Transformer operation state real-time analysis method and system oriented to edge computing

Abstract

This invention discloses a method and system for real-time analysis of transformer operating status oriented towards edge computing, specifically relating to the field of power equipment condition monitoring and edge computing data processing technology. The method includes acquiring load values, oil temperature values, winding temperature values, cooling status values, and status monitoring values ​​collected from the transformer site. It generates three types of change symbols (increase, decrease, and remain unchanged) for each type of value according to the sampling order, and binds each change symbol to its corresponding sampling sequence number, outputting ordered transformer operating data. By converting the load values, oil temperature values, winding temperature values, cooling status values, and status monitoring values ​​collected from the transformer site into change symbols with sampling sequence numbers, it identifies load change segments or cooling change points and extracts the current temperature rise response segment. Then, it compares the current temperature rise response segment with existing temperature rise response segments in response order, and generates transformer operating status analysis results based on the order differences formed by oil temperature reversal, winding temperature changes, and status monitoring value changes.

Description

Technical Field

[0001] This invention relates to the field of power equipment condition monitoring and edge computing data processing technology, and more specifically, to a method and system for real-time analysis of transformer operating status oriented towards edge computing. Background Technology

[0002] In the field of real-time analysis of transformer operating status, the mainstream practice in the industry is to use monitoring data such as oil temperature, winding temperature, load current, cooler status, partial discharge, vibration and ambient temperature and humidity to make online judgments on whether the transformer is overheating, insulation deterioration, mechanical loosening or abnormal discharge trend. Specifically, fixed threshold discrimination, historical trend comparison, cloud model diagnosis or deployment of lightweight models to edge computing nodes are often used for preliminary analysis. In practical application scenarios such as distribution substations, unattended substations, or new energy booster stations, edge computing nodes are usually deployed at the station or transformer side. Due to limitations in local computing power, storage capacity, communication bandwidth, and power grid security strategies, it is difficult to store all high-frequency data for a long time, and it is also difficult to continuously rely on the cloud for real-time verification. At the same time, during transformer operation, real operational disturbances such as load jumps, cooler start-up and shutdown, sudden changes in ambient temperature, and short-term overloads will frequently occur. Under this constraint, the mainstream approach will consistently expose a key flaw: it often weakens the above disturbances as noise, interference terms, or normal operating condition corrections. As a result, the edge side can only observe whether the current value exceeds the limit or whether the trend is abnormal, but cannot use verifiable phenomena such as the temperature rise rate before and after the disturbance, the cooling drop amplitude, the vibration recovery process, and the duration of partial discharge fluctuations to determine whether the response capability of the same transformer has degraded. Thus, early weak anomalies are masked by normal load fluctuations before the limit is exceeded. The technical problem this application aims to solve is: how to perform real-time, low-storage, and verifiable analysis of the transformer's operating status by utilizing the short-time response process triggered by actual operating disturbances of the transformer under the constraints of edge computing. Summary of the Invention

[0003] To overcome the aforementioned deficiencies of the prior art, embodiments of the present invention provide a real-time analysis method and system for transformer operating status oriented towards edge computing. This method converts load values, oil temperature values, winding temperature values, cooling status values, and status monitoring values ​​collected on-site from the transformer into change symbols with sampling sequence numbers. It identifies load change segments or cooling change points and extracts the current temperature rise response segment. The current temperature rise response segment is then compared with existing temperature rise response segments in response order. Based on the sequence differences formed by oil temperature reversal, winding temperature changes, and status monitoring value changes, the transformer operating status analysis result is generated, thus solving the problems mentioned in the background art.

[0004] To achieve the above objectives, the present invention provides the following technical solution: a real-time analysis method for transformer operating status oriented towards edge computing, characterized in that it includes: S1. Acquire the load value, oil temperature value, winding temperature value, cooling status value and status monitoring value collected on the transformer site, generate three types of change symbols (increase, decrease and remain unchanged) for each type of value according to the sampling order, bind each change symbol with the corresponding sampling sequence number, and output the orderly operation data of the transformer. S2. Using the orderly operation data of the transformer as input, merge the consecutive identical increase or decrease symbols in the load value into a load change segment. Record the sampling sequence number of the cooling status value from shutdown to start or from start to shutdown as the cooling change point. Use the load change segment or cooling change point as the transformer operation disturbance. Extract the data between the most recent load symbol reversal point before the disturbance and the first reversal point of the oil temperature value after the disturbance, and output the current temperature rise response segment. S3. Read the disturbance source, disturbance direction, pre-disturbance cooling state, pre-disturbance load sequence, first change sequence of oil temperature, first change sequence of winding temperature, and first change sequence of status monitoring value in the current temperature rise response segment. Select segments with the same disturbance source, the same disturbance direction, and the same pre-disturbance cooling state from the stored temperature rise response segments, and arrange them from near to far according to the pre-disturbance load sequence. Output the reference temperature rise response segment sequence. S4. Compare the current temperature rise response segment with the reference temperature rise response segment sequence segment by segment, and generate the current response sequence string and the reference response sequence string according to the sampling order. The response sequence string consists of the order of occurrence of load change, cooling change, oil temperature change, winding temperature change, status monitoring value change and oil temperature reversal. Read the position of the same event in the two segments one by one and output the sequence difference item of the current temperature rise response segment. S5. Generate transformer operating status analysis results based on sequence difference items, and write the current temperature rise response segment, the current response sequence string, and the recovery sampling position after oil temperature reversal into the stored temperature rise response segment, and output the updated stored temperature rise response segment.

[0005] In a preferred embodiment, S1 includes: S1-1: Taking the nth sample value and the (n-1)th sample value in the same acquisition channel as input, compare the size relationship between the nth sample value and the (n-1)th sample value. If the nth sample value is greater than the (n-1)th sample value, generate an increment sign; if the nth sample value is less than the (n-1)th sample value, generate a decrement sign; if the nth sample value is equal to the (n-1)th sample value, generate a neutral sign. Output the channel change sign with the sampling sequence number n. S1-2. Using the channel change symbols of each acquisition channel as input, merge consecutive identical channel change symbols within the same acquisition channel into a symbol segment, and record the first sampling number, the last sampling number, the first sample value, and the last sample value of the symbol segment, and output the channel symbol segment; S1-3. Using the channel symbol segments of each acquisition channel as input, arrange them in order of first sampling sequence number, and bind each channel symbol segment with the corresponding acquisition channel name, symbol type, first sampling sequence number, last sampling sequence number, first sampling value and last sampling value, and output the orderly operation data of the transformer.

[0006] In a preferred embodiment, S2 includes: S2-1. Using the load value channel symbol segment in the orderly operation data of the transformer as input, the load value channel symbol segment with the symbol type of increase symbol or decrease symbol is recorded as the load change segment, and the symbol type of the load change segment is recorded as the disturbance direction, the starting sampling number is recorded as the disturbance start point, and the ending sampling number is recorded as the disturbance end point. Output the load disturbance record. S2-2. Using the cooling status value channel symbol segment in the orderly operation data of the transformer as input, read the sampling sequence number corresponding to the cooling status value from shutdown to start or from start to shutdown, record the sampling sequence number corresponding to the cooling status value from shutdown to start or from start to stop as the cooling change point, and record the cooling change point as the disturbance start point and disturbance end point, and output the cooling disturbance record. S2-3. Using load disturbance records or cooling disturbance records as input, before the disturbance start point, read the sampling number of the most recent load value symbol type that changes from increasing to decreasing or from decreasing to increasing as the segment start point, and after the disturbance end point, read the sampling number of the oil temperature value symbol type that changes from increasing to decreasing or from decreasing to increasing as the segment end point. Extract the orderly operation data of the transformer between the segment start point and the segment end point, and output the current temperature rise response segment.

[0007] In a preferred embodiment, S3 includes: S3-1. Taking the disturbance source, disturbance direction, pre-disturbance cooling state, pre-disturbance load sequence, first change sequence of oil temperature, first change sequence of winding temperature, and first change sequence of status monitoring value in the current temperature rise response segment as input, retain segments with the same disturbance source, the same disturbance direction, and the same pre-disturbance cooling state from the existing temperature rise response segments. Then, assemble the pre-disturbance load sequence, first change sequence of oil temperature, first change sequence of winding temperature, and first change sequence of status monitoring value of each retained segment into a segment sequence vector according to the same field order, and output the candidate segment vector set.

[0008] In a preferred embodiment, S3 further includes: S3-2. Taking the candidate segment vector set and the current sequence vector of the current temperature rise response segment as input, the candidate segment vector set is arranged into a candidate matrix by column. The candidate matrix is ​​orthogonally decomposed to obtain an orthogonal matrix and an upper triangular matrix. The columns with diagonal elements of zero and their corresponding candidate segments in the upper triangular matrix are deleted to obtain a non-zero column candidate matrix. The inverse matrix of the transpose product matrix of the non-zero column candidate matrix is ​​calculated. The redundant candidate segment set and the inverse matrix are output. S3-3. Taking the set of candidate segments for redundancy removal, the candidate matrix with non-zero columns, the inverse matrix, and the current order vector as input, calculate the reconstruction coefficient of each candidate segment for redundancy removal to the current order vector through the inverse matrix. Remove each candidate segment for redundancy removal one by one and calculate the change in reconstruction residual before and after removal. Arrange the candidate segments for redundancy removal in order of increasing load order difference before disturbance and decreasing reconstruction residual change, and output the reference temperature rise response segment sequence.

[0009] In a preferred embodiment, S4 includes: S4-1. Using the current temperature rise response segment and the reference temperature rise response segment sequence as input, read the first sampling sequence number of load change, cooling change, oil temperature change, winding temperature change, condition monitoring value change and oil temperature reversal respectively. Arrange each event in ascending order of the first sampling sequence number to generate the current response sequence string and the reference response sequence string. Convert each response sequence string into a sequence vector composed of event position, adjacent event interval and event missing marker. Output the current sequence vector and the reference sequence vector set.

[0010] In a preferred embodiment, S4 further includes: S4-2. Using the reference order vector set as input, construct a third-order order tensor according to event type, fragment number and order field. Perform tensor decomposition on the third-order order tensor to obtain the event factor matrix, fragment factor matrix and field factor matrix. Reconstruct the reference order basis using the event factor matrix and field factor matrix, and output the reference order basis matrix after removing duplicate order relations. S4-3. Using the current order vector and the reference order basis matrix as input, construct a regularized linear equation with the goal of minimizing the sum of the linear reconstruction error, the sum of squared reconstruction coefficients, and the event missing penalty term between the current order vector and the reference order basis matrix. Solve to obtain the reconstruction coefficients of the current order vector on the reference order basis matrix. Subtract the reconstructed order vector from the current order vector event by event, and output the order difference term including event position shift, event position shift, change of adjacent event interval, and event missing.

[0011] In a preferred embodiment, S5 includes: S5-1. Taking the sequential difference item as input, read the order shift items, order shift items, and missing event items of oil temperature change, winding temperature change, status monitoring value change, and oil temperature reverse relative to load change or cooling change, and generate a status difference table according to event name, difference type and corresponding sampling number, and output the transformer status difference record. S5-2. Using the transformer condition difference record as input, read the missing reverse oil temperature item, the reverse oil temperature position shifted item, the winding temperature change earlier than the oil temperature change item, and the condition monitoring value change earlier than the oil temperature change item; correspond the missing reverse oil temperature item and the reverse oil temperature position shifted item as temperature rise recovery anomaly, correspond the winding temperature change earlier than the oil temperature change item as winding thermal response advance, correspond the condition monitoring value change earlier than the oil temperature change item as condition monitoring pre-anomaly, and correspond the condition difference record without a corresponding relationship as disturbance response consistency, and output the transformer operating condition analysis result.

[0012] In a preferred embodiment, S5 further includes: S5-3. Using the current temperature rise response segment, the current response sequence string, the recovery sampling position after the oil temperature reverses, and the transformer operating status analysis result as inputs, bind the four according to the same segment number and write them into the stored temperature rise response segment. Then, use the written segment number, the recovery sampling position, and the operating status analysis result as index fields to output the updated stored temperature rise response segment.

[0013] In a preferred embodiment, the real-time transformer operation status analysis system for edge computing includes a symbol encoding module, a disturbance truncation module, a reference sorting module, a sequence comparison module, and a result write-back module. The symbol encoding module is used to acquire load values, oil temperature values, winding temperature values, cooling status values ​​and status monitoring values ​​collected on-site by the transformer. It generates three types of change symbols (increase, decrease, and remain unchanged) for each type of value according to the sampling order, and binds each change symbol with the corresponding sampling sequence number to output the orderly operation data of the transformer. The disturbance truncation module is used to take the orderly operation data of the transformer as input, merge the consecutive identical increase or decrease symbols in the load value into a load change segment, record the sampling sequence number of the cooling status value from shutdown to start or from start to shutdown as the cooling change point, and take the load change segment or cooling change point as the transformer operation disturbance, extract the data between the most recent load symbol reversal point before the disturbance and the first reversal point of the oil temperature value after the disturbance, and output the current temperature rise response segment. The reference sorting module is used to read the disturbance source, disturbance direction, pre-disturbance cooling state, pre-disturbance load sequence, first change sequence of oil temperature, first change sequence of winding temperature, and first change sequence of status monitoring value in the current temperature rise response segment. It selects segments with the same disturbance source, the same disturbance direction, and the same pre-disturbance cooling state from the stored temperature rise response segments, and arranges them from near to far according to the pre-disturbance load sequence, and outputs the reference temperature rise response segment sequence. The sequence comparison module is used to compare the current temperature rise response segment with the reference temperature rise response segment sequence segment by segment. It generates the current response sequence string and the reference response sequence string according to the sampling order. The response sequence string consists of the order of occurrence of load change, cooling change, oil temperature change, winding temperature change, status monitoring value change and oil temperature reversal. It reads the position of the same event in the two bit by bit and outputs the sequence difference item of the current temperature rise response segment. The result write-back module generates transformer operating status analysis results based on the sequence difference items, and writes the current temperature rise response segment, the current response sequence string, and the recovery sampling position after the oil temperature reverses into the stored temperature rise response segment, and outputs the updated stored temperature rise response segment.

[0014] The technical effects and advantages of this invention are as follows: By using load change segments or cooling change points as trigger conditions for temperature rise response analysis, and extracting the current temperature rise response segments before and after the disturbance at the edge computing node, the actual operating disturbance is transformed from an interference item into an analysis basis, thereby relatively improving the problem of early weak anomalies being masked by normal load fluctuations under edge computing constraints. By converting the values ​​of each acquisition channel into increasing, decreasing, and flat change symbols at the edge computing node and merging them into channel symbol segments, the analysis object on the edge side is transformed from the full amount of raw data into records of change direction and sampling sequence number, thereby relatively reducing the local storage volume and the burden of continuous high-frequency data processing. By selecting segments with the same disturbance source, disturbance direction, and pre-disturbance cooling state from the existing temperature rise response segments, and sorting them according to load sequence and reconstruction contribution, the reference object is made closer to the current operating conditions, thereby relatively reducing the comparison deviation caused by the mixing of segments from different operating conditions. By converting the current temperature rise response segment and the reference temperature rise response segment into a response sequence string, and comparing the order of occurrence of load, cooling, oil temperature, winding temperature, condition monitoring value and oil temperature reversal, the condition judgment is changed from numerical limit exceeding to process comparison, thereby improving the interpretability of edge side analysis. By mapping the oil temperature reverse missing, the oil temperature reverse position shifted backward, the winding temperature change advanced, and the status monitoring value change advanced into operating status results respectively, the sequential difference items can be transformed into clear status conclusions, thereby relatively alleviating the problem of general alarms being difficult to locate the cause. By writing the current temperature rise response segment, the current response sequence string, the restored sampling position, and the analysis results into the stored temperature rise response segment according to the segment number, the analysis results of this analysis are entered into the subsequent reference sorting process, thereby forming a continuous update mechanism for historical segments on the edge side. Attached Figure Description

[0015] Figure 1 This is a flowchart of the present invention.

[0016] Figure 2 This is a schematic diagram of the system modules of the present invention. Detailed Implementation

[0017] 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.

[0018] Refer to the instruction manual appendix Figure 1-2 The present invention provides a real-time analysis method for transformer operating status based on edge computing, comprising: S1. Acquire the load value, oil temperature value, winding temperature value, cooling status value and status monitoring value collected on the transformer site, generate three types of change symbols (increase, decrease and remain unchanged) for each type of value according to the sampling order, bind each change symbol with the corresponding sampling sequence number, and output the orderly operation data of the transformer. In this embodiment, S1 is used to convert the load value, oil temperature value, winding temperature value, cooling status value, and status monitoring value continuously collected on-site from the transformer into ordered transformer operation data that can be directly called by subsequent disturbance segmentation, reference sorting, and sequential comparison. Since the physical meaning, sampling frequency, and numerical units of each acquisition channel are different, directly using the original values ​​for unified comparison can easily cause problems with cross-channel data connection. Therefore, this embodiment first generates three types of change symbols (increase, decrease, and flat) within each acquisition channel according to the sampling order, then merges consecutive identical change symbols into symbol segments, and finally arranges the symbol segments of different acquisition channels uniformly according to the sampling sequence number, so that subsequent steps can identify the transformer operation disturbance and temperature rise response process based on the change direction and occurrence sequence. This implementation process includes the following steps: In S1-1, the nth and (n-1)th sampled values ​​within the same acquisition channel are used as inputs, where n is the sampling sequence number within the same acquisition channel, and n starts taking values ​​from the second sampled value; the first sampled value is only used as the first comparison reference. For the load value channel, oil temperature value channel, winding temperature value channel, cooling status value channel, and status monitoring value channel, adjacent sampled values ​​are compared within their respective channels; sampled values ​​from different channels are not compared with each other. Specifically, the nth and (n-1)th sampled values ​​are read, and their magnitude relationship is calculated. When the nth sampled value is greater than the (n-1)th sampled value, it indicates that the channel is showing an increasing change relative to the previous sampling point at sampling sequence number n, generating an increment sign. When the nth sampled value is less than the (n-1)th sampled value, it indicates that the channel is showing a decreasing change relative to the previous sampling point at sampling sequence number n, generating a decrement sign. When the nth sampled value is equal to the (n-1)th sampled value, it indicates that the channel is showing a decreasing change relative to the previous sampling point at sampling sequence number n, generating a decrement sign. When there are n-1 sampled values, it indicates that the channel remains unchanged relative to the previous sampled point at sampled number n, generating a flat symbol. For the cooling status value channel, the shutdown state can be recorded as one discrete value, and the startup state as another discrete value, so that the cooling status value can also form an increase, decrease, or flat symbol by comparing adjacent sampled values. For the status monitoring value channel, if the status monitoring value is one of the partial discharge value, vibration value, noise value, or oil gas characteristic value, then adjacent comparisons are performed according to the digital quantity output by the corresponding acquisition device. After the comparison is completed, the generated increase, decrease, or flat symbol is bound to the sampled number n, and the corresponding acquisition channel name and the nth sampled value are retained, thereby outputting the channel change symbol with the sampled number n. Through this processing, each acquisition channel forms a sequence of change symbols arranged in order of sampling, and subsequent steps can directly read the change direction of a certain channel at any sampled number. In S1-2, using the channel change symbols of each acquisition channel as input, consecutive identical channel change symbols are merged within each acquisition channel. The merging is limited to channel change symbols with the same symbol type and consecutive sampling numbers within the same acquisition channel; merging across channels or identical symbols separated by other symbols is not performed. Specifically, the process starts reading from the first channel change symbol in the same acquisition channel, using its symbol type as the symbol type of the current symbol segment and its sampling number as the first sampling number of the current symbol segment. The process continues reading the channel change symbol with the next sampling number. If the next channel change symbol has the same symbol type as the current symbol segment, it is merged into the current symbol segment, and its sampling number is updated to the last sampling number of the current symbol segment. If the next channel change symbol has a different symbol type than the current symbol segment, the process ends. The system first generates an increasing symbol segment, and then starts a new symbol segment with the next channel change symbol. Each symbol segment records the first sampling number, the last sampling number, the first sample value, and the last sample value, where the first sample value is the sample value corresponding to the first sampling number, and the last sample value is the sample value corresponding to the last sampling number. For example, if the load value channel generates increasing symbols continuously from sampling number 2 to 5, then sampling numbers 2 to 5 are merged into one load value increasing symbol segment, and the load value at sampling number 2 and the load value at sampling number 5 are recorded. If a flat symbol is generated at sampling number 6, then the load value increasing symbol segment ends at sampling number 5, and a new load value flat symbol segment is started at sampling number 6. Through this processing, the continuous change process with the same direction is compressed into a symbol segment, which not only retains the change direction, start and end positions, and start and end values, but also reduces the number of data entries that the edge computing nodes need to process when identifying load change segments and temperature rise response processes in the future. In S1-3, using the channel symbol segments of each acquisition channel as input, the load value channel symbol segments, oil temperature value channel symbol segments, winding temperature value channel symbol segments, cooling status value channel symbol segments, and status monitoring value channel symbol segments are aggregated and arranged in ascending order according to the first sampling sequence number of each channel symbol segment. When channel symbol segments of different acquisition channels have the same first sampling sequence number, they are arranged in a fixed order: load value channel, cooling status value channel, oil temperature value channel, winding temperature value channel, and status monitoring value channel, to avoid uncertainty in the order of events across channels with the same sampling sequence number. After arrangement, each channel symbol segment is bound to the corresponding acquisition channel name, symbol type, first sampling sequence number, last sampling sequence number, first sampling value, and last sampling value to form a sequence that can be... The subsequent steps read ordered records; among them, the acquisition channel name is used to distinguish whether the record comes from load value, oil temperature value, winding temperature value, cooling status value or status monitoring value; the symbol type is used to indicate whether the record corresponds to an increasing symbol, decreasing symbol or flat symbol; the first sampling sequence number and the last sampling sequence number are used to determine the start and end positions of the symbol segment in time sequence; the first sampling value and the last sampling value are used to retain the starting point and ending point of the numerical change of the symbol segment; all ordered records constitute the transformer ordered operation data; through this data structure, the subsequent S2 can directly read continuous increasing symbol segments or continuous decreasing symbol segments from the load value channel symbol segment and determine the load change segment, and can also read the cooling status change position from the cooling status value channel symbol segment, thereby avoiding retracing the entire original sampling data; Through the above implementation method, S1 unifies the transformer field data of different types, units, and sampling sources into orderly transformer operation data with sampling sequence number as the main line, change symbol as the core, and symbol segment as the recording unit. This enables subsequent steps to directly identify the order of occurrence of load changes, cooling changes, temperature changes, and status monitoring value changes within the edge computing node, reducing the dependence on continuous storage of all original data and complex numerical models, while retaining the change direction, start and end positions, and channel sources required for transformer operation status analysis. In practical applications: for example, the edge computing node of a distribution transformer receives load values, oil temperature values, winding temperature values, cooling status values, and vibration values ​​at fixed sampling cycles. If the load value increases continuously from sampling number 2 to 6, S1-1 generates a continuous increase symbol. S1-2 merges these symbols into a single load value increase symbol segment. S1-3 arranges this load value increase symbol segment along with the cooling state value symbol segment, oil temperature value symbol segment, winding temperature value symbol segment, and vibration value symbol segment from the same period according to the sampling number. When the oil temperature value starts to increase at sampling number 8, the winding temperature value starts to increase at sampling number 9, and the vibration value changes at sampling number 10, these changes are all entered into the transformer's orderly operation data in the form of channel symbol segments. This allows subsequent steps to use the load value increase symbol segment as a source of disturbance, continue to extract the current temperature rise response segment, and analyze the transformer's response process under this actual operating disturbance.

[0019] S2. Using the orderly operation data of the transformer as input, merge the consecutive identical increase or decrease symbols in the load value into a load change segment. Record the sampling sequence number of the cooling status value from shutdown to start or from start to shutdown as the cooling change point. Use the load change segment or cooling change point as the transformer operation disturbance. Extract the data between the most recent load symbol reversal point before the disturbance and the first reversal point of the oil temperature value after the disturbance, and output the current temperature rise response segment. In this embodiment, S2 is used to identify transformer operating disturbances that can trigger temperature rise response analysis in the orderly transformer operation data output by S1, and to extract the current temperature rise response segment according to the order of changes before and after the disturbance. Since the changes in transformer oil temperature and winding temperature usually lag behind the changes in load or cooling status, if data is directly extracted according to a fixed time window, it is easy to extract data that is irrelevant to the current operating disturbance. Therefore, this embodiment does not use a preset time window, but first identifies the load change segment from the load value channel symbol segment, identifies the cooling change point from the cooling status value channel symbol segment, and then uses the load change segment or cooling change point as the disturbance trigger object, searches forward to find the most recent reverse position of the load symbol as the segment start point, and searches backward to find the first reverse position of the oil temperature value as the segment end point, so that the extracted current temperature rise response segment can cover the operating status before the disturbance, the disturbance process, and the oil temperature response transition process. This implementation process includes the following steps: In S2-1, the load value channel symbol segment in the transformer's ordered operation data is used as input. Ordered records with the channel name "Load Value Channel" are read one by one, and the symbol type in each ordered record is determined. When the symbol type is an increasing symbol, it indicates that the load value is continuously increasing within the sampling interval corresponding to that symbol segment. When the symbol type is a decreasing symbol, it indicates that the load value is continuously decreasing within the sampling interval corresponding to that symbol segment. When the symbol type is a neutral symbol, it indicates that the load value remains unchanged within the sampling interval corresponding to that symbol segment; this neutral symbol segment is not considered a load change segment. For load value channel symbol segments with increasing or decreasing symbol types, these load value channel symbol segments are recorded as load change segments. The sign type of the load change segment is recorded as the disturbance direction, where the increasing sign corresponds to the load rising direction and the decreasing sign corresponds to the load falling direction. The first sampling number of the load change segment is recorded as the disturbance start point, and the last sampling number of the load change segment is recorded as the disturbance end point. At the same time, the first and last sampling values ​​of the load change segment are retained to represent the load values ​​at the disturbance start point and disturbance end point. After completing the above recording, the load disturbance record is output. Through this processing, the continuously rising or continuously falling load process is treated as a complete disturbance object, rather than being split into multiple single-point changes. When subsequently extracting the current temperature rise response segment, the response range can be determined around the load change segment. In S2-2, the cooling status value channel symbol segment in the transformer's ordered operation data is used as input. The ordered records with the channel name "cooling status value channel" are read one by one, and the first sampled value, last sampled value, first sampled number, and last sampled number of the channel symbol segment are read. The cooling status value is represented by discrete values, where "stop" indicates the cooler is not in operation, and "start" indicates the cooler is in operation. When an adjacent sampled status changes from "stop" to "start," it indicates the cooler is in operation at the corresponding sampled number; when an adjacent sampled status changes from "start" to "stop," it indicates the cooler is out of operation at the corresponding sampled number. Specifically, within the cooling status value channel, two adjacent cooling status values ​​are read according to their sampled numbers. If the previous sampled value is "stop" and the next sampled value is... If the current state is "on", the sampling number corresponding to the next sampled value is recorded as the cooling change point. If the previous sampled value is "on" and the next sampled value is "off", the sampling number corresponding to the next sampled value is recorded as the cooling change point. If two adjacent cooling state values ​​are both "off" or both are "on", no cooling change point is generated. Since the cooling state change is a point event rather than a continuous change segment, the cooling change point is recorded as both the disturbance start point and the disturbance end point, and the cooling change type is recorded as "off to start" or "start to stop", and the cooling disturbance record is output. Through this processing, the operation or shutdown of the cooler can be used as an independent transformer operation disturbance to participate in the temperature rise response segment interception, so that subsequent analysis can distinguish between the temperature rise change caused by the load and the oil temperature transition caused by the cooling action. In S2-3, using load disturbance records or cooling disturbance records as input, the disturbance start and end points in the disturbance records are read first. Then, in the load value channel symbol segment before the disturbance start, the sampling number of the most recent load value symbol type that changed from an increasing symbol to a decreasing symbol or from a decreasing symbol to an increasing symbol is searched backward, and this sampling number is used as the segment start. In specific execution, if there is a continuous relationship between increasing, flat, and decreasing symbol segments in the load value channel symbol segment, the flat symbol segment is not considered as a reverse object, and the first sampling number of the decreasing symbol segment is taken as the sampling number for the change from increasing to decreasing. If there is a continuous relationship between decreasing, flat, and increasing symbol segments, the first sampling number of the increasing symbol segment is taken as the sampling number for the change from decreasing to increasing. If there is no reverse position of load value symbol type before the disturbance start, the first sampling number in the transformer's orderly operation data that is earlier than the disturbance start is used as the segment start. Subsequently, in the oil temperature value channel symbol segment after the disturbance end, the first oil temperature value symbol type that changed from an increasing symbol to an increasing symbol is searched backward. The sampling number that changes from a decrement sign to an increment sign is used as the segment endpoint. If the oil temperature value first appears as a flat sign segment and then as a reverse sign segment, the first sampling number of the reverse sign segment is used as the segment endpoint. If the oil temperature value sign type has not yet reversed after the disturbance endpoint, the current data segment is recorded as an unclosed temperature rise response segment, and the segment endpoint is searched again after subsequent sampling data enters the transformer's orderly operation data. After determining the segment start point and segment endpoint, the transformer's orderly operation data between the segment start point and segment endpoint is extracted, and the load value channel sign segment, oil temperature value channel sign segment, winding temperature value channel sign segment, cooling status value channel sign segment, and status monitoring value channel sign segment are retained. The current temperature rise response segment is then output. Through this processing, the starting point of the current temperature rise response segment comes from the most recent load direction change before the disturbance, and the ending point comes from the oil temperature response direction change after the disturbance, so that the segment can cover the temperature rise response process of the transformer after being triggered by the actual operation disturbance. Through the above implementation method, S2 transforms the continuous load changes and cooling state switching in the orderly operation data of the transformer into executable transformer operation disturbances, and uses the reversal of load sign and oil temperature sign as the segment boundaries to avoid missing response processes or mixing in irrelevant data due to fixed time window interception; at the same time, both load disturbance records and cooling disturbance records are represented by disturbance start point, disturbance end point and disturbance direction, so that S3 can continue to read the disturbance source, disturbance direction, cooling state before disturbance and load sequence before disturbance, and select the reference temperature rise response segment sequence from the stored temperature rise response segments; in practical applications: for example, if the load value of a transformer continuously increases between sampling sequence number 20 and 35, S2-1 records the load value increase segment as the load change segment, and records the sampled data... Sample number 20 is recorded as the disturbance start point and sample number 35 as the disturbance end point. If the most recent change in load symbol from decreasing to increasing before sample number 18 occurs at sample number 16, then S2-3 takes sample number 16 as the segment start point. If the oil temperature value continues to increase after the load increases, and changes from an increasing symbol to a decreasing symbol at sample number 52, then sample number 52 is taken as the segment end point, and the orderly operation data of the transformer between sample numbers 16 and 52 is extracted as the current temperature rise response segment. If the cooling status value changes from a stopped transformer to a started transformer at sample number 40 in another scenario, then S2-2 takes sample number 40 as the cooling change point, and S2-3 continues to search forward for the reverse position of the load symbol and backward for the reverse position of the oil temperature symbol, thus forming the current temperature rise response segment triggered by the cooler being put into operation.

[0020] S3. Read the disturbance source, disturbance direction, pre-disturbance cooling state, pre-disturbance load sequence, first change sequence of oil temperature, first change sequence of winding temperature, and first change sequence of status monitoring value in the current temperature rise response segment. Select segments with the same disturbance source, the same disturbance direction, and the same pre-disturbance cooling state from the stored temperature rise response segments, and arrange them from near to far according to the pre-disturbance load sequence. Output the reference temperature rise response segment sequence. In this embodiment, S3 is used to select historical segments that can serve as reference objects for the current temperature rise response segment from the existing temperature rise response segments based on the current temperature rise response segment output by S2, and to perform redundancy removal and sorting processing on the historical segments. Since the temperature rise response sequence of the same transformer is not the same under different load levels, different cooling states, and different disturbance directions, if the search is only based on time proximity or simple field similarity, it is easy to select historical segments that are not comparable to the current operating conditions. Therefore, this embodiment first performs hard consistency screening based on disturbance source, disturbance direction, and cooling state before disturbance, and then forms a segment sequence vector by combining the load sequence before disturbance, the first change sequence of oil temperature, the first change sequence of winding temperature, and the first change sequence of state monitoring value. Subsequently, linearly repeating candidate segments are deleted through orthogonal decomposition, and the reconstruction contribution of each candidate segment to the current sequence vector is calculated using the inverse matrix. Finally, the segments are arranged according to both load proximity and reconstruction contribution to form a reference temperature rise response segment sequence. This implementation process includes the following steps: In S3-1, the inputs are the disturbance source, disturbance direction, pre-disturbance cooling state, pre-disturbance load sequence, first oil temperature change sequence, first winding temperature change sequence, and first status monitoring value change sequence in the current temperature rise response segment. The disturbance source is either a load disturbance or a cooling disturbance; the disturbance direction is an increase or decrease sign in load disturbances and a stop-start or start-stop sign in cooling disturbances; the pre-disturbance cooling state is the last state value in the cooling state value channel before the disturbance start point in the current temperature rise response segment; the pre-disturbance load sequence is the position of the load value before the disturbance start point in the ordered list of stored load values ​​in the current temperature rise response segment; and the first oil temperature change sequence, first winding temperature change sequence, and first status monitoring value change sequence are the first occurrence positions of oil temperature change, winding temperature change, and status monitoring value change in the event sequence of the current temperature rise response segment, respectively. Execution... First, the disturbance source, disturbance direction, and pre-disturbance cooling state of each stored temperature rise response segment are read sequentially. Only historical segments with the same disturbance source, disturbance direction, and pre-disturbance cooling state are retained. For each retained segment, a segment sequence vector is formed according to a fixed field order of pre-disturbance load sequence, oil temperature first change sequence, winding temperature first change sequence, and condition monitoring value first change sequence. The current temperature rise response segment is also formed into a current sequence vector according to the same field order. If an oil temperature change, winding temperature change, or condition monitoring value change does not occur in a certain retained segment, an event missing code is written in the corresponding field. This event missing code serves as the determined value of the field in subsequent matrix operations without changing the field order. After the above processing is completed, multiple segment sequence vectors form a candidate segment vector set, and the current sequence vector serves as the target vector for subsequent reconstruction calculations. In S3-2, the candidate segment vector set and the current sequence vector of the current temperature rise response segment are used as inputs. Each segment sequence vector in the candidate segment vector set is used as a column of the candidate matrix, and the columns are arranged according to the reading order of the candidate segments in the stored temperature rise response segments to obtain the candidate matrix. The rows of the candidate matrix correspond to the pre-disturbance load sequence, the first change sequence of oil temperature, the first change sequence of winding temperature, and the first change sequence of condition monitoring value. The columns of the candidate matrix correspond to each candidate segment. Then, orthogonal decomposition is performed on the candidate matrix to obtain an orthogonal matrix and an upper triangular matrix. The orthogonal matrix is ​​used to represent the projection relationship of the candidate segment vectors in mutually orthogonal directions, and the upper triangular matrix is ​​used to represent the linear dependence relationship between each candidate segment vector. The diagonal elements of the upper triangular matrix are read. When a diagonal element is zero, it indicates that the corresponding element is zero. The candidate segment vector of a column can be linearly represented by the preceding candidate segment vector. Since this column no longer provides a new reference direction, the candidate matrix column corresponding to the diagonal element is deleted, and the candidate segment corresponding to this column is deleted simultaneously. After deletion, the remaining candidate matrix columns form a non-zero candidate matrix, and the remaining candidate segments form a set of de-redundant candidate segments. Since there is no linear repetition between the column vectors of the non-zero candidate matrix represented by zero diagonal elements, the transpose product matrix of the non-zero candidate matrix can be used for inverse matrix calculation. In practice, the non-zero candidate matrix is ​​transposed and multiplied with the non-zero candidate matrix to obtain the transpose product matrix. Then, the transpose product matrix is ​​inverted to obtain the inverse matrix. This inverse matrix and the non-zero candidate matrix are used together to calculate the reconstruction coefficients of each de-redundant candidate segment for the current order vector. In S3-3, the input consists of a set of candidate fragments to be deduplicated, a non-zero column candidate matrix, an inverse matrix, and the current order vector. The linear reconstruction coefficients of the current order vector on the non-zero column candidate matrix are calculated using the inverse matrix. Specifically, the transpose of the non-zero column candidate matrix is ​​multiplied by the current order vector to obtain the projected column vectors of the current order vector in each candidate fragment direction. Then, the inverse matrix is ​​multiplied by these projected column vectors to obtain the reconstruction coefficients corresponding to each deduplicated candidate fragment. These reconstruction coefficients represent the contribution of the corresponding candidate fragment to the reconstruction of the current order vector. Subsequently, a removal calculation is performed on each candidate fragment in the set of candidate fragments to be deduplicated: each time, a candidate fragment and its corresponding column in the non-zero column candidate matrix are removed. The remaining columns are used to reconstruct the current order vector linearly. The residual between the reconstruction result before removal and the current order vector is calculated, and the residual between the reconstruction result after removal and the current order vector is calculated. The residual between the order vectors is calculated by subtracting the residual before removal from the residual after removal, resulting in the change in the reconstruction residual corresponding to the candidate segment. The change in the reconstruction residual indicates the degree of deviation of the current order vector after the removal of the candidate segment. The larger the change in the reconstruction residual, the stronger the reference contribution of the candidate segment to the current temperature rise response segment. After calculating the change in the reconstruction residual for all redundancy-removed candidate segments, the order difference between the pre-disturbance load order of each redundancy-removed candidate segment and the pre-disturbance load order of the current temperature rise response segment is calculated and arranged in ascending order of order difference. When the order difference of two redundancy-removed candidate segments is the same, they are arranged in descending order of reconstruction residual change to obtain the reference temperature rise response segment sequence. This sorting rule prioritizes segments with similar load conditions as references, while retaining segments with higher reconstruction contribution to the current order vector when the load order is the same or similar. Through the above implementation method, S3 does not simply retrieve historical segments from the existing temperature rise response segments based on the same field. Instead, it first ensures that the historical segments and the current temperature rise response segment are under the same disturbance semantics by considering the disturbance source, disturbance direction, and pre-disturbance cooling state. Then, it transforms the load position and the first occurrence position of multiple response events into calculable objects through the segment sequence vector. Furthermore, it removes duplicate reference directions through orthogonal decomposition and determines the reference contribution of each candidate segment to the current temperature rise response segment through the inverse matrix and the change in the reconstruction residual. This results in the output of a reference temperature rise response segment sequence that is both consistent with the transformer operating conditions and has computational distinguishability. In practical applications, for example, the disturbance source of the current temperature rise response segment is a load disturbance, the disturbance direction is an increasing sign, the pre-disturbance cooling state is stopped, the pre-disturbance load sequence is 120, and the oil temperature is... The first change sequence is 3, the first change sequence of winding temperature is 4, and the first change sequence of status monitoring value is 5. The edge computing node first retains historical segments from the existing temperature rise response segments that are all load disturbances, all have the same increase sign, and all have the same cooling state of being stopped before the disturbance. Then, each historical segment is converted into a four-field segment sequence vector. If the four-field sequence relationship of some historical segments is linearly repeated, the repeated columns are deleted through orthogonal decomposition. Then, the reconstruction coefficient of each historical segment to the current four-field sequence vector is calculated using the inverse matrix of the non-zero column candidate matrix, and the reconstruction residual change is calculated one by one by removing historical segments. Finally, historical segments with a load sequence close to 120 before the disturbance and a large residual change after removal are placed at the beginning of the reference temperature rise response segment sequence for subsequent S4 to compare the response order of the current temperature rise response segment and the reference temperature rise response segment.

[0021] S4. Compare the current temperature rise response segment with the reference temperature rise response segment sequence segment by segment, and generate the current response sequence string and the reference response sequence string according to the sampling order. The response sequence string consists of the order of occurrence of load change, cooling change, oil temperature change, winding temperature change, status monitoring value change and oil temperature reversal. Read the position of the same event in the two segments one by one and output the sequence difference item of the current temperature rise response segment. In this embodiment, S4 is used to compare the sequence of reference temperature rise response segments output by S3 with the current temperature rise response segments output by S2, and generate a sequence difference term that can characterize the deviation of the current temperature rise response process. Since the oil temperature change, winding temperature change, condition monitoring value change and oil temperature reversal do not occur simultaneously after the transformer is triggered by load change or cooling change, but form a response chain according to a certain sampling order, this embodiment does not directly compare the magnitude of individual values. Instead, it first converts each temperature rise response segment into a response sequence string, then converts the response sequence string into a computable sequence vector, then uses the reference sequence vector set to construct a third-order sequence tensor and extracts the reference sequence basis. Finally, it obtains the reconstruction result of the current sequence vector relative to the reference sequence basis through regularized linear solution, and outputs the sequence difference term based on the event-by-event difference between the reconstruction result and the current sequence vector. This implementation process includes the following steps: In S4-1, using the current temperature rise response segment and the reference temperature rise response segment sequence as input, the first sampling sequence number of load change, cooling change, oil temperature change, winding temperature change, condition monitoring value change, and oil temperature reversal is read within each temperature rise response segment. Specifically, the first sampling sequence number of load change is taken from the disturbance start point of the load change segment; the first sampling sequence number of cooling change is taken from the cooling change point; the first sampling sequence number of oil temperature change is taken from the position where the oil temperature value channel symbol segment first shows an increase or decrease sign after the disturbance start point; the first sampling sequence number of winding temperature change is taken from the position where the winding temperature value channel symbol segment first shows an increase or decrease sign after the disturbance start point; the first sampling sequence number of condition monitoring value change is taken from the position where the condition monitoring value channel symbol segment first shows an increase or decrease sign after the disturbance start point; and the first sampling sequence number of oil temperature reversal is taken from the position where the oil temperature value channel symbol type first changes from increase to decrease or from decrease to increase after the disturbance end point. For the current... For each temperature rise response segment, the events are arranged in ascending order of their first sampling numbers to generate the current response sequence string. For each reference temperature rise response segment in the reference temperature rise response segment sequence, a corresponding reference response sequence string is generated according to the same rules. If an event does not appear in a temperature rise response segment, it is not written into the sorting position of the response sequence string, and an event missing marker is written for the event in the subsequent sequence vector. Subsequently, each response sequence string is converted into a sequence vector, which includes three fields: event position, adjacent event interval, and event missing marker. The event position is the position of the event in the response sequence string, the adjacent event interval is the difference between the first sampling numbers of two adjacent events, and the event missing marker is used to record that the event did not appear in the temperature rise response segment. After the conversion, the current response sequence string forms the current sequence vector, the reference response sequence strings form reference sequence vectors, and multiple reference sequence vectors form a reference sequence vector set. In S4-2, a third-order order tensor is constructed using a reference order vector set as input, based on event type, segment number, and order field. The event type dimension corresponds to load change, cooling change, oil temperature change, winding temperature change, condition monitoring value change, and oil temperature reversal; the segment number dimension corresponds to the arrangement position of each reference temperature rise response segment in the reference temperature rise response segment sequence; and the order field dimension corresponds to the event position, adjacent event interval, and event missing marker. During construction, the event position, adjacent event interval, and event missing marker for the same event type in each reference order vector are written into the corresponding event type dimension, segment number dimension, and order field dimension of the third-order order tensor, so that the order relationship in the reference temperature rise response segment sequence is represented as unified tensor data. Subsequently, tensor decomposition is performed on the third-order order tensor to obtain event factor moments. The system consists of an event factor matrix, a fragment factor matrix, and a field factor matrix. The event factor matrix represents the common direction of change for different event types in the reference order relationship. The fragment factor matrix represents the contribution of different reference temperature rise response segments to the common order relationship. The field factor matrix represents the field contribution of event position, adjacent event interval, and event missing marker in the order relationship. After decomposition, the event factor matrix and field factor matrix are combined to reconstruct the reference order basis, compressing the repeated event sequence and adjacent interval relationships in the reference temperature rise response segment sequence into a reference order basis matrix. The fragment factor matrix is ​​used to weaken the repeated expression of the same order relationship by repeated fragments during the reconstruction process, so that the reference order basis matrix retains the stable order relationship in the reference order vector set and reduces the redundancy caused by multiple similar reference temperature rise response segments. In S4-3, a regularized linear equation is constructed using the current sequence vector and the reference sequence basis matrix as input, enabling the current sequence vector to be linearly reconstructed from the reference sequence basis matrix. Specifically, the reference sequence basis matrix is ​​multiplied by the coefficients to be reconstructed to obtain the reconstructed sequence vector. The sum of squared differences between the current sequence vector and the reconstructed sequence vector is used as the linear reconstruction error, and the sum of squared reconstruction coefficients is used as a regularization term to prevent the reconstruction result from being overly dominated by a single reference sequence basis. An event missing penalty term is used as an event integrity constraint, calculated based on whether the event missing markers for the same event are consistent in the current and reconstructed sequence vectors (consistent is recorded as zero, inconsistent is recorded as one), and the calculation results for each event are summed. The goal is to minimize the sum of the linear reconstruction error, the sum of squared reconstruction coefficients, and the event missing penalty term to solve the regularized linear equation. The equation yields the reconstruction coefficients of the current sequence vector on the reference sequence basis matrix, and a reconstructed sequence vector is generated based on these coefficients. Subsequently, the reconstructed sequence vector is aligned with the current sequence vector according to event type, and the event position, adjacent event interval, and event missing flag for the same event are read. When the event position count in the current sequence vector is less than that in the reconstructed sequence vector, the event position is shifted forward; when the event position count in the current sequence vector is greater than that in the reconstructed sequence vector, the event position is shifted backward; when the interval between adjacent events is different, the adjacent event interval is changed; when the event missing flag for the same event differs between the current and reconstructed sequence vectors, the event is missing. These outputs together form the sequence difference term, which is used in subsequent S5 generation of transformer operating status analysis results. Through the above implementation, S4 transforms the comparison object between the current temperature rise response segment and the reference temperature rise response segment sequence from a single sample value to the relationship of event occurrence order, event interval, and event missing events. This enables edge computing nodes to identify the position shift, position shift, interval change, and event missing of the current temperature rise response process relative to the historical reference order, rather than relying on fixed thresholds or single-point exceedance judgments. Simultaneously, the third-order order tensor and tensor decomposition are used to extract stable order relationships from multiple reference temperature rise response segments, and the regularized linear equation is used to project the current order vector onto the reference order basis and output verifiable order difference terms. This provides clear criteria for abnormal temperature rise recovery, early winding thermal response, pre-emptive anomalies in condition monitoring, and consistent disturbance response in S5. Data source; In practical applications: For example, in the current temperature rise response segment, the order of events is load change, oil temperature change, winding temperature change, condition monitoring value change, and oil temperature reversal. The reference sequence base formed after tensor decomposition of multiple reference temperature rise response segments indicates that under the same load disturbance, the usual order is load change, oil temperature change, winding temperature change, oil temperature reversal, and condition monitoring value change. Then, when comparing events one by one, S4-3 will find that the condition monitoring value change in the current sequence vector is earlier than the reconstructed reference sequence position, and the oil temperature reversal position is shifted backward. Thus, two sequence difference items are output: the position of condition monitoring value change is shifted forward and the position of oil temperature reversal is shifted backward. Subsequently, S5 can generate the corresponding transformer operation status analysis results based on these sequence difference items.

[0022] S5. Generate transformer operating status analysis results based on sequence difference items, and write the current temperature rise response segment, the current response sequence string and the recovery sampling position after oil temperature reversal into the stored temperature rise response segment, and output the updated stored temperature rise response segment. In this embodiment, S5 is used to convert the sequence difference items obtained in S4 into transformer operating status analysis results, and write the data generated in this analysis into the stored temperature rise response segment for subsequent reference and sorting. Since the sequence difference items only indicate changes in event position or missing events, they cannot be directly used as operating status conclusions. Therefore, this step first generates transformer status difference records, then generates status results according to event correspondence, and finally writes them back according to segment numbers. This implementation process includes the following steps: In S5-1, taking the sequence difference item as input, it reads the order shift items, order shift items, and event missing items formed by oil temperature change, winding temperature change, status monitoring value change, and oil temperature reversal relative to load change or cooling change. In specific execution, if the event position in the current response sequence string is less than the event position in the reconstructed sequence vector, it is recorded as order shift; if the event position in the current response sequence string is greater than the event position in the reconstructed sequence vector, it is recorded as order shift; if an event does not appear in the current temperature rise response segment, it is recorded as event missing. Subsequently, a status difference table is generated according to the event name, difference type, and corresponding sampling number. The event name includes oil temperature change, winding temperature change, status monitoring value change, and oil temperature reversal; the difference type includes order shift, order shift, and event missing; the corresponding sampling number is the first sampling number of the event in the current temperature rise response segment; when an event is missing, the corresponding sampling number is written as a missing flag; and the transformer status difference record is output. In S5-2, using transformer state difference records as input, it reads missing items for reverse oil temperature, items with reverse oil temperature shifted position, items where winding temperature change precedes oil temperature change, and items where state monitoring value change precedes oil temperature change. During execution, for records with the event name "Oil Temperature Reversal" and the difference type "Event Missing," a missing oil temperature reverse item is generated; for records with the event name "Oil Temperature Reversal" and the difference type "Shifted Position," an item with reverse oil temperature shifted position is generated; if the event position of winding temperature change in the current response sequence is less than the event position of oil temperature change, an item with winding temperature change preceding oil temperature change is generated; if the event position of state monitoring value change in the current response sequence is less than the event position of oil temperature change, an item with state monitoring value change preceding oil temperature change is generated; the missing oil temperature reverse item and the item with reverse oil temperature shifted position are associated with temperature rise recovery anomalies. The reason is that the oil temperature is not reversed or the oil temperature reversed is delayed, indicating that the thermal response after the disturbance did not fall back in the reference order; the item that changes in winding temperature earlier than oil temperature is corresponding to the item that the winding thermal response is advanced, because the winding temperature changes before the oil temperature, indicating that the thermal change occurs first on the winding side; the item that changes in status monitoring value earlier than oil temperature is corresponding to the item that the status monitoring value changes before the oil temperature, indicating that the discharge, vibration or other status quantities change first; if the above four types of items do not exist in the transformer status difference record, then the current temperature rise response segment is corresponding to the disturbance response consistency, and the transformer operation status analysis result is output; if the same current temperature rise response segment forms multiple correspondences, then multiple status names are retained in the transformer operation status analysis result, and the corresponding event name, difference type and corresponding sampling sequence number are bound to them respectively; In S5-3, the current temperature rise response segment, the current response sequence, the recovery sampling position after oil temperature reversal, and the transformer operating status analysis results are used as inputs to generate a segment number for the current temperature rise response segment, and the four data items are written into the stored temperature rise response segment according to the segment number; the current temperature rise response segment stores the orderly operation data of the transformer from the segment start point to the segment end point; the current response sequence stores the order of occurrence of load change, cooling change, oil temperature change, winding temperature change, status monitoring value change, and oil temperature reversal; the recovery sampling position after oil temperature reversal stores the oil temperature value during the oil temperature reversal... The sampling sequence number is used to enter the oil temperature range before the disturbance; the transformer operation status analysis results are saved for abnormal temperature rise recovery, early winding thermal response, pre-anomaly in status monitoring, or consistent disturbance response; if the recovery sampling position after the oil temperature reversal is not formed during writing, the recovery sampling position is written to the missing marker, and when the oil temperature value enters the oil temperature range before the disturbance in subsequent sampling data, the recovery sampling position is written according to the same segment number; after writing, the segment number, recovery sampling position, and transformer operation status analysis results are used as index fields to output the updated stored temperature rise response segment; Through the above implementation method, S5 transforms the position changes and event omissions in the sequence difference items into operating status results, and writes the current temperature rise response segment into the stored temperature rise response segment, so that S3 can continue to call this segment as reference data. This processing avoids only outputting abnormal scores or general alarms, but instead maps the sequence differences of oil temperature reversal, winding temperature change, and status monitoring value change to temperature rise recovery, winding thermal response, and status monitoring pre-change, respectively. In practical applications: for example, if a current temperature rise response segment is triggered by a load increase, S4 outputs that the oil temperature reversal position shifts backward and the status monitoring value change occurs earlier than the oil temperature change. S5-1 writes the two items into the transformer status difference record, S5-2 outputs temperature rise recovery anomaly and status monitoring pre-change anomaly, and S5-3 writes the current temperature rise response segment, the current response sequence string, the recovery sampling position after oil temperature reversal, and the two status results into the stored temperature rise response segment according to the same segment number. When a similar load increase disturbance occurs later, the edge computing node calls this segment to participate in reference sorting and sequence comparison.

[0023] Furthermore, it also includes: a real-time analysis system for transformer operating status oriented to edge computing, comprising a symbol encoding module, a disturbance truncation module, a reference sorting module, a sequence comparison module, and a result write-back module, characterized in that; The symbol encoding module is used to acquire load values, oil temperature values, winding temperature values, cooling status values ​​and status monitoring values ​​collected on-site by the transformer. It generates three types of change symbols (increase, decrease, and remain unchanged) for each type of value according to the sampling order, and binds each change symbol with the corresponding sampling sequence number to output the orderly operation data of the transformer. The disturbance truncation module is used to take the orderly operation data of the transformer as input, merge the consecutive identical increase or decrease symbols in the load value into a load change segment, record the sampling sequence number of the cooling status value from shutdown to start or from start to shutdown as the cooling change point, and take the load change segment or cooling change point as the transformer operation disturbance, extract the data between the most recent load symbol reversal point before the disturbance and the first reversal point of the oil temperature value after the disturbance, and output the current temperature rise response segment. The reference sorting module is used to read the disturbance source, disturbance direction, pre-disturbance cooling state, pre-disturbance load sequence, first change sequence of oil temperature, first change sequence of winding temperature, and first change sequence of status monitoring value in the current temperature rise response segment. It selects segments with the same disturbance source, the same disturbance direction, and the same pre-disturbance cooling state from the stored temperature rise response segments, and arranges them from near to far according to the pre-disturbance load sequence, and outputs the reference temperature rise response segment sequence. The sequence comparison module is used to compare the current temperature rise response segment with the reference temperature rise response segment sequence segment by segment. It generates the current response sequence string and the reference response sequence string according to the sampling order. The response sequence string consists of the order of occurrence of load change, cooling change, oil temperature change, winding temperature change, status monitoring value change and oil temperature reversal. It reads the position of the same event in the two bit by bit and outputs the sequence difference item of the current temperature rise response segment. The result write-back module generates transformer operating status analysis results based on the sequence difference items, and writes the current temperature rise response segment, the current response sequence string, and the recovery sampling position after the oil temperature reverses into the stored temperature rise response segment, and outputs the updated stored temperature rise response segment.

[0024] Working Principle: The system utilizes naturally occurring load changes and cooler start-up / shutdown during transformer operation as analysis trigger points. First, load values, oil temperature values, winding temperature values, cooling status values, and status monitoring values ​​are converted into increase, decrease, and equilibrium symbols to form ordered operating data. Then, load change segments or cooling change points are identified, and the current temperature rise response segment is formed by extracting the segments before and after the disturbance. Subsequently, segments with the same disturbance source, disturbance direction, and cooling status are selected from the existing temperature rise response segments as references. The order of occurrence of load changes, cooling changes, oil temperature changes, winding temperature changes, status monitoring value changes, and reverse oil temperature changes in the current segment and the reference segment are compared to obtain the order difference item. Finally, based on this, the system outputs abnormal temperature rise recovery, premature winding thermal response, pre-emptive status monitoring anomaly, or consistent disturbance response, and writes this segment back as a subsequent reference. For example, in an unattended power distribution room, edge computing nodes continuously collect load, oil temperature, winding temperature, cooling status, and vibration data of a transformer. When the load continuously increases, the system treats it as an operational disturbance, extracts data from before the load increases to the point where the oil temperature reverses, and compares it with the temperature rise response segments under similar load increases in the past. If the oil temperature reverses later than the reference sequence or does not occur, it is judged as an abnormal temperature rise recovery. If the winding temperature changes earlier than the oil temperature, it is judged as an early winding thermal response. If the vibration and other status monitoring values ​​change earlier than the oil temperature, it is judged as a pre-emergence anomaly in status monitoring. In this way, abnormal trends can be judged at the edge using real operational fluctuations.

[0025] In conclusion, the above description is only 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 transformer operating state real-time analysis method for edge computing, characterized in that, include: S1. Acquire the load value, oil temperature value, winding temperature value, cooling status value and status monitoring value collected on the transformer site, generate three types of change symbols (increase, decrease and remain unchanged) for each type of value according to the sampling order, bind each change symbol with the corresponding sampling sequence number, and output the orderly operation data of the transformer. S2. Using the orderly operation data of the transformer as input, merge the consecutive identical increase or decrease symbols in the load value into a load change segment. Record the sampling sequence number of the cooling status value from shutdown to start or from start to shutdown as the cooling change point. Use the load change segment or cooling change point as the transformer operation disturbance. Extract the data between the most recent load symbol reversal point before the disturbance and the first reversal point of the oil temperature value after the disturbance, and output the current temperature rise response segment. S3. Read the disturbance source, disturbance direction, pre-disturbance cooling state, pre-disturbance load sequence, first change sequence of oil temperature, first change sequence of winding temperature, and first change sequence of status monitoring value in the current temperature rise response segment. Select segments with the same disturbance source, the same disturbance direction, and the same pre-disturbance cooling state from the stored temperature rise response segments, and arrange them from near to far according to the pre-disturbance load sequence. Output the reference temperature rise response segment sequence. S4. Compare the current temperature rise response segment with the reference temperature rise response segment sequence segment by segment, and generate the current response sequence string and the reference response sequence string according to the sampling order. The response sequence string consists of the order of occurrence of load change, cooling change, oil temperature change, winding temperature change, status monitoring value change and oil temperature reversal. Read the position of the same event in the two segments one by one and output the sequence difference item of the current temperature rise response segment. S5. Generate transformer operating status analysis results based on sequence difference items, and write the current temperature rise response segment, the current response sequence string, and the recovery sampling position after oil temperature reversal into the stored temperature rise response segment, and output the updated stored temperature rise response segment.

2. The real-time analysis method for transformer operating status based on edge computing as described in claim 1, characterized in that: S1 includes: S1-1: Taking the nth sample value and the (n-1)th sample value in the same acquisition channel as input, compare the size relationship between the nth sample value and the (n-1)th sample value. If the nth sample value is greater than the (n-1)th sample value, generate an increment sign; if the nth sample value is less than the (n-1)th sample value, generate a decrement sign; if the nth sample value is equal to the (n-1)th sample value, generate a neutral sign. Output the channel change sign with the sampling sequence number n. S1-2. Using the channel change symbols of each acquisition channel as input, merge consecutive identical channel change symbols within the same acquisition channel into a symbol segment, and record the first sampling number, the last sampling number, the first sample value, and the last sample value of the symbol segment, and output the channel symbol segment; S1-3. Using the channel symbol segments of each acquisition channel as input, arrange them in order of first sampling sequence number, and bind each channel symbol segment with the corresponding acquisition channel name, symbol type, first sampling sequence number, last sampling sequence number, first sampling value and last sampling value, and output the orderly operation data of the transformer.

3. The real-time analysis method for transformer operating status based on edge computing according to claim 2, characterized in that: S2 includes: S2-1. Using the load value channel symbol segment in the orderly operation data of the transformer as input, the load value channel symbol segment with the symbol type of increase symbol or decrease symbol is recorded as the load change segment, and the symbol type of the load change segment is recorded as the disturbance direction, the starting sampling number is recorded as the disturbance start point, and the ending sampling number is recorded as the disturbance end point. Output the load disturbance record. S2-2. Using the cooling status value channel symbol segment in the orderly operation data of the transformer as input, read the sampling sequence number corresponding to the cooling status value from shutdown to start or from start to shutdown, record the sampling sequence number corresponding to the cooling status value from shutdown to start or from start to stop as the cooling change point, and record the cooling change point as the disturbance start point and disturbance end point, and output the cooling disturbance record. S2-3. Using load disturbance records or cooling disturbance records as input, before the disturbance start point, read the sampling number of the most recent load value symbol type that changes from increasing to decreasing or from decreasing to increasing as the segment start point, and after the disturbance end point, read the sampling number of the oil temperature value symbol type that changes from increasing to decreasing or from decreasing to increasing as the segment end point. Extract the orderly operation data of the transformer between the segment start point and the segment end point, and output the current temperature rise response segment.

4. The real-time analysis method for transformer operating status based on edge computing according to claim 3, characterized in that: S3 includes: S3-1. Taking the disturbance source, disturbance direction, pre-disturbance cooling state, pre-disturbance load sequence, first change sequence of oil temperature, first change sequence of winding temperature, and first change sequence of status monitoring value in the current temperature rise response segment as input, retain segments with the same disturbance source, the same disturbance direction, and the same pre-disturbance cooling state from the existing temperature rise response segments. Then, assemble the pre-disturbance load sequence, first change sequence of oil temperature, first change sequence of winding temperature, and first change sequence of status monitoring value of each retained segment into a segment sequence vector according to the same field order, and output the candidate segment vector set.

5. The real-time analysis method for transformer operating status based on edge computing according to claim 4, characterized in that: S3 also includes: S3-2. Taking the candidate segment vector set and the current sequence vector of the current temperature rise response segment as input, the candidate segment vector set is arranged into a candidate matrix by column. The candidate matrix is ​​orthogonally decomposed to obtain an orthogonal matrix and an upper triangular matrix. The columns with diagonal elements of zero and their corresponding candidate segments in the upper triangular matrix are deleted to obtain a non-zero column candidate matrix. The inverse matrix of the transpose product matrix of the non-zero column candidate matrix is ​​calculated. The redundant candidate segment set and the inverse matrix are output. S3-3. Taking the set of candidate segments for redundancy removal, the candidate matrix with non-zero columns, the inverse matrix, and the current order vector as input, calculate the reconstruction coefficient of each candidate segment for redundancy removal to the current order vector through the inverse matrix. Remove each candidate segment for redundancy removal one by one and calculate the change in reconstruction residual before and after removal. Arrange the candidate segments for redundancy removal in order of increasing load order difference before disturbance and decreasing reconstruction residual change, and output the reference temperature rise response segment sequence.

6. The real-time analysis method for transformer operating status based on edge computing according to claim 5, characterized in that: S4 includes: S4-1. Using the current temperature rise response segment and the reference temperature rise response segment sequence as input, read the first sampling sequence number of load change, cooling change, oil temperature change, winding temperature change, condition monitoring value change and oil temperature reversal respectively. Arrange each event in ascending order of the first sampling sequence number to generate the current response sequence string and the reference response sequence string. Convert each response sequence string into a sequence vector composed of event position, adjacent event interval and event missing marker. Output the current sequence vector and the reference sequence vector set.

7. The real-time analysis method for transformer operating status based on edge computing according to claim 6, characterized in that: S4 also includes: S4-2. Using the reference order vector set as input, construct a third-order order tensor according to event type, fragment number and order field. Perform tensor decomposition on the third-order order tensor to obtain the event factor matrix, fragment factor matrix and field factor matrix. Reconstruct the reference order basis using the event factor matrix and field factor matrix, and output the reference order basis matrix after removing duplicate order relations. S4-3. Using the current order vector and the reference order basis matrix as input, construct a regularized linear equation with the goal of minimizing the sum of the linear reconstruction error, the sum of squared reconstruction coefficients, and the event missing penalty term between the current order vector and the reference order basis matrix. Solve to obtain the reconstruction coefficients of the current order vector on the reference order basis matrix. Subtract the reconstructed order vector from the current order vector event by event, and output the order difference term including event position shift, event position shift, change of adjacent event interval, and event missing.

8. The real-time analysis method for transformer operating status based on edge computing according to claim 7, characterized in that: S5 includes: S5-1. Taking the sequential difference item as input, read the order shift items, order shift items, and missing event items of oil temperature change, winding temperature change, status monitoring value change, and oil temperature reverse relative to load change or cooling change, and generate a status difference table according to event name, difference type and corresponding sampling number, and output the transformer status difference record. S5-2. Using the transformer condition difference record as input, read the missing reverse oil temperature item, the reverse oil temperature position shifted item, the winding temperature change earlier than the oil temperature change item, and the condition monitoring value change earlier than the oil temperature change item; correspond the missing reverse oil temperature item and the reverse oil temperature position shifted item as temperature rise recovery anomaly, correspond the winding temperature change earlier than the oil temperature change item as winding thermal response advance, correspond the condition monitoring value change earlier than the oil temperature change item as condition monitoring pre-anomaly, and correspond the condition difference record without a corresponding relationship as disturbance response consistency, and output the transformer operating condition analysis result.

9. The real-time analysis method for transformer operating status based on edge computing according to claim 8, characterized in that: S5 also includes: S5-3. Using the current temperature rise response segment, the current response sequence string, the recovery sampling position after the oil temperature reverses, and the transformer operating status analysis result as inputs, bind the four according to the same segment number and write them into the stored temperature rise response segment. Then, use the written segment number, the recovery sampling position, and the operating status analysis result as index fields to output the updated stored temperature rise response segment.

10. A real-time transformer operation status analysis system for edge computing, comprising a symbol encoding module, a disturbance truncation module, a reference sorting module, a sequence comparison module, and a result write-back module, characterized in that: The symbol encoding module is used to acquire load values, oil temperature values, winding temperature values, cooling status values ​​and status monitoring values ​​collected on-site by the transformer. It generates three types of change symbols (increase, decrease, and remain unchanged) for each type of value according to the sampling order, and binds each change symbol with the corresponding sampling sequence number to output the orderly operation data of the transformer. The disturbance truncation module is used to take the orderly operation data of the transformer as input, merge the consecutive identical increase or decrease symbols in the load value into a load change segment, record the sampling sequence number of the cooling status value from shutdown to start or from start to shutdown as the cooling change point, and take the load change segment or cooling change point as the transformer operation disturbance, extract the data between the most recent load symbol reversal point before the disturbance and the first reversal point of the oil temperature value after the disturbance, and output the current temperature rise response segment. The reference sorting module is used to read the disturbance source, disturbance direction, pre-disturbance cooling state, pre-disturbance load sequence, first change sequence of oil temperature, first change sequence of winding temperature, and first change sequence of status monitoring value in the current temperature rise response segment. It selects segments with the same disturbance source, the same disturbance direction, and the same pre-disturbance cooling state from the stored temperature rise response segments, and arranges them from near to far according to the pre-disturbance load sequence, and outputs the reference temperature rise response segment sequence. The sequence comparison module is used to compare the current temperature rise response segment with the reference temperature rise response segment sequence segment by segment. It generates the current response sequence string and the reference response sequence string according to the sampling order. The response sequence string consists of the order of occurrence of load change, cooling change, oil temperature change, winding temperature change, status monitoring value change and oil temperature reversal. It reads the position of the same event in the two bit by bit and outputs the sequence difference item of the current temperature rise response segment. The result write-back module generates transformer operating status analysis results based on the sequence difference items, and writes the current temperature rise response segment, the current response sequence string, and the recovery sampling position after the oil temperature reverses into the stored temperature rise response segment, and outputs the updated stored temperature rise response segment.

Images