighboring node counter-proof value, link node counter-proof value, and job node counter-proof value are all non-zero, it indicates that neighboring node counter-proof, link node counter-proof, and job node counter-proof exist simultaneously under the same device number and the same time window number. For proof by contradiction, edge nodes establish proof by contradiction closure edges between neighboring proof by contradiction nodes, link proof by contradiction nodes, and job proof by contradiction nodes. If zero values ​​exist among the three types of proof by contradiction values, edge nodes only retain proof by contradiction nodes with non-zero values ​​and do not establish proof by contradiction closure edges. Proof by contradiction closure edges are used to constrain multiple proof by contradiction nodes to maintain linkage during the same conflict resolution process, preventing one proof by contradiction node from being retained while other strongly related proof by contradiction nodes under the same time window number are arbitrarily flipped. When there are no proof by contradiction closure edges, the closure difference is zero, and the proof by contradiction nodes still participate in subsequent processing through mutual exclusion edges. S4-3. Edge nodes establish mutual exclusion edges between event nodes and neighboring counter-proof nodes, link counter-proof nodes, and job counter-proof nodes. Each mutual exclusion edge connects an event node and a counter-proof node, representing the conflict relationship between the event's validity and the corresponding counter-proof's validity. Edge nodes count the number of edges where both ends of a mutual exclusion edge are in a retained state, and this number is taken as the conflict count. Edge nodes also count the number of edges where the states of the counter-proof nodes at both ends of a counter-proof closure edge are inconsistent, and this number is taken as the closure difference. The closure difference is included when one end is retained and the other end is flipped, and not included when both ends are retained or both are flipped. When an event node is flipped, the event node weight of that event node is included in the flipping loss. No flipping loss is added when only the counter-proof node is flipped. If the conflict count is not zero, edge nodes generate a conflict cost sequence according to the arrangement of conflict count, closure difference, and flipping loss. If the conflict count is zero, it means that there is no mutual exclusion conflict in a retained state between the event node and the current counter-proof node, and edge nodes directly form a resolution piece from the event node state, counter-proof node state, and mutual exclusion edge state. S4-4. Edge nodes use the MaxWalkSAT conflict resolution algorithm (MARKSAT) to perform a flip search on the contradiction proof conflict graph. During the search, the nodes are flipped sequentially according to the order of nodes in the node table. Event nodes, neighboring contradiction proof nodes, link contradiction proof nodes, and job contradiction proof nodes are flipped in turn. After each candidate flip, the edge node recalculates the number of conflicts, closure difference, and flip loss, obtaining the current conflict cost sequence. The current conflict cost sequence is compared sequentially with the previous conflict cost sequence: first, the number of conflicts is compared; if the number of conflicts decreases, the flip is retained. If the number of conflicts is the same, the closure difference is compared; if the closure difference decreases, the flip is retained. If both the number of conflicts and the closure difference are the same, the flip loss is compared; if the flip loss decreases, the flip is retained. If none of the three decreases, the flip is withdrawn. A round of traversal is completed by checking all nodes in the node table for one candidate flip. If no flip is retained in this round of traversal, a resolution piece is generated based on the current event node state, contradiction proof node state, and mutual exclusion edge state. If a retained flip exists in this round of traversal, the next round of traversal is entered based on the updated node state. Based on the results of the above steps, S4 transforms the event judgment from a single posterior value judgment into a state search process with contradictory proof closure constraints and mutually exclusive edge constraints. Whether an event node is retained no longer depends solely on the alarm intensity, but also on the contradictory proof structure formed by changes in adjacent measurement points, transmission anomalies, and job occupation under the same time window number. The number of conflicts in the conflict cost sequence is used to prioritize eliminating direct conflicts between event nodes and contradictory proof nodes. The closure difference is used to maintain the linkage state of multiple contradictory proof nodes under the same time window number. The flip loss is used to avoid arbitrarily flipping event nodes when the cost has not decreased. In practical applications: A device forms event nodes under the same time window number. The posterior slice provides the corrected posterior value, and the counter-proof slice provides the reverse change, retransmission count, and overlap duration. Based on this, edge nodes form neighbor counter-proof nodes, link counter-proof nodes, and job counter-proof nodes. Since all three types of counter-proof values ​​are non-zero, counter-proof closure edges are established. If the event node and the three counter-proof nodes are initially reserved, all three mutually exclusive edges will cause conflicts. The flip search first attempts to flip the counter-proof node or the event node. Each time, only flips that make the number of conflicts, closure difference, or flip loss decrease in order are retained until no flips are retained in one round of traversal. Then, the resolution slice is output.

[0021] S5. Based on the event node status, counter-evidence node status, mutual exclusion edge status and event node weight in the resolution slice, divide it into direct response slice, supplementary verification slice, false alarm retention slice and ordinary recording slice, and encapsulate the division result, alarm slice, counter-evidence slice and necessary original fragments into edge data transmission packets and send them to the main station to generate emergency response results. In this embodiment, S5 is used to convert the resolution piece generated by S4 into a station edge data transmission object, so that the master station receives not only the alarm conclusion, but also the event node status, the counter-evidence node status, the mutual exclusion edge status and the necessary original fragments behind the conclusion; the event node status in the resolution piece is used to indicate whether the fault event is retained, the counter-evidence node status is used to indicate whether the counter-evidence information is retained, the mutual exclusion edge status is used to indicate whether the event node and the counter-evidence node are still in the state of simultaneous retention, and the event node weight is used to indicate the corrected posterior value passed from the posterior piece to the event node; Furthermore, in this embodiment, the specific processing is as follows: S5-1: The edge node reads the event node status, counter-evidence node status, mutex status, and event node weight in the decryption piece, and first determines whether the event node is in a retained state. If the event node is in a retained state, the edge node continues to read the node status at both ends of each mutex. If the nodes at both ends of each mutex are not retained at the same time, and the event node weight is not zero, it means that the event node corresponding to the alarm piece has not been overturned by the counter-evidence node, and the edge node generates a direct response piece. If the event node is in a retained state, but at least one mutex is retained at both ends, it means that the event node corresponding to the alarm piece still conflicts with at least one type of counter-evidence node in the counter-evidence piece, and the edge node generates a supplementary verification piece. When the mutex is empty, it is processed as if the nodes at both ends of the mutex are not retained at the same time, to accommodate the case in S4 where there is no counter-evidence node and a decryption piece is directly formed. S5-2. If the edge node does not generate a direct response slice or a supplementary verification slice, it continues to determine whether the event node is in a flipped state. If the event node is in a flipped state and at least one counter-evidence node is in a retained state, it means that the event node corresponding to the alarm slice has been excluded by the counter-evidence node in the counter-evidence slice, and the edge node generates a false alarm retention slice. If the event node is in a flipped state and all counter-evidence nodes are in a flipped state, or if the resolution slice does not fall into the direct response slice, supplementary verification slice, or false alarm retention slice, the edge node generates a normal recording slice. The edge node uses the direct response slice, supplementary verification slice, false alarm retention slice, and normal recording slice as the partitioning result, so that each resolution slice enters only one slice type. When the event node is retained, it is partitioned only between the direct response slice and the supplementary verification slice. When the event node is flipped, it is partitioned only between the false alarm retention slice and the normal recording slice. S5-3. Edge nodes extract necessary original fragments from alarm fragments and counter-evidence fragments according to the partitioning results. For alarm fragments, edge nodes read the original sampled values ​​within the time window number where the trigger segment is located according to the device number, time window number, and trigger segment identifier, and use this part as the original sampled value of the trigger segment. For counter-evidence fragments, edge nodes read the original sampled values ​​of adjacent measurement points in the time window number where the adjacent measurement point changes, the original record of the link log in the time window number where the transmission anomaly occurs, and the original record of the job ticket in the time window number where the job occupancy occurs, and use the above content as the original record. If a certain type of counter-evidence does not exist in the counter-evidence fragment, the original record of that type is empty, which does not affect the extraction of other original records. Edge nodes merge the original sampled values ​​and original records of the trigger segment into necessary original fragments to support the main station in reviewing the partitioning results. S5-4. Edge nodes use the same device number and the same time window number as the encapsulation index to encapsulate the partitioning results, alarm fragments, counter-evidence fragments, and necessary original fragments into edge data transmission packets. The edge data transmission packet includes at least the device number, time window number, partitioning result, alarm fragment identifier, counter-evidence fragment identifier, necessary original fragments, and resolution fragment index. The resolution fragment index is used to associate the event node status, counter-evidence node status, mutual exclusion edge status, and event node weight. After the edge node sends the edge data transmission packet to the master station, the master station generates emergency response results according to the partitioning results: direct response fragments enter emergency handling prompts, supplementary verification fragments enter verification prompts, false alarm retention fragments enter false alarm archives, and ordinary recording fragments enter ordinary archives. Based on the results of the above steps, S5 converts the digestion fragments into hierarchical transmission results, enabling the station edge nodes to complete the classification processing of event retention, counter-evidence retention, and mutual exclusion edge states before sending data; after the master station receives the edge data transmission packet, it does not need to pull the full monitoring stream again, and can obtain the correspondence between alarm fragments, counter-evidence fragments, and necessary original fragments. In practical applications: In the resolution slice of the fifteenth time window of a certain device, the event node is in a reserved state, the three counter-proof nodes are all in a flipped state, the nodes at both ends of the mutual exclusion edge are not reserved at the same time, and the weight of the event node is not zero. Then the edge node generates a direct response slice, and extracts the original sampled value of the trigger segment of the fifteenth time window and the corresponding original counter-proof record and encapsulates it into an edge data transmission packet. If the event node of another device is in a reserved state, but the link counter-proof node is also in a reserved state and the nodes at both ends of the mutual exclusion edge are reserved at the same time, then the edge node generates a supplementary verification slice and sends the original sampled value of the trigger segment and the original record of the link log to the main station.

[0022] Working principle: This implementation method completes monitoring data filtering, counter-evidence generation, post-hoc correction, conflict resolution, and edge data transmission within the station edge node. The station first acquires monitoring streams, link logs, and work ticket records, forms device windows according to device number and time window number, and excludes disturbed device windows affected by communication or work using retransmission count and work occupancy value. Then, a device baseline band is formed from the undisturbed device windows. Trigger segments are extracted from device windows that exceed the device baseline band and alarm pieces are generated. Subsequently, for the device number, time window number, and trigger segment in the alarm piece, the neighbor difference, retransmission count, and overlap duration are calculated respectively to obtain changes in adjacent measurement points, transmission anomalies, and work occupancy, and these are merged to generate counter-evidence pieces. Then, the out-of-bounds distances of the trigger segment relative to the device baseline are concatenated into a fault likelihood sequence. Reverse changes, transmission anomalies, and job occupancy are mapped to the same time window number to form a counter-evidence occupancy sequence. The corrected posterior value is obtained through Bayesian odds recursion, and a posterior fragment is generated. Based on this, event nodes are formed with alarm fragments, counter-evidence nodes are formed with counter-evidence fragments, and the corrected posterior value is used as the event node weight. A counter-evidence conflict graph is constructed through counter-evidence closure edges and mutual exclusion edges. MaxWalkSAT flip search is performed according to the number of conflicts, closure difference, and flip loss to generate a resolution fragment. Finally, according to the event node status, counter-evidence node status, mutual exclusion edge status, and event node weight in the resolution fragment, the station-end event is divided into direct response fragments, supplementary verification fragments, false alarm retention fragments, and ordinary recording fragments. The division results, alarm fragments, counter-evidence fragments, and necessary original fragments are encapsulated into edge data transmission packets and sent to the main station to generate emergency response results. In practical applications, if a temperature rise occurs at a main transformer temperature measurement point within a time window, the station edge node first determines whether there is a link retransmission or work ticket coverage within that time window. If there is a retransmission or work occupation, the time window will not enter the equipment baseline band to avoid treating communication or maintenance disturbances as equipment status benchmarks. For trigger segments outside the equipment baseline band in unaffected time windows, the edge node continues to read the sampled values ​​of adjacent measurement points under the same time window number, forming a neighbor difference and determining whether adjacent measurement points change in the same direction. At the same time, it reads the number of retransmissions in the link log and the overlap duration in the work ticket record to generate a counter-evidence. Subsequently, the system forms a fault likelihood sequence based on the time window number for the out-of-bounds distance of the trigger segment, maps reverse changes, transmission anomalies, and job occupancy to a counter-evidence placeholder sequence, weakens the out-of-bounds judgment affected by counter-evidence placeholders through Bayesian odds recursion, and then puts the posterior result into the counter-evidence conflict graph for flip search; if the event node is retained and the nodes at both ends of the mutually exclusive edge are not retained at the same time, the edge node generates a direct response slice; if the event node still conflicts with the counter-evidence node, a supplementary verification slice is generated; if the event node is overturned by the counter-evidence node, a false alarm retention slice is generated; thus, the main station receives not a single temperature out-of-bounds data, but an edge data transmission packet containing the trigger segment, counter-evidence slice, resolution result, and necessary original fragments, which can be directly used for remote emergency judgment.

[0023] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for unmanned operation of power data monitoring and emergency response in substations, characterized in that, include: S1. At the station, obtain monitoring stream, link log and work ticket record, form device windows according to device number and time window number, and generate device baseline band after removing disturbed device windows by retransmission count and work occupancy value. Then, extract trigger segments from device windows that exceed the device baseline band and generate alarm pieces. S2. For the device number, time window number and trigger segment of the alarm video, generate changes in adjacent measurement points by adjacent point difference, generate transmission anomalies by retransmission count, generate job occupancy by overlap duration, and merge them to generate counter-evidence video. S3. Generate a fault likelihood sequence based on the trigger segment and the equipment baseline band, generate a counter-evidence placeholder sequence based on changes in adjacent measurement points, transmission anomalies and job occupancy, and perform Bayesian odds recursion on the fault likelihood sequence and the counter-evidence placeholder sequence to generate a posterior piece. S4. Form a node table with alarm pieces, counter-proof pieces and posterior pieces. Establish counter-proof closure edges based on the counter-proof values ​​in the counter-proof pieces. Establish mutually exclusive edges based on the event nodes and counter-proof nodes in the node table. Perform MaxWalkSAT flip search according to the number of collisions, closure difference and flip loss to generate resolution pieces. S5. Based on the event node status, counter-evidence node status, mutual exclusion edge status, and event node weight in the digestion slice, divide it into direct response slice, supplementary verification slice, false alarm retention slice, and ordinary recording slice. Then, encapsulate the division result, alarm slice, counter-evidence slice, and necessary original fragments into an edge data transmission packet and send it to the main station to generate emergency response results.

2. The method for unmanned substation power data monitoring and emergency response according to claim 1, characterized in that: S1 includes: S1-1. In the station-end monitoring stream, the sampled values ​​of the same device are merged by device number, and the sampled values ​​under the same device number are divided by time window number to form device window segments. S1-2. In the link log, count the number of retransmitted packets using the device number and time window number. In the job ticket record, count the overlap between the job ticket effective time and the device window segment using the device number and time window number. Use the number of retransmitted packets and the overlap duration as the retransmission count and job occupation value, respectively, to generate the window segment occupation. S1-3. Form a device baseline band using device windows with zero retransmission count and zero job occupancy value under the same device number, and use device windows with zero retransmission count, zero job occupancy value and exceeding the device baseline band as trigger segments to generate alarm clips containing device number, time window number and trigger segment.

3. The method for unmanned substation power data monitoring and emergency response according to claim 2, characterized in that: S2 includes: S2-1. Read the device number, time window number and trigger segment from the alarm chip. Take the sampled value of the trigger segment and the sampled value of the adjacent measuring point under the same time window number in the monitoring stream. Subtract the sampled value of the trigger segment from the sampled value of the adjacent measuring point to form the adjacent point difference. S2-2. Take the direction of the trigger segment's crossover relative to the equipment baseline as the trigger direction, and compare the positive and negative directions of the adjacent point difference with the trigger direction. When the positive and negative directions of the adjacent point difference are consistent with the trigger direction, a change in the same direction is formed. When the positive and negative directions of the adjacent point difference are opposite to the trigger direction, a change in the opposite direction is formed. Changes in adjacent measuring points are generated by changes in the same direction or in the opposite direction.

4. The method for unmanned substation power data monitoring and emergency response according to claim 3, characterized in that: S2 further includes: S2-3. Count the number of retransmissions within the same time window in the link log along the device number and time window number, and treat the number of retransmissions as a transmission anomaly. Count the overlap between the job ticket effective time and the trigger segment in the job ticket record along the device number and time window number, and treat the overlap duration as job occupancy. S2-4. According to the same device number and the same time window number, merge the changes of adjacent measurement points, transmission anomalies and job occupancy with the trigger segments in the alarm video to generate a counter-evidence video.

5. The method for unmanned substation power data monitoring and emergency response according to claim 4, characterized in that: S3 includes: S3-1. Read the trigger segment and device baseline band in the alarm chip, calculate the cross-boundary distance of the sampled value of the trigger segment relative to the device baseline band according to each time window number in the trigger segment, and concatenate the cross-boundary distances in the order of the time window numbers to generate a fault likelihood sequence. S3-2. Read the changes, transmission anomalies, and job occupancy of adjacent measurement points in the counter-evidence slice. Map the time window number where the reverse change is located, the time window number where the transmission anomaly is located, and the time window number covered by the job occupancy to the same time window number of the fault likelihood sequence, and generate the counter-evidence occupancy sequence.

6. The method for unmanned substation power data monitoring and emergency response according to claim 5, characterized in that: S3 further includes: S3-3. Take the out-of-bounds distance in the fault likelihood sequence that is not occupied by the counter-evidence placeholder sequence as the positive likelihood segment, and take the out-of-bounds distance in the fault likelihood sequence that is occupied by the counter-evidence placeholder sequence as the counter-evidence deduction segment. Perform Bayesian odds recursion on the positive likelihood segment and the counter-evidence deduction segment in the order of time window number to generate the posterior recursion sequence. S3-4. Take the posterior value of the last window number in the posterior recursive sequence as the corrected posterior value, and generate a posterior chip with the device number, window number and corrected posterior value.

7. The method for unmanned substation power data monitoring and emergency response according to claim 6, characterized in that: S4 includes: S4-1. Read alarm chips, counter-proof chips, and post-proof chips. Form event nodes with the same device number and the same time window number. Form neighbor counter-proof nodes, link counter-proof nodes, and job counter-proof nodes with changes in adjacent measurement points, transmission anomalies, and job occupancy in the counter-proof chips, respectively. Use the corrected post-proof value as the weight of the event nodes. When there are event nodes and at least one counter-proof node under the same device number and the same time window number, form a node table. Otherwise, directly form the event node into a dissolution chip. S4-2. The neighboring point counter-proof value is formed according to the number of time windows of reverse change in the neighboring counter-proof node, the link counter-proof value is formed according to the number of retransmissions in the link counter-proof node, and the job counter-proof value is formed according to the overlap duration in the job counter-proof node. When the neighboring point counter-proof value, the link counter-proof value, and the job counter-proof value are all non-zero, a counter-proof closure edge is established between the three counter-proof nodes. Otherwise, only the counter-proof nodes with non-zero values ​​are retained and the establishment of counter-proof closure edges is prohibited.

8. The method for unmanned substation power data monitoring and emergency response according to claim 7, characterized in that: S4 further includes: S4-3. Establish mutual exclusion edges between the event node and the neighboring counter-proof node, the link counter-proof node, and the job counter-proof node, respectively. The number of nodes at both ends of the mutual exclusion edge that are simultaneously retained is used to form the conflict number. The number of nodes at both ends of the counter-proof closure edge that are inconsistent in state is used to form the closure difference. The event node weight of the flipped event node is used to form the flip loss. When the conflict number is not zero, generate a conflict cost sequence in the order of conflict number, closure difference, and flip loss. Otherwise, directly form a resolution piece by the event node state, the counter-proof node state, and the mutual exclusion edge state. S4-4. The MaxWalkSAT conflict resolution algorithm for Markov logic networks is adopted. The state flipping is performed on each event node, neighbor node, link node, and job node, and the conflict cost sequence is recalculated after each flipping. If the current conflict cost sequence is less than the previous conflict cost sequence in the order of conflict number, closure difference, and flipping loss, the current flipping is retained; otherwise, the current flipping is withdrawn. When no flipping is retained in a continuous round of traversal, a resolution piece containing the state of the event node, the state of the node, and the state of the mutually exclusive edge is generated; otherwise, the next round of traversal is entered.

9. The method for unmanned substation power data monitoring and emergency response according to claim 8, characterized in that: S5 includes: S5-1. Read the event node status, proof node status, mutual exclusion edge status and event node weight in the unsolvable piece. When the event node is retained, the nodes at both ends of each mutual exclusion edge are not retained at the same time and the event node weight is not zero, a direct response piece is generated. Otherwise, when the event node is retained and at least one mutual exclusion edge has both ends of the node retained at the same time, a supplementary verification piece is generated. S5-2. When an event node is flipped and at least one counter-evidence node is retained, a false alarm retention slice is generated. Otherwise, the resolution slice that has not generated a direct response slice, a supplementary verification slice, and a false alarm retention slice is generated into a normal record slice, and the partitioning result is generated using the direct response slice, the supplementary verification slice, the false alarm retention slice, and the normal record slice.

10. The method for unmanned substation power data monitoring and emergency response according to claim 9, characterized in that: The S5 also includes: S5-3. According to the division results, extract the original sampled value of the trigger segment from the alarm segment, extract the original record of the time window number where the adjacent measurement point changes, transmission anomalies and operation occupancy are located from the counter-evidence segment, and merge the original sampled value of the trigger segment and the original record into the necessary original segment. S5-4. Encapsulate the partitioning results, alarm fragments, counter-evidence fragments, and necessary original fragments into an edge data transmission packet with the same device number and the same time window number, and send the edge data transmission packet to the main station to generate the emergency response result.