An electrical integrated control test equipment monitoring method based on edge computing

Abstract

The application discloses an electrical integrated control test equipment monitoring method based on edge computing and relates to the technical field of edge computing, which comprises the following steps: collecting and preprocessing original test monitoring data to generate a standardized data set; generating an event template and a current instruction echo actual measurement chain; screening a historical normal segment to generate an instruction echo shadow chain; adopting Smith-Waterman comparison to compare the actual measurement chain and the shadow chain to generate an event alignment table and residual data; constructing a residual entropy Bloom grass map to generate a DeepESN state update queue; inputting an improved DeepESN model in combination with a physical topology mask matrix to generate a broken chain identification result; executing a bypass micro-retest to correct an abnormal source mark; and constructing a broken chain recursive index rope to generate a monitoring result. The application realizes real-time identification, source determination and recurrence tracing of a broken chain abnormality of the electrical integrated control test equipment.

Description

Technical Field

[0001] This invention relates to the field of edge computing technology, and in particular to a monitoring method for electrical integrated control test equipment based on edge computing. Background Technology

[0002] With the increasing application of electrical integrated control testing equipment in relay protection testing, integrated automation commissioning, channel loop verification, and load response verification scenarios, test scripts, test commands, communication messages, relay actions, power output, sampling channels, load responses, and reset recovery data are continuously generated. Existing monitoring methods mostly collect operational data at the equipment end or host computer end, identify test anomalies through threshold judgment, timing statistics, message comparison, or neural network models, and output alarm results.

[0003] Existing methods mostly rely on single-channel sample values, communication status codes, or device operating curves for judgment, lacking an event chain expression oriented towards the execution sequence of test scripts. They cannot connect command issuance, bus confirmation, relay action, power output, load response, sampling stabilization, feedback return, and reset recovery into a comparable command echo chain. When test events are delayed, missing, misaligned, or abnormally transmitted across channels, existing methods struggle to distinguish between internal device chain breaks, external wiring abnormalities, and abnormalities in the tested object.

[0004] Some methods incorporate edge computing or deep learning models to process field data, but most directly input preprocessed sampling features into the model, lacking a shadow chain benchmark composed of historical normal test segments, and lacking event residual screening, physical topology constraints, bypass micro-retesting, and recurrence chain break index structures. When short-term residual mutations, topology propagation breaks, and repeatability anomalies occur during the testing process, existing methods are prone to misjudgments, omissions, and difficulties in tracing, making it difficult to meet the real-time monitoring needs of electrical integrated control testing equipment in the field.

[0005] Therefore, how to provide a monitoring method for electrical integrated control test equipment based on edge computing is a problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0006] One objective of this invention is to propose a monitoring method for electrical integrated control test equipment based on edge computing. This method utilizes command echo chain comparison, residual entropy Bloom plot, improved DeepESN model, and bypass micro-retest to achieve chain break identification and anomaly source marking, and has the advantages of strong real-time performance, high positioning accuracy, and strong traceability.

[0007] A monitoring method for electrical integrated control test equipment based on edge computing according to an embodiment of the present invention includes: Collect raw test and monitoring data and preprocess it to generate a standardized test and monitoring dataset; Based on the standardized test monitoring dataset, a test script semantic event template is generated, and the current instruction echo test chain is generated according to the order of test events. Filter historical event chains from historical normal test segments to generate instruction echo shadow chains; The Smith-Waterman algorithm is used to perform event chain comparison, sparse unit evaluation, bidirectional expansion of central anchor events, and relative sequence difference encoding on the current command echo measured chain and command echo shadow chain to generate an event alignment table and event residual data. Based on the event residual data, construct a residual entropy Bloom plot, filter the entropy trigger event window, and generate a DeepESN state update queue; A physical topology mask matrix is ​​generated based on a standardized test and monitoring dataset. The DeepESN state update queue, event alignment table, event residual data and physical topology mask matrix are input into an improved DeepESN model containing a topology perforation reservoir, a single-step inverse time echo excitation unit and an instantaneous mutual exclusion freeze unit to generate chain break identification results. Based on the set of broken links, suspected abnormal channels are extracted. Standard load bypass micro-retest is performed on the suspected abnormal channels to generate bypass micro-retest results. The initial abnormal source marker is corrected and an abnormal source marker is generated. Based on the chain break identification results and bypass micro-retest results, a chain break recursive index rope is constructed. The recurrence records of the same chain break edge are embedded into the corresponding rope nodes and a chain break tracing index is generated. The monitoring results of the electrical integrated control test equipment are generated and output according to the chain break identification results, abnormal source markers, bypass micro-retest results and chain break tracing index.

[0008] Optionally, the original test monitoring data includes test scripts, test instructions, communication messages, device channel topology data, relay action data, power output data, sampling channel data, load response data, reset recovery data, historical normal test segments, standard load configuration data, bypass switching status, safety interlock status, test task pause marker, monitoring record library, and historical chain break recursive index.

[0009] Optionally, the preprocessing includes timestamp calibration, field unification, duplicate message deletion, missing event marking, unit unification, and event number association.

[0010] Optionally, the step of generating a test script semantic event template based on a standardized test monitoring dataset and generating the current instruction echo test chain according to the order of test events includes: Read the standardized test monitoring dataset and extract the test instruction number, test step number, target channel number, relay number, sampling channel number, load level mark, feedback status code and event timestamp; According to the test steps in the test script, each test instruction is broken down into instruction issuance event, bus confirmation event, relay action event, power output event, load response event, sampling stabilization event, feedback return event, and reset recovery event, and the event sequence number and the test instruction number to which the event belongs are written. Write the target channel number, relay number, sampling channel number, load level mark, and feedback status code into the corresponding event to generate a test script semantic event template; Arrange the events corresponding to the current test command according to the semantic event template and event sequence number of the test script, and write the event timestamp, event duration, sampling amplitude, load response value and feedback status code to generate the current command echo test chain.

[0011] Optionally, the generation of the instruction echo shadow chain includes: Read the test instruction number, target channel number, relay number, load level mark, and event sequence number from the historical normal test segments and the current instruction echo test chain; From historical normal test segments, select those that have the same test command number, target channel number, relay number, and load level mark as the current command echo test chain, and generate candidate historical test segments; Extract corresponding events from candidate historical test segments according to the semantic event template of the test script, and generate a historical event chain; The historical event chain is aligned according to the event sequence number. Based on the event timestamp, event duration, sampling amplitude, load response value and feedback status code corresponding to the same event sequence number in the candidate historical test segment, the field baseline value of each event is generated, and the instruction echo shadow chain is generated.

[0012] Optionally, generating the event alignment table and event residual data includes: Read the event type, event number, test command number, target channel number, relay number, load level mark, event timestamp, event duration, sampling amplitude, load response value, and feedback status code from the current command echo test chain and command echo shadow chain; The event character code is generated according to the event type, target channel number, relay number, and load level mark, and then modulo hash processing is performed to generate the event hash code; Establish a Smith-Waterman alignment matrix, calculate the matching score for matrix cells with the same event hash code, and write empty cell markers for matrix cells with inconsistent event hash codes to complete the sparse cell evaluation; Filter events from the semantic event template of the test script and select the central anchor event, and perform bidirectional local comparison starting from the central anchor event; Read the event sequence number difference of each event relative to the central anchor event, write the event sequence number difference into the comparison symbol, generate an event alignment table based on the comparison path, and generate event residual data according to the event timestamp difference, event duration difference, sampling amplitude difference, load response value difference, and feedback status code difference between aligned events, and write it into the causal edge number.

[0013] Optionally, generating the DeepESN state update queue includes: Read the event number, residual type, event timestamp, causal edge number, event timestamp residual, event duration residual, sampling amplitude residual, load response value residual, and feedback status code difference from the event residual data; Read the historical event residuals corresponding to the same test instruction number, target channel number, relay number and load level mark in the historical normal test segment, and generate the positive judgment boundary and the negative judgment boundary of the residual; The event residual data is converted into a symbolic residual sequence, and the corresponding symbols are written according to the comparison results between the residual and the positive decision boundary of the residual and the negative decision boundary of the residual. The symbolic residual sequence is divided according to the event timestamp order, and the window residual entropy value is generated based on the proportion of the corresponding symbols in each event window. Based on the distribution of window residual entropy values ​​corresponding to historical normal test segments, an entropy decision boundary is generated. Event window numbers whose window residual entropy values ​​exceed the entropy decision boundary are written into the entropy trigger blotting bucket in the residual entropy blotting diagram. The event window number, causal edge number, and residual type are associated and written into the residual entropy blotting diagram. Read the event window number in the entropy trigger cloning bucket, perform forward and reverse cumulative detection on the corresponding event residuals, generate residual cumulative judgment boundary based on the residual cumulative result distribution corresponding to the historical normal test segments, and write the event number, residual type, residual direction and causal edge number that exceed the residual cumulative judgment boundary into the DeepESN state update queue.

[0014] Optionally, generating the chain break identification result includes: Read the device channel topology data from the standardized test and monitoring dataset, extract the connection relationships between communication interfaces, control loops, relay nodes, power output nodes, test channels, load nodes, sampling loops, and feedback message nodes, and generate a physical topology mask matrix; An improved DeepESN model is constructed, which includes a topological perforated reservoir, a single-step reverse-time echo excitation unit, an instantaneous mutual exclusion freezing unit, and a chain break readout layer. The topology perforation reservoir reads the physical topology mask matrix, writes empty connection markers to the positions in the reservoir connection matrix that do not belong to the connection relationship, and writes the DeepESN state update queue, event alignment table and event residual data to the corresponding reservoir node according to the event number and causal edge number to generate the topology perforation reservoir state. The single-step reverse-time echo excitation unit reads the state of the topological perforated reservoir, performs a reverse-time echo calculation based on the transpose of the reservoir connection matrix, generates a reverse-time echo state, and splices the state of the topological perforated reservoir with the reverse-time echo state to generate an echo splicing state. The instantaneous mutual exclusion freeze unit reads the missing event markers in the event alignment table and the trigger event numbers in the DeepESN state update queue, performs sign inversion and one event step freeze on the reserve pool node corresponding to the missing event marker, and generates a freeze pulse state. The broken chain readout layer reads the echo splicing state and the frozen pulse state, and reads the causal edge number and abnormal channel number from the physical topology mask matrix and the DeepESN state update queue to generate the broken chain identification result. The broken chain identification result includes the broken chain edge set, the broken chain event number, the abnormal channel number, and the initial marker of the abnormal source.

[0015] Optionally, generating the anomaly source marker includes: Read the chain break identification results and extract the abnormal channel numbers corresponding to the chain break edges pointing to relay nodes, power output nodes, test channels, load nodes and sampling circuits in the chain break edge set as suspected abnormal channels. Read the standard load configuration data, bypass switching status, safety interlock status and test task pause flag corresponding to the suspected abnormal channel, and write the suspected abnormal channel that meets the requirements of valid bypass switching status, allowed safety interlock status and valid test task pause flag into the bypass micro retest channel set. Perform standard load bypass micro-retest on suspected abnormal channels in the bypass micro-retest channel set to generate bypass command echo chain; Compare the bypass instruction echo chain with the instruction echo shadow chain execution event chain to generate bypass micro-retest results; When the bypass micro-retest result has a broken chain edge that is the same as the broken chain edge set, the initial mark of the abnormal source is corrected to the internal fault mark of the electrical integrated control test equipment. When the bypass micro-retest result does not have a broken chain edge that is the same as the broken chain edge set, the initial mark of the abnormal source is corrected to the non-equipment internal abnormal mark, and an abnormal source mark is generated.

[0016] Optionally, generating and outputting the monitoring results of the electrical integrated control test equipment includes: Read the broken chain edge set, broken chain event number and abnormal channel number from the broken chain identification results; read the abnormal source marker, bypass micro-retest results and historical broken chain recursive index rope in the monitoring record library. Establish the top-level node of the recursive index rope of the broken chain according to the broken chain edge number in the broken chain edge set, and write the broken chain edge number, broken chain event number, abnormal channel number, abnormal source marker and bypass micro-retest result to the top-level node; Search for the top-level rope node that matches the broken chain edge number in the historical broken chain recursive index rope, and embed the current broken chain event number, abnormal channel number, abnormal source marker and bypass micro-retest result into the sub-rope node of the consistent top-level rope node. Generate the number of chain break recurrences, the first chain break event number, the most recent chain break event number, and the chain break recurrence event sequence according to the hierarchical relationship between the top-level rope node and the sub-rope nodes; A chain break tracing index is generated based on the chain break edge number, chain break event number, abnormal channel number, and chain break recurrence event sequence. Monitoring results of electrical integrated control test equipment are generated according to the chain break identification results, abnormal source markers, bypass micro-retest results, chain break recursion index rope, and chain break tracing index.

[0017] The beneficial effects of this invention are: This invention generates a standardized test monitoring dataset by collecting test scripts, test commands, communication messages, device channel topology data, relay action data, power output data, sampling channel data, load response data, reset and recovery data, and historical normal test segments. It also constructs a test script semantic event template, a current command echo measured chain, and a command echo shadow chain. This allows the monitoring of electrical integrated control test equipment to no longer rely solely on single-point sampling values, communication status codes, or fixed thresholds for judgment. Instead, it expresses the test process through a complete event chain consisting of command issuance, bus confirmation, relay action, power output, load echo, sampling stabilization, feedback return, and reset and recovery, thereby improving the structural integrity of test anomaly identification.

[0018] This invention utilizes the Smith-Waterman algorithm to perform sparse unit evaluation, bidirectional expansion of central anchor events, and relative sequence difference encoding on the current command echo measured chain and command echo shadow chain. This generates an event alignment table and event residual data. Based on the event residual data, a residual entropy Bloom plot is constructed to filter entropy-triggered event windows and generate a DeepESN state update queue. This allows delayed events, missing events, misaligned events, and cross-channel residual mutations to be filtered hierarchically, reducing misjudgments caused by normal test fluctuations and improving the accuracy of chain break anomaly localization and edge processing efficiency.

[0019] This invention improves the topological perforation reservoir in the DeepESN model by constraining it with a physical topological mask matrix. It also generates chain break identification results by combining a single-step reverse-time echo excitation unit and an instantaneous mutual exclusion freezing unit. Furthermore, it corrects the initial marker of the abnormal source by using a standard load bypass micro-retest. By recording the recurrence record and chain break traceability index of the same chain break edge through a chain break recursive index rope, it enables internal equipment faults, non-internal equipment anomalies, and recurring chain break records to form traceable monitoring results, thereby improving the ability of real-time on-site monitoring, anomaly source determination, and recurring anomaly retrieval. Attached Figure Description

[0020] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They explain the invention together with the embodiments of the invention and do not constitute a limitation thereof. In the drawings: Figure 1 This is an overall flowchart of a monitoring method for electrical integrated control test equipment based on edge computing proposed in this invention; Figure 2 This is a schematic diagram of the Smith-Waterman algorithm for a monitoring method of electrical integrated control test equipment based on edge computing proposed in this invention. Figure 3 This is a schematic diagram of the structure of the improved DeepESN model of the monitoring method for electrical integrated control test equipment based on edge computing proposed in this invention; Detailed Implementation

[0021] The present invention will now be described in further detail with reference to the accompanying drawings. These drawings are simplified schematic diagrams, illustrating only the basic structure of the invention, and therefore only show the components relevant to the invention.

[0022] refer to Figure 1 , Figure 2 and Figure 3 A monitoring method for electrical integrated control test equipment based on edge computing, comprising: Collect raw test and monitoring data and preprocess it to generate a standardized test and monitoring dataset; Based on the standardized test monitoring dataset, a test script semantic event template is generated, and the current instruction echo test chain is generated according to the order of test events. Filter historical event chains from historical normal test segments to generate instruction echo shadow chains; The Smith-Waterman algorithm is used to perform event chain comparison, sparse unit evaluation, bidirectional expansion of central anchor events, and relative sequence difference encoding on the current command echo measured chain and command echo shadow chain to generate an event alignment table and event residual data. Based on the event residual data, construct a residual entropy Bloom plot, filter the entropy trigger event window, and generate a DeepESN state update queue; A physical topology mask matrix is ​​generated based on a standardized test and monitoring dataset. The DeepESN state update queue, event alignment table, event residual data and physical topology mask matrix are input into an improved DeepESN model containing a topology perforation reservoir, a single-step inverse time echo excitation unit and an instantaneous mutual exclusion freeze unit to generate chain break identification results. Based on the set of broken links, suspected abnormal channels are extracted. Standard load bypass micro-retest is performed on the suspected abnormal channels to generate bypass micro-retest results. The initial abnormal source marker is corrected and an abnormal source marker is generated. Based on the chain break identification results and bypass micro-retest results, a chain break recursive index rope is constructed. The recurrence records of the same chain break edge are embedded into the corresponding rope nodes and a chain break tracing index is generated. The monitoring results of the electrical integrated control test equipment are generated and output according to the chain break identification results, abnormal source markers, bypass micro-retest results and chain break tracing index.

[0023] In this embodiment, the original test monitoring data includes test scripts, test instructions, communication messages, device channel topology data, relay action data, power output data, sampling channel data, load response data, reset recovery data, historical normal test segments, standard load configuration data, bypass switching status, safety interlock status, test task pause marker, monitoring record library, and historical chain break recursive index.

[0024] In this embodiment, the preprocessing includes timestamp calibration, field unification, duplicate message deletion, missing event marking, unit unification, and event number association.

[0025] In this embodiment, the step of generating a test script semantic event template based on a standardized test monitoring dataset and generating the current instruction echo test chain according to the test event sequence includes: Read the standardized test monitoring dataset and extract the test instruction number, test step number, target channel number, relay number, sampling channel number, load level mark, feedback status code and event timestamp; Following the test steps in the test script, each test instruction is broken down into instruction issuance event, bus confirmation event, relay action event, power output event, load response event, sampling stabilization event, feedback return event, and reset recovery event. The event sequence number and the test instruction number to which the event belongs are then written into the script. Specifically, writing the event sequence number and the test instruction number to which the event belongs is as follows: Read the test command issuance time, bus confirmation message time, relay action time, power output start time, load response start time, sample value stabilization time, feedback message return time, and reset recovery completion time corresponding to each test command number. Record the test command issuance time to generate a command issuance event, the bus confirmation message time to generate a bus confirmation event, the relay action time to generate a relay action event, the power output start time to generate a power output event, the load response start time to generate a load response event, the sample value stabilization time to generate a sample stabilization event, the feedback message return time to generate a feedback return event, and the reset recovery completion time to generate a reset recovery event. For the eight types of events under the same test instruction number, write the event sequence numbers 1 to 8 in the order of instruction issuance event, bus confirmation event, relay action event, power output event, load response event, sampling stabilization event, feedback return event, and reset recovery event. Copy the test instruction number to the test instruction number field of the event belonging to the eight types of events, and copy the test step number to the test step number field of the event belonging to the eight types of events to generate an event sequence table within the instruction. Write the target channel number, relay number, sampling channel number, load level marker, and feedback status code into the corresponding event to generate a test script semantic event template. Specifically, generating the test script semantic event template involves: Read the event template records in ascending order according to the test step number, arrange the 8 types of events under the same test step number according to the event sequence number, establish script adjacency relationship between adjacent events, and write the command issuance event to bus confirmation event, bus confirmation event to relay action event, relay action event to power output event, power output event to load echo event, load echo event to sampling stabilization event, sampling stabilization event to feedback return event, and feedback return event to reset recovery event into the causal edge number respectively to generate the test script semantic event template; Arrange the events corresponding to the current test command according to the semantic event template and event sequence number of the test script, and write the event timestamp, event duration, sampling amplitude, load response value and feedback status code to generate the current command echo test chain. Specifically, generating the current command echo test chain is as follows: Each event record in the current test instruction event sorting table is converted into a chain node. The event's associated test instruction number, test step number, event sequence number, event type, target channel number, relay number, sampling channel number, load level marker, event timestamp, event duration, sampling amplitude, load response value, and feedback status code are written into the chain node. Adjacent chain nodes are connected in ascending order of event sequence number. When the test step number increments, the reset / recovery event of the previous test step is connected to the instruction issuance event of the next test step, generating the current instruction echo test chain.

[0026] In this embodiment, the generation of the instruction echo shadow chain includes: Read the test instruction number, target channel number, relay number, load level mark, and event sequence number from the historical normal test segments and the current instruction echo test chain; Candidate historical test segments are generated by selecting historical test segments from the historical normal test segments that have the same test command number, target channel number, relay number, and load level marker as the current command echo test chain. Specifically, the selection of historical test segments from the historical normal test segments that have the same test command number, target channel number, relay number, and load level marker as the current command echo test chain involves: Read the segment number, test command number, target channel number, relay number, load level mark and event number from the historical normal test segment table. Write the segments in the historical normal test segments whose test command number matches the test command number in the current chain matching key into the first filter table. Write the segments in the first filter table whose target channel number, relay number and load level mark match the current chain matching key into the second filter table. Write the segments in the second filter table whose event number set covers the current chain event number set into the candidate historical test segments. According to the semantic event template of the test script, the corresponding events are extracted from the candidate historical test segments to generate a historical event chain, where: Extracting corresponding events from candidate historical test segments according to the semantic event template of the test script, specifically: Read the event sequence number, event type, target channel number, relay number, sampling channel number, load level mark and feedback status code from the test script semantic event template. Read the historical event records from the candidate historical test segments. Using the segment number and event sequence number as the matching key, extract the historical event records in the candidate historical test segments that match the event sequence number of the test script semantic event template as the corresponding events, and write the event type and template fields to the corresponding events. Generate a historical event chain, specifically as follows: Group the corresponding events according to the segment number, arrange the corresponding events in ascending order of event sequence number within each segment number, establish chain edge connection between events corresponding to adjacent event sequence numbers, and write the test instruction number, target channel number, relay number, load level mark, event sequence number, event timestamp, event duration, sampling amplitude, load response value and feedback status code into the chain node to generate the historical event chain corresponding to each candidate historical test segment; The historical event chain is aligned according to the event sequence number. Based on the event timestamp, event duration, sampling amplitude, load response value, and feedback status code corresponding to the same event sequence number in the candidate historical test segment, the field baseline values ​​of each event are generated, and an instruction echo shadow chain is generated, wherein: Align the historical event chain according to the event sequence number, specifically as follows: Read the fragment number and event sequence number from all historical event chains. Use the event sequence number as the alignment column identifier. Write the chain nodes with the same event sequence number in different fragment numbers into the same alignment column. Arrange the chain nodes in each alignment column in ascending order according to the fragment number to generate a historical event chain alignment table. Generate the baseline values ​​for each event's fields, specifically: Read the event timestamp, event duration, sampling amplitude, load response value, and feedback status code corresponding to the same event number in the historical event chain alignment table. Take the instruction issuance event timestamp in each candidate historical test segment as the segment start point. Subtract the segment start point from each event timestamp to obtain the event relative timestamp. Sort the event relative timestamp, event duration, sampling amplitude, and load response value under the same event number in ascending order of numerical value. When the number of samples is odd, take the middle value as the field baseline value. When the number of samples is even, take the arithmetic mean of the two middle values ​​as the field baseline value. Sort the feedback status codes under the same event number in descending order of occurrence. For feedback status codes with the same number of occurrences, sort them in ascending order of status code value. Take the first-ranked feedback status code as the feedback status code baseline value. The process of generating instruction echo shadow chains is as follows: Read the baseline values ​​of each event field in ascending order of event sequence number, establish a shadow chain node for each event sequence number, and write the test command number, target channel number, relay number, load level mark, event sequence number, event relative timestamp baseline value, event duration baseline value, sampling amplitude baseline value, load response value baseline value, and feedback status code baseline value into the shadow chain node. Connect the shadow chain nodes corresponding to adjacent event sequences in ascending order of event sequence number to generate the command echo shadow chain.

[0027] In this embodiment, generating the event alignment table and event residual data includes: Read the event type, event number, test command number, target channel number, relay number, load level mark, event timestamp, event duration, sampling amplitude, load response value, and feedback status code from the current command echo test chain and command echo shadow chain; The event character code is generated based on the event type, target channel number, relay number, and load level marker, and then modulo hashing is performed to generate the event hash code, where: The generated event character encoding is as follows: Read the event type, target channel number, relay number, and load position mark from the measured event field table and the shadow event field table. Convert the event type into an event type character, the target channel number into a channel character, the relay number into a relay character, and the load position mark into a load character. Concatenate the event type character, channel character, relay character, and load character in the order of event type character, channel character, relay character, and load character, and write separator characters between adjacent characters to generate the event character code. Perform the modulo hash operation, specifically: Read the ASCII code value corresponding to each character in the event character encoding, assign a position number to each ASCII code value according to the character arrangement order, multiply each ASCII code value by the corresponding position number and sum them to generate an encoding sum value, divide the encoding sum value by 1009 and read the remainder, use the remainder as the event hash code, and write the event hash code into the corresponding row of the actual event field table and the shadow event field table; A Smith-Waterman alignment matrix is ​​constructed. Matching scores are calculated for matrix elements with matching event hash codes. Matrix elements with inconsistent event hash codes are marked as empty elements. This completes the sparse element evaluation. The Smith-Waterman alignment matrix is ​​constructed as follows: Read the number of measured events in the measured event field table, read the number of shadow events in the shadow event field table, and establish a comparison matrix with the number of rows equal to the number of measured events plus 1 and the number of columns equal to the number of shadow events plus 1. Write 0 to all rows and columns 1, and assign the measured events in the current instruction echo measured chain to rows 2 to the last row, and assign the shadow events in the instruction echo shadow chain to columns 2 to the last column. The calculation of the matching score and the writing of the empty cell marker are as follows: Read each matrix cell in the comparison matrix in row and column order, and read the actual event hash code and shadow event hash code corresponding to the matrix cell. When the actual event hash code and the shadow event hash code are consistent, calculate the event type consistency score, target channel number consistency score, relay number consistency score, load level mark consistency score, event duration difference score, sampling amplitude difference score, load response value difference score, and feedback status code consistency score. Add the scores to generate a matching score and write it into the matrix cell score according to the Smith-Waterman recursion rule. When the actual event hash code and the shadow event hash code are inconsistent, write the matrix cell into the empty cell mark and write the matrix cell score into 0 to complete the sparse cell evaluation. Events are filtered from the semantic event template of the test script, and a central anchor event is selected. A bidirectional local comparison is then performed starting from the central anchor event, where: Filter events from the test script semantic event template and select the central anchor event, specifically: Read the event types and event numbers from the semantic event template of the test script, count the number of times each event type appears in the semantic event template of the test script, write the event types with an occurrence count of 1 into the unique event type table, read the event types from the current instruction echo test chain and instruction echo shadow chain, write the events that exist simultaneously in the unique event type table, the current instruction echo test chain and the instruction echo shadow chain into the candidate anchor event table, calculate the absolute value of the difference between the event number of the candidate anchor event and the middle sequence number of the event chain, and select the candidate anchor event with the smallest absolute value of the difference as the central anchor event; Starting with the central anchor event, a bidirectional local comparison is performed, specifically as follows: Read the measured event sequence number of the central anchor event in the current command echo measured chain, read the shadow event sequence number of the central anchor event in the command echo shadow chain, classify the measured events with a measured event sequence number less than the central anchor event as measured front-end events, classify the shadow events with a shadow event sequence number less than the central anchor event as shadow front-end events, classify the measured events with a measured event sequence number greater than the central anchor event as measured back-end events, and classify the shadow events with a shadow event sequence number greater than the central anchor event as shadow back-end events. Perform Smith-Waterman local alignment on the measured front-end events and shadow front-end events to generate a front-end alignment path, and perform Smith-Waterman local alignment on the measured back-end events and shadow back-end events to generate a back-end alignment path. Merge the front-end alignment path, the central anchor event alignment record, and the back-end alignment path in ascending order of event sequence number to generate an alignment path. Read the event sequence number difference of each event relative to the central anchor event, write the event sequence number difference into the alignment symbol, generate an event alignment table based on the alignment path, and generate event residual data according to the event timestamp difference, event duration difference, sampling amplitude difference, load response value difference, and feedback status code difference between aligned events, and write it into the causal edge number, where: The event sequence number difference is written into the comparison symbol. Specifically, each measured event and shadow event in the comparison path is read, the event sequence number of the central anchor event is read, the measured event sequence number is subtracted from the central anchor event sequence number to obtain the measured relative sequence number difference, the shadow event sequence number is subtracted from the central anchor event sequence number to obtain the shadow relative sequence number difference, the measured relative sequence number difference and the shadow relative sequence number difference are written into the comparison symbol, and the event type, event hash code, measured relative sequence number difference and shadow relative sequence number difference are concatenated in the field order to generate the relative sequence number difference code. An event alignment table is generated based on the comparison path, specifically as follows: Read the comparison path and relative sequence number difference code, build the event alignment table row by row according to the comparison path order, and write the measured event row number, shadow event row number, measured event sequence number, shadow event sequence number, event type, event hash code, measured relative sequence number difference, shadow relative sequence number difference, matrix unit score and empty unit mark into the event alignment table. When there is a measured event but no corresponding shadow event, write the measured missing alignment mark. When there is a shadow event but no corresponding measured event, write the shadow missing alignment mark. Event residual data is generated based on the differences in event timestamps, event durations, sampling amplitudes, load response values, and feedback status codes between aligned events, and then written into the causal edge number. Specifically: Read each alignment record in the event alignment table, read the corresponding measured event field and shadow event field of the alignment record, subtract the shadow event timestamp from the measured event timestamp to obtain the event timestamp difference, subtract the shadow event duration from the measured event duration to obtain the event duration difference, subtract the shadow sampling amplitude from the measured sampling amplitude to obtain the sampling amplitude difference, subtract the shadow load response value from the measured load response value to obtain the load response value difference, write records where the measured feedback status code and the shadow feedback status code are inconsistent to the feedback status code difference value 1, and write records where they are consistent to the feedback status code difference value 0, read the causal edge number between adjacent events in the semantic event template of the test script, write the causal edge number to the corresponding event residual record, and generate event residual data.

[0028] In this embodiment, generating the DeepESN state update queue includes: Read the event number, residual type, event timestamp, causal edge number, event timestamp residual, event duration residual, sampling amplitude residual, load response value residual, and feedback status code difference from the event residual data; Read the historical event residuals corresponding to the same test instruction number, target channel number, relay number, and load level mark from the historical normal test segments, and generate positive and negative residual decision boundaries. Specifically, generating the positive and negative residual decision boundaries involves: The historical event residuals are grouped according to residual type and causal edge number. The residuals of each group of historical events are arranged in ascending order of value. When the sample size is not less than 20, the value of the sorting position is the sample size multiplied by 0.95 and then rounded up as the positive decision boundary of the residual. The value of the sorting position is the sample size multiplied by 0.05 and then rounded up as the negative decision boundary of the residual. When the sample size is less than 20, the median of the historical event residuals in the same group is calculated. The absolute value of the difference between each historical event residual and the median is calculated. The median of the absolute values ​​is taken as the median absolute deviation. The median plus 3 times the median absolute deviation is taken as the positive decision boundary of the residual. The median minus 3 times the median absolute deviation is taken as the negative decision boundary of the residual. The event residual data is converted into a symbolic residual sequence, and the corresponding symbols are written according to the comparison results between the residual and the positive and negative decision boundaries of the residual. Specifically, the conversion of the event residual data into a symbolic residual sequence is as follows: Read the event number, residual type, event timestamp, causal edge number, and residual value from the residual expansion table. Read the corresponding positive and negative decision boundaries of the residual according to the residual type and causal edge number. Write the symbol 1 when the residual value is greater than the positive decision boundary of the residual, write the symbol -1 when the residual value is less than the negative decision boundary of the residual, and write the symbol 0 when the residual value is neither greater than the positive decision boundary of the residual nor less than the negative decision boundary of the residual. Sort the event number, residual type, event timestamp, causal edge number, residual value, and symbol in ascending order of event timestamp to generate a symbolic residual sequence. The symbolic residual sequence is divided according to the event timestamp order. A window residual entropy value is generated based on the proportion of corresponding symbols within each event window. Specifically, the division of the symbolic residual sequence according to the event timestamp order is as follows: Read the test instruction number and event sequence number from the symbolized residual sequence. Divide the symbolized residual records corresponding to event sequence numbers 1 to 8 within the same test instruction number into one event window. Write the event window number as the test instruction number plus the window sequence number. Count the number of symbols 1, negative 1, and 0 in each event window. Divide the number of each type of symbol by the total number of symbols in the event window to get the proportion. Multiply the proportion of each non-zero number by the logarithm of the proportion of the proportion to the base 2 and take the negative. Add the results corresponding to the three types of symbols to get the window residual entropy value. Write the result corresponding to the symbol with a proportion of 0 into 0. An entropy decision boundary is generated based on the distribution of window residual entropy values ​​corresponding to historical normal test segments. Event window numbers whose window residual entropy values ​​exceed the entropy decision boundary are written into the entropy trigger blotting bucket in the residual entropy blotting diagram. The event window number, causal edge number, and residual type are then associated and written into the residual entropy blotting diagram, where: The entropy determination boundary is generated as follows: The symbolic residual sequence corresponding to the historical normal test segment is processed according to the same event window division method to generate historical window residual entropy value. The historical window residual entropy value corresponding to the same residual type and the same causal edge number is arranged in ascending order of value. The sorting position is the value of the number of historical windows multiplied by 0.95 and rounded up as the entropy decision boundary. The entropy decision boundary is bound to the residual type and causal edge number to generate an entropy boundary table. Write the residual entropy Bloom plot as follows: Read the window residual entropy value and entropy boundary table. Write the event window number that is greater than the entropy judgment boundary into the entropy trigger window table. Multiply the number of event windows by 4 and round up to get the length of the Bloom bit array. If the length of the Bloom bit array is less than 256, take 256 as the length of the Bloom bit array. Concatenate the event window number, causal edge number, and residual type in the field order to generate the Bloom key. Read the ASCII code value of each character in the Bloom key. Sum all the ASCII code values ​​and divide by the length of the Bloom bit array and take the remainder to get the position of the first Bloom bucket. Multiply each ASCII code value by the character position number and sum them. Add 31 to the sum and divide by the length of the Bloom bit array and take the remainder to get the position of the second Bloom bucket. Write the positions of the first and second Bloom buckets to 1. Write the event window number into the entropy trigger Bloom bucket in the residual entropy Bloom sketch. Write the event window number, causal edge number, residual type, position of the first Bloom bucket, and position of the second Bloom bucket into the residual entropy Bloom sketch association table. The residual entropy Bloom sketch is as follows: The residual entropy Bloom plot consists of a residual expansion table, a symbolic residual sequence table, a windowed residual entropy table, an entropy boundary table, a Bloom bit array, a Bloom key mapping table, and an entropy trigger Bloom bucket. The generation process involves reading the event timestamp residual, event duration residual, sampling amplitude residual, load response value residual, and feedback status code difference from the event residual data. Each event residual is split into multiple residual expansion records according to residual type. Then, historical normal test segments with the same test command number, target channel number, relay number, and load level marker are read and processed according to residual type and... Causal edge numbers generate positive and negative residual decision boundaries. The current residual is converted into a symbolic residual sequence composed of 1, negative 1, and 0. Event windows are divided according to event timestamps. The proportion of the three types of symbols in each event window is counted and the window residual entropy value is calculated. Entropy decision boundaries are generated based on historical window residual entropy values. Event window numbers, causal edge numbers, and residual types that exceed the entropy decision boundaries are concatenated into Bloom keys. The keys are written into the Bloom bit array and the entropy trigger Bloom bucket through modulo hashing to form a residual entropy Bloom sketch for quickly locating abnormal residual windows. Read the event window number from the entropy trigger cloning bucket, perform forward and reverse cumulative detection on the corresponding event residuals, generate a residual cumulative decision boundary based on the residual cumulative result distribution corresponding to historical normal test segments, and write the event number, residual type, residual direction, and causal edge number exceeding the residual cumulative decision boundary into the DeepESN state update queue, where: Forward and reverse cumulative checks are performed on the corresponding event residuals, specifically as follows: Read the event window number from the entropy trigger Bloom bucket, read the corresponding causal edge number and residual type from the residual entropy Bloom sketch association table, extract the residual records with the same event window number, the same causal edge number and the same residual type from the residual expansion table, sort the residual records in ascending order by event timestamp, read the median of the corresponding historical event residuals as the residual baseline value, subtract the residual baseline value from the current residual value to get the positive increment, subtract the current residual value from the residual baseline value to get the negative increment, add the positive increment to the previous positive cumulative value to get the candidate positive cumulative value, write 0 to the positive cumulative value when the candidate positive cumulative value is less than 0, write the positive cumulative value to the candidate positive cumulative value when the candidate positive cumulative value is not less than 0, add the negative increment to the previous negative cumulative value to get the candidate negative cumulative value, write 0 to the negative cumulative value when the candidate negative cumulative value is less than 0, write the negative cumulative value to the candidate negative cumulative value when the candidate negative cumulative value is not less than 0, and generate the residual accumulation result; The event number, residual type, residual direction, and causal edge number that exceed the residual accumulation decision boundary are written to the DeepESN state update queue, specifically: Read the historical event residuals corresponding to the historical normal test segments. Generate historical forward and backward cumulative results in the same way as forward and backward cumulative detection in S56. Sort the historical forward cumulative results corresponding to the same residual type and the same causal edge number in ascending order of value. Take the position of the number of historical forward cumulative results multiplied by 0.95 and rounded up as the judgment boundary for forward residual accumulation. Sort the historical backward cumulative results corresponding to the same residual type and the same causal edge number in ascending order of value. Take the position of the number of historical backward cumulative results multiplied by 0.95 and rounded up as the judgment boundary for backward residual accumulation. When the current forward cumulative value is greater than the judgment boundary for forward residual accumulation, write the event number, residual type, forward residual direction, causal edge number, event timestamp, and forward cumulative value into the DeepESN state update queue. When the current backward cumulative value is greater than the judgment boundary for backward residual accumulation, write the event number, residual type, backward residual direction, causal edge number, event timestamp, and backward cumulative value into the DeepESN state update queue.

[0029] In this embodiment, generating the chain break identification result includes: Read the device channel topology data from the standardized test and monitoring dataset, extract the connection relationships between communication interfaces, control loops, relay nodes, power output nodes, test channels, load nodes, sampling loops, and feedback message nodes, and generate a physical topology mask matrix. Specifically, generating the physical topology mask matrix involves: Read the device channel topology table from the standardized test and monitoring dataset. The device channel topology table contains node number, node type, connection start number, connection end number, connection direction, channel number, and interface number. Write the records of node types such as communication interface, control loop, relay node, power output node, test channel, load node, sampling loop, and feedback message node into the topology node table. Sort the topology node table in ascending order of node number and assign matrix row and column numbers to each node. Create an initial matrix with the same number of rows and columns as the number of topology nodes. Write 0 to all cells in the initial matrix. Read each connection relationship and locate the row number corresponding to the connection start number and the column number corresponding to the connection end number as matrix cells. Write 1 to the matrix cells. Write 1 to the reverse matrix cells corresponding to bidirectional connection relationships. Generate a physical topology mask matrix. An improved DeepESN model is constructed, comprising a topological perforated reservoir, a single-step reverse-time echo excitation unit, an instantaneous mutual exclusion freezing unit, and a chain-breaking readout layer. Specifically, the improved DeepESN model is constructed as follows: This invention constructs an improved DeepESN model based on the original DeepESN input mapping, multi-layer reservoir state recursion, and linear readout backbone structure. It uses the DeepESN state update queue as the event trigger input, the event alignment table and event residual data as the chain break residual input, and the physical topology mask matrix as the reservoir structure constraint input. The improved DeepESN model is formed by sequentially connecting a topology perforated reservoir, a single-step reverse-time echo excitation unit, an instantaneous mutual exclusion freeze unit, and a chain break readout layer. A topology perforated reservoir is inserted before the reservoir state recursion, and nodes in the reservoir connection matrix that do not belong to the communication interface, control loop, or relay node are connected. The positions of the actual connection relationships of the power output node, test channel, load node, sampling loop and feedback message node are written into the empty connection mark. After the recursive output of the reservoir state, a single-step reverse time echo excitation unit is embedded. The transpose matrix of the reservoir connection matrix is ​​read to perform one reverse echo calculation on the current reservoir state and splice it with the forward reservoir state to generate an echo splicing state. When the missing event mark is triggered, an instantaneous mutual exclusion freeze unit is embedded to perform sign inversion and one event step freeze on the reservoir node corresponding to the missing event, forming an improved DeepESN model that includes topology constraint propagation, reverse time echo compensation, missing event pulse holding and disconnection field reading. The improved DeepESN model was trained using the mean squared error between the chain break score output by the chain break readout layer and the true chain break label as the loss function. The weights and chain break readout biases of the chain break readout layer were trained. The topological perforation reservoir connection matrix was fixed according to the physical topological mask matrix. The single-step reverse time echo excitation unit was fixed according to the transpose of the reservoir connection matrix. The instantaneous mutual exclusion freeze unit was fixed according to the missing event label. When the rate of change of the loss function in 10 consecutive training rounds was not greater than 0.001, the improved DeepESN model was determined to have reached the convergence state and training was stopped. Read historical normal test segments, labeled broken chain samples, event alignment table, event residual data, DeepESN state update queue, and physical topology mask matrix. Write the true broken chain label of the causal edge corresponding to the historical normal test segment to 0, write the true broken chain label of the causal edge manually labeled as broken chain in the labeled broken chain samples to 1, write the reserve pool connection matrix unit corresponding to the position with a value of 0 in the physical topology mask matrix to 0, write the connection weight corresponding to the position with a value of 1 in the physical topology mask matrix to the reserve pool connection matrix unit, and take 1 and divide it by the number of matrix units with a value of 1 in the corresponding row. Input the DeepESN state update queue, event alignment table, and event residual data into the improved DeepESN model to obtain the broken chain readout vector corresponding to each causal edge. The broken link readout layer reads the broken link readout vector. The initial values ​​of the broken link readout weights and the broken link readout bias are all set to 0. The values ​​of each dimension in each broken link readout vector are multiplied by the corresponding broken link readout weights and summed. The broken link readout bias is then added to obtain the broken link score of the corresponding causal edge. The numerical difference between the broken link score and the true broken link label is calculated. The numerical differences are squared and summed. Then, the sum is divided by the number of causal edge samples involved in the calculation to generate the mean squared error loss function value. Adjust the chain break readout weight and chain break readout bias based on the mean squared error loss function value. Calculate the change in the mean squared error loss function value before and after adjusting the chain break readout weight and the chain break readout bias. When the mean squared error loss function value decreases after parameter adjustment, retain the current parameters. When the mean squared error loss function value increases after parameter adjustment, halve the parameter adjustment range and recalculate the mean squared error loss function value. Generate the optimization direction corresponding to the chain break readout weight and chain break readout bias. The broken link readout weights and biases are updated along the optimization direction with a learning rate of 0.001. After each training round, the mean squared error loss function value is recalculated, and the absolute value of the difference between the current round's mean squared error loss function value and the previous round's mean squared error loss function value is calculated. The absolute value of the difference is then divided by the previous round's mean squared error loss function value to generate the loss function change rate. When the loss function change rate in 10 consecutive training rounds is not greater than 0.001, training is stopped, and the broken link readout weights, broken link readout biases, physical topology mask matrix, reservoir connection matrix, and broken link score decision boundary are written into the improved DeepESN model parameter table. The broken link score decision boundary is the 0.95th quantile value of the broken link score corresponding to the historical normal test segment, sorted in ascending order of numerical values. The topology perforation reservoir reads the physical topology mask matrix, writes empty connection markers to positions in the reservoir connection matrix that do not belong to the connection relationship, and writes the DeepESN state update queue, event alignment table, and event residual data to the corresponding reservoir node according to the event number and causal edge number, generating the topology perforation reservoir state, where: Topological perforated reservoirs include: Topology mask reader: Stores the physical topology mask matrix, topology node number, node type, connection start number, connection end number, causal edge number, and abnormal channel number, and generates a topology node index table; The reservoir connection matrix puncher stores the reservoir connection matrix, empty connection markers, and connection weights. It writes 0 to the positions with values ​​of 0 in the physical topology mask matrix and writes the connection weights to the positions with values ​​of 1 in the physical topology mask matrix. The connection weight is 1 divided by the number of matrix cells with values ​​of 1 in the corresponding row. Event State Mapper: Stores event number, causal edge number, residual type, residual direction, event timestamp and cumulative value, and maps the DeepESN state update queue to the corresponding reserve pool node according to the causal edge number; Residual channel writer: Stores event timestamp residuals, event duration residuals, sampling amplitude residuals, load response value residuals, and feedback status code differences, and generates timestamp residual channels, duration residual channels, sampling amplitude residual channels, load response residual channels, and status code residual channels; Alignment state writer: Stores matrix cell scores, empty cell markers, measured missing alignment markers, and shadow missing alignment markers in the event alignment table, and generates alignment state channels; The reservoir state recursor stores the input vector, the reservoir state vector with the previous event timestamp, the reservoir connection matrix, and the candidate state vector. It performs a hyperbolic tangent transformation on the sum of the input vector and the reservoir connection matrix multiplied by the reservoir state vector with the previous event timestamp to generate the topological perforated reservoir state. Node state cache: Stores the state of the reserve pool nodes, node number, causal edge number, and abnormal channel number for each event timestamp, and forms a reserve pool state cache table in ascending order of event timestamp; Topology State Outputter: Stores the topology perforation reserve pool status, node status cache table, causal edge number and abnormal channel number, and outputs the topology perforation reserve pool status to the single-step reverse time echo excitation unit; The state of the topological perforated reservoir is generated as follows: Establish a reservoir connection matrix, where the number of rows and columns equals the number of rows in the physical topology mask matrix. Read the physical topology mask matrix cell corresponding to each matrix cell in the reservoir connection matrix. When the physical topology mask matrix cell is 0, write 0 to the corresponding cell in the reservoir connection matrix and mark it as an empty connection. When the physical topology mask matrix cell is 1, write the connection weight to the corresponding cell in the reservoir connection matrix. The connection weight is 1 divided by the number of cells with a value of 1 in the corresponding row of the physical topology mask matrix. Read the event number, residual type, residual direction, causal edge number, event timestamp, and cumulative value from the DeepESN state update queue, and sort them according to the causal edge number. Retrieve the corresponding start and end nodes of the connection in the physical topology mask matrix, write the cumulative value into the input vector channel corresponding to the end node, write the matrix unit score and missing event marker in the event alignment table into the alignment state channel of the same node, and write the event timestamp residual, event duration residual, sampling amplitude residual, load response value residual and feedback status code difference in the event residual data into the residual state channel of the same node. For each event timestamp, calculate the candidate state by adding the input vector to the reservoir connection matrix and multiplying it by the sum of the reservoir state vector of the previous event timestamp. Then, perform hyperbolic tangent transformation on the candidate state element by element to generate the topology perforation reservoir state. The single-step reverse-time echo excitation unit reads the state of the topologically perforated reservoir, performs a reverse-time echo calculation based on the transpose of the reservoir connection matrix, generates a reverse-time echo state, and concatenates the topologically perforated reservoir state with the reverse-time echo state to generate an echo concatenation state, wherein: The single-step reverse-time echo excitation unit includes: Transposed area of ​​reservoir connection matrix: Stores the reservoir connection matrix output by the topology perforated reservoir. The reservoir connection matrix is ​​generated by swapping rows and columns. The transposed matrix is ​​used to represent the reverse echo propagation relationship from the feedback message node, sampling loop node, load node to the power output node, relay node and control loop node. Current Reserve Pool Status Input Area: Stores the topology perforated reserve pool status corresponding to the current event timestamp. The topology perforated reserve pool status includes the node status vector, node number, causal edge number, and abnormal channel number of each reserve pool node. Reverse-time propagation vector calculation area: Stores the transpose matrix and the current reservoir state. Multiply the transpose matrix with the current reservoir state vector to generate the reverse-time propagation vector. The reverse-time propagation vector is used to express the state response after feedback delay, sampling stabilization delay and load response hysteresis are propagated in reverse along the test chain. Residual compensation input area: Stores the event timestamp residual, event duration residual, sampling amplitude residual, load response value residual, and feedback status code difference in the event residual data. Write the five types of residuals into the residual status channel of the corresponding node according to the event number and causal edge number. Reverse-time candidate state region: stores the reverse-time propagation vector and the residual state channel vector. The reverse-time propagation vector and the residual state channel vector are added according to the corresponding nodes to generate the reverse-time candidate vector; Reverse-time echo activation region: Stores reverse-time candidate vectors. Performs hyperbolic tangent transform on each element of the reverse-time candidate vectors to generate reverse-time echo states. The output range of the hyperbolic tangent transform is from -1 to 1. The echo splicing state is generated as follows: Read the state of the topological perforated reservoir and the reservoir connection matrix. Transpose the reservoir connection matrix by row and column to generate a transpose matrix. Multiply the transpose matrix by the topological perforated reservoir state vector of the current event timestamp to obtain the reverse propagation vector. Add the reverse propagation vector to the residual state channel vector of the current event timestamp to generate a reverse candidate vector. Perform hyperbolic tangent transformation on the reverse candidate vector element by element to generate a reverse echo state. Arrange the node state vectors in the topological perforated reservoir state first and the node state vectors in the reverse echo state second. Concatenate them in ascending order of node number to form an echo concatenation state. The instantaneous mutual exclusion freeze unit reads the missing event markers from the event alignment table and the trigger event numbers from the DeepESN state update queue, performs sign inversion and one event step freeze on the reserve pool node corresponding to the missing event marker, and generates a freeze pulse state, where: The instantaneous mutual exclusion freeze unit includes: Missing event marker input area: stores the measured missing alignment marker, shadow missing alignment marker, event number, event sequence number, event type, and matrix cell score in the event alignment table; Triggered event matching area: Match the missing event marker with the corresponding event number and the triggered event number one by one. Records that match are written to the frozen candidate table. Reserve pool node location area: Map the causal edge numbers to the corresponding reserve pool nodes, and read the node state vector of the reserve pool node in the topological perforated reserve pool state; Mutually exclusive freeze selection area: Stores freeze candidate records that point to the same reserve pool node within the same event step. The freeze candidate records are sorted in descending order of the chain break score. Only the freeze candidate record ranked first in the sorting is retained for the same reserve pool node. If there are records with the same chain break score in the first sorting position, the freeze candidate record with the smaller event sequence number is retained. A mutually exclusive freeze table is generated. Sign Inverted State Area: Stores the reserve pool node number in the mutual exclusion freeze table, the current node state vector, and the sign inverted state. Multiply each element of the current node state vector by -1 to generate the sign inverted state. Event Step Freeze Area: Stores the event step number, freeze start event number, freeze end event number, and freeze status flag. The event step is defined as the event number interval of 1 between two adjacent event records. Within the current event step, the sign is kept inverted, and the corresponding reserve pool node is prohibited from participating in the recursive update of the reserve pool connection matrix. Freeze-off area: Stores the input vector for the next event step, the freeze-off flag, and the recovery update state. When the event sequence number enters the next event step, it reads the new input vector, writes the freeze-off flag to 1, and restores the topology perforated reserve pool state update of the corresponding reserve pool node. The freeze pulse state is generated as follows: Read the missing event markers in the event alignment table, which include measured missing alignment markers and shadow missing alignment markers. Read the trigger event number in the DeepESN state update queue. Match the event number corresponding to the missing event marker with the trigger event number. If the match is consistent, read the corresponding causal edge number and locate the reserve pool node. Read the topology perforated reserve pool state of the corresponding reserve pool node at the current event timestamp. Multiply each element in the node state vector by -1 to generate a sign-inverted state. Define the event step size as the event sequence number interval between two adjacent event records as 1. Keep the sign-inverted state in the current event step size and do not participate in the reserve pool connection matrix update. In the next event step size, read the new input vector and resume the reserve pool update to generate a frozen pulse state. The chain break readout layer reads the echo splicing state and the frozen pulse state, and reads the causal edge number and abnormal channel number from the physical topology mask matrix and the DeepESN state update queue to generate a chain break identification result. The chain break identification result includes the chain break edge set, chain break event number, abnormal channel number, and initial marker of the abnormal source. Specifically, generating the chain break identification result is as follows: The system reads the echo splicing state, frozen pulse state, causal edge numbers in the physical topology mask matrix, abnormal channel numbers and event timestamps in the DeepESN state update queue. It then splices the echo splicing state and frozen pulse state in ascending order of node numbers to form a broken link readout vector. The broken link readout layer reads the readout training vectors from historical normal test segments and labeled broken link samples. The broken link readout weights are calculated using ridge regression with a regularization coefficient of 0.001. The product of the broken link readout vector and the broken link readout weights is used as the broken link score. When the broken link score is greater than the 0.95th percentile of the broken link score in historical normal test segments, the corresponding causal edge number is written into the broken link edge set, the corresponding event number is written into the broken link event number, the channel number associated with the causal edge number is written into the abnormal channel number, and the initial abnormal source marker is written into the retest marker, generating the broken link identification result.

[0030] In this embodiment, generating the anomaly source marker includes: Read the chain break identification results, and extract the abnormal channel numbers corresponding to the chain break edges pointing to relay nodes, power output nodes, test channels, load nodes, and sampling circuits from the chain break edge set as suspected abnormal channels. Specifically, the extraction of the abnormal channel numbers corresponding to the chain break edges pointing to relay nodes, power output nodes, test channels, load nodes, and sampling circuits from the chain break edge set as suspected abnormal channels is as follows: Read the broken link node type table, and write the broken links that belong to relay nodes, power output nodes, test channels, load nodes and sampling circuits of the connection start node type or connection end node type into the reproducible broken link table. Read the abnormal channel number, target channel number and sampling channel number from the reproducible broken link table, use the abnormal channel number as the main field, use the target channel number and sampling channel number as the correlation fields, delete the records with duplicate abnormal channel numbers, keep the record with the highest broken link score, and generate the suspected abnormal channel. Read the standard load configuration data, bypass switching status, safety interlock status, and test task pause flag corresponding to the suspected abnormal channel. Write the suspected abnormal channels that meet the following criteria into the bypass micro-retest channel set: (Specific details are missing from the original text.) Read the abnormal channel number in the suspected abnormal channel, read the corresponding bypass switching status, safety interlock status and test task pause mark. When the bypass switching status is valid, the access interlock status is allowed, the grounding interlock status is allowed, the overcurrent protection status is not triggered, and the test task pause mark is valid, write the abnormal channel number, standard load number, standard load resistance value, standard load rated voltage, standard load rated current, retest duration and chain break edge number into the bypass micro retest channel set. Perform standard load bypass micro-retest on suspected abnormal channels in the bypass micro-retest channel set, generating a bypass command echo chain, where: Perform standard load bypass micro-retest on suspected abnormal channels in the bypass micro-retest channel set, specifically as follows: Read the bypass micro-retest channel set, select the channel to be retested in ascending order of the abnormal channel number, read the standard load number and bypass switch number corresponding to the channel to be retested, send a bypass switching command to the bypass switch number, read the bypass switching feedback status code, connect the standard load corresponding to the standard load number when the bypass switching feedback status code is successful, send a micro-retest command consistent with the original test command number to the channel to be retested, and collect the command issuance time, bus confirmation time, relay action time, power output start time, load response start time, sampling stabilization time, feedback message return time, and reset recovery completion time; Generate a bypass instruction echo chain, specifically as follows: Read the command issuance time, bus confirmation time, relay action time, power output start time, load response start time, sampling stabilization time, feedback message return time, and reset recovery completion time collected by the standard load bypass micro-retest. Write the event sequence numbers 1 to 8 in the order of command issuance event, bus confirmation event, relay action event, power output event, load echo event, sampling stabilization event, feedback return event, and reset recovery event. Write the abnormal channel number, standard load number, bypass switch number, sampling amplitude, load response value, and feedback status code into the corresponding chain node. Connect the chain nodes corresponding to adjacent event sequence numbers in the ascending direction of the event sequence number to generate a bypass command echo chain. The bypass instruction echo chain is compared with the instruction echo shadow chain execution event chain to generate bypass micro-retest results, where: The bypass instruction echo chain is compared with the instruction echo shadow chain execution event chain, specifically as follows: Read the event type, event number, target channel number, relay number, load level marker, event timestamp, event duration, sampling amplitude, load response value, and feedback status code from the bypass command echo chain and command echo shadow chain. Generate bypass event character codes according to the event type, target channel number, relay number, and load level marker. Perform modulo hash processing on the bypass event character codes to generate bypass event hash codes. Establish a bypass Smith-Waterman comparison matrix. Calculate the matching score for matrix cells with matching bypass event hash codes. Write empty cell markers for matrix cells with inconsistent bypass event hash codes. Generate a bypass comparison path. Generate bypass micro-retest results, specifically: Read the alignment events in the bypass comparison path, calculate the difference between the bypass command echo chain event timestamp and the command echo shadow chain event timestamp, calculate the difference between the bypass event duration and the shadow chain event duration baseline value, calculate the difference between the bypass sampling amplitude and the shadow chain sampling amplitude baseline value, calculate the difference between the bypass load response value and the shadow chain load response value baseline value, write records where the bypass feedback status code and the shadow chain feedback status code baseline value are inconsistent into the feedback status code difference value 1, write records where they are consistent into the feedback status code difference value 0, write the causal edge number where the difference exceeds the corresponding residual positive decision boundary or is lower than the corresponding residual reverse decision boundary into the bypass broken chain edge set, and write the bypass broken chain edge set, bypass event residual data, bypass feedback status code, and bypass retest time into the bypass micro retest result; When the bypass micro-retest result contains a broken link edge that matches the broken link edge set, the initial anomaly source marker is corrected to an internal fault marker of the electrical integrated control test equipment. When the bypass micro-retest result does not contain a broken link edge that matches the broken link edge set, the initial anomaly source marker is corrected to a non-equipment internal anomaly marker, and an anomaly source marker is generated, wherein: The initial flag for the anomaly source has been corrected to a non-device internal anomaly flag, specifically: Read the broken chain edge set and the initial mark of the abnormal source in the broken chain identification result, read the bypass broken chain edge set in the bypass micro retest result, compare the broken chain edge number in the broken chain edge set with the bypass broken chain edge number in the bypass broken chain edge set one by one, if there is a matching broken chain edge number, correct the initial mark of the abnormal source to the internal fault mark of the electrical integrated control test equipment, if there is no matching broken chain edge number, correct the initial mark of the abnormal source to the non-equipment internal abnormal mark; Generate an exception source marker, specifically: Read the corrected initial anomaly source marker, chain break event number, anomaly channel number, chain break edge number, and bypass micro-retest results. Write the corresponding records of the internal fault markers of the electrical integrated control test equipment into the internal fault source table, and write the corresponding records of the non-equipment internal anomaly markers into the non-equipment internal anomaly source table. Arrange the chain break event number, anomaly channel number, chain break edge number, bypass chain break edge set, bypass event residual data, and corrected source markers in field order to generate anomaly source markers.

[0031] In this embodiment, generating and outputting the monitoring results of the electrical integrated control test equipment includes: Read the broken chain edge set, broken chain event number and abnormal channel number from the broken chain identification results; read the abnormal source marker, bypass micro-retest results and historical broken chain recursive index rope in the monitoring record library. Establish the top-level node of the recursive index rope of the broken chain according to the broken chain edge number in the broken chain edge set, and write the broken chain edge number, broken chain event number, abnormal channel number, abnormal source marker, and bypass micro-retest result to the top-level node, where: Establish the top-level rope node of the broken-chain recursive index rope, specifically as follows: Read the broken chain edge number from the broken chain index input table, use the broken chain edge number as the primary key of the top-level rope node, and create the top-level rope node in ascending order of the broken chain edge number. The top-level rope node field includes the top-level rope node number, broken chain edge number, connection start node number, connection end node number, causal edge number, abnormal channel number, node creation time and cumulative number of broken chains. The top-level rope node number is generated by concatenating the broken chain edge number and the node creation time. Write a field to the top-level rope node, specifically: Read the chain break event number, abnormal channel number, abnormal source marker and bypass micro-retest result corresponding to the same chain break edge number, write the chain break event number into the current chain break event field of the top rope node, write the abnormal channel number into the channel field of the top rope node, write the source marker in the abnormal source marker into the source field of the top rope node, write the bypass retest time, bypass feedback status code, bypass event residual data and bypass chain break edge set in the bypass micro-retest result into the retest field of the top rope node, and generate the current chain break recursive index rope; Retrieve the top-level rope node with the same chain break edge number from the historical chain break recursive index rope. Embed the current chain break event number, anomaly channel number, anomaly source marker, and bypass micro-retest result into the child rope node of the consistent top-level rope node, where: Retrieve the top-level rope node with the same number as the broken edge in the historical broken chain recursive index rope, specifically: Read the top-level rope node table in the historical broken chain recursive index rope, build a historical top-level rope node index according to the broken chain edge number, use the broken chain edge number in the current broken chain recursive index rope as the search key, search for the top-level rope node with the same broken chain edge number in the historical top-level rope node index, when a matching top-level rope node is found, read the matching top-level rope node number, when no matching top-level rope node is found, use the top-level rope node as the current top-level rope node. Embedding sub-rope nodes, specifically: Read the current chain break event number, abnormal channel number, abnormal source marker, and bypass micro-retest result. Write the consistent top-level rope node number into the parent node number field of the child rope node. Write the current chain break event number into the event field of the child rope node. Write the abnormal channel number into the channel field of the child rope node. Write the abnormal source marker into the source field of the child rope node. Write the bypass micro-retest result into the retest field of the child rope node. Generate the child rope node number by concatenating the chain break edge number, chain break event number, and bypass retest time. Append the child rope node to the end of the child rope node chain of the consistent top-level rope node. Generate the number of chain break recurrences, the first chain break event number, the most recent chain break event number, and the chain break recurrence event sequence according to the hierarchical relationship between the top-level rope node and the sub-rope nodes; A chain break tracing index is generated based on the chain break edge number, chain break event number, abnormal channel number, and chain break recurrence event sequence. Monitoring results for the electrical integrated control testing equipment are then generated according to the chain break identification results, abnormal source markers, bypass micro-retest results, chain break recursion index, and chain break tracing index. Specifically, the generation of the chain break tracing index based on the chain break edge number, chain break event number, abnormal channel number, and chain break recurrence event sequence is as follows: Read the broken chain edge number, broken chain event number, abnormal channel number, and broken chain recurrence event sequence. Use the broken chain edge number as the first-level index field, the abnormal channel number as the second-level index field, the broken chain event number as the event location field, and the broken chain recurrence event sequence as the recurrence location field. Concatenate the broken chain edge number, abnormal channel number, broken chain event number, and broken chain recurrence event sequence in the field order to generate a traceability index key. Write the traceability index key into the historical broken chain traceability index table to generate the broken chain traceability index.

[0032] Example 1: In a continuous commissioning and monitoring scenario for an electrical integrated control test equipment, the test object includes one set of electrical integrated control test equipment and one set of control loops under test. Edge computing nodes are deployed beside the test cabinet. The continuous test cycle lasts 6 hours. The test script contains 86 test instructions, which are executed repeatedly 30 times, resulting in 2580 test instruction records. The test tasks cover communication handshake, relay drive, power output, load access, sampling feedback, and reset recovery, involving 18 target channels, 32 relay nodes, 24 sampling channels, and 12 standard loads with resistance values ​​ranging from 5Ω to 200Ω. The system receives 41280 communication messages, 20640 relay action records, 15480 power output records, 928800 sampling channel records, 185760 load response records, 2580 reset recovery records, and 146 equipment channel topology connection records. Manual verification revealed 79 anomalies, including 21 instances of delayed relay operation, 17 instances of intermittent sampling circuit operation, 14 instances of loose load wiring, 9 instances of no response from the tested object, and 18 instances of communication feedback delay.

[0033] After the data enters the processing flow, the edge computing nodes perform timestamp calibration, field unification, duplicate message deletion, missing event marking, unit unification, and event number association. Before processing, the communication message duplication rate was 2.8%, the relay action timestamp offset range was 0–42ms, and there were 1836 records of mixed units in the sampling channels. After processing, duplicate communication messages were reduced to 0, the relay action timestamp offset was controlled within 0–3ms, and the sampling channel units were unified to V, A, and ms. Using one round of scripts as the current monitoring batch, the system breaks down each test command into command issuance events, bus confirmation events, relay action events, power output events, load echo events, sampling stabilization events, feedback return events, and reset recovery events. 86 test commands generate 688 event template nodes, and write the event timestamp, event duration, sampling amplitude, load response value, and feedback status code to form the current command echo measurement chain.

[0034] The system filters historical normal test segments that share the same test command number, target channel number, relay number, and load level marker. The training samples include 4200 historical normal test segments, 620 broken chain marked samples, and 310 non-equipment internal anomaly samples. Under each event number, the system reads the event's relative timestamp, event duration, sampling amplitude, load response value, and feedback status code from the historical segment, using the median to generate a baseline value for the field, forming a command echo shadow chain. In test command T047, the baseline value for the duration of the shadow chain relay action event is 23ms, the measured event duration is 71ms, and the duration residual is 48ms. The load response value of the shadow chain load echo event is 2.31A, the measured value is 1.42A, and the load response value residual is -0.89A.

[0035] During the event chain comparison phase, the system generates event character codes based on event type, target channel number, relay number, and load level marker, and obtains event hash codes through modulo hashing. The Smith-Waterman comparison matrix only calculates matching scores for matrix units with matching hash codes; matrix units with inconsistent hash codes are marked as empty units. The theoretical matrix units for 688 measured events and 688 shadow events total 473,344. After sparse unit evaluation, 84,672 matrix units actually participate in the matching calculation, reducing edge computation time from 238ms for full matrix comparison to 61ms. The system selects stable sampled events as central anchor events, performs bidirectional local comparisons in both the front and back segments, and writes relative sequence difference codes, generating an event alignment table of 688 rows and 3,440 event residual data entries.

[0036] Event residual data is converted into a residual entropy Bloom plot. The system reads historical event residuals from 4200 historical normal test segments, generating positive and negative decision boundaries for residuals of the same residual type and causal edge number. A sign 1 is written when the current residual exceeds the positive decision boundary, a sign -1 is written when it falls below the negative decision boundary, and a sign 0 is written when it lies between the two boundaries. Every 8 consecutive events are divided into one event window, and one round of scripting creates 86 event windows. The average residual entropy value for normal windows is 0.31, and the average residual entropy value for abnormal windows is 1.12. The entropy decision boundary is generated based on the 0.95th quantile of the historical window residual entropy value, with a value of 0.76. In the current test batch, 24 event windows exceeded the entropy decision boundary and were written to the entropy-triggered Bloom bucket. The Bloom bit array length is 344 bits, and 127 state update records are written to the DeepESN state update queue.

[0037] The DeepESN model is improved by reading the DeepESN state update queue, event alignment table, event residual data, and physical topology mask matrix. The physical topology mask matrix is ​​generated from 146 device channel topology connection records, with a matrix size of 94×94. The topology perforation reservoir writes physically unreachable connections to 0, retaining real connections. The single-step reverse-time echo excitation unit reads the transpose matrix of the reservoir connection matrix to supplement feedback delay and sampling stabilization delay information. The instantaneous mutual exclusion freeze unit performs a one-event-step freeze on missing event markers. The broken-chain readout layer uses 620 broken-chain labeled samples and 4200 historical normal samples to generate readout weights, with a ridge regression regularization coefficient of 0.001. In the current test batch, the model outputs 63 broken-chain edges, 63 broken-chain event numbers, and 38 abnormal channel numbers. The initial markers for abnormal sources are all written into the retest markers.

[0038] The system performed standard load bypass micro-retests on 38 suspected abnormal channels. Channels with valid bypass switching, permitted safety interlocks, and valid test task pause markers were included in the bypass micro-retest channel set. Ultimately, 34 channels were included in the retest, while the remaining 4 channels were not included due to invalid bypass switching or prohibited safety interlocks. The average time for each bypass micro-retest was 3 seconds. After comparing the bypass command echo chain with the command echo shadow chain again, 21 channels still had the same broken chain edge and were corrected to internal fault markers of the electrical integrated control test equipment. The broken chain edge disappeared in 13 channels and was corrected to non-internal equipment anomaly markers. In the current test batch, 11 broken chain edges appeared in the historical record, with a broken chain recurrence rate ranging from 2 to 5 times. The system generated 63 broken chain trace indexes.

[0039] The comparative experiment used the same training and test samples. The test samples consisted of 30 rounds of script test data, totaling 2580 test commands, with 79 anomalies manually reviewed. The traditional threshold method correctly identified 61 anomalies, with 18 missed detections, 22 false positives, and a total of 83 alarms. The accuracy rate for broken link edge location was 68.3%, the accuracy rate for anomaly source determination was 64.6%, and the average computation time at the edge was 486ms. The traditional single DeepESN method correctly identified 68 anomalies, with 11 missed detections, 15 false positives, and a total of 83 alarms. The accuracy rate for broken link edge location was 78.9%, the accuracy rate for anomaly source determination was 72.2%, and the average computation time at the edge was 314ms. The method of this invention correctly identified 76 anomalies, with 3 missed detections, 6 false positives, and a total of 82 alarms. The accuracy rate for broken link edge location was 91.4%, the accuracy rate for anomaly source determination was 88.6%, the average computation time for edge broken link identification was 128ms, and the average computation time for recurring broken link retrieval was 12ms.

[0040] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A monitoring method for electrical integrated control testing equipment based on edge computing, characterized in that, include: Collect raw test and monitoring data and preprocess it to generate a standardized test and monitoring dataset; Based on a standardized test monitoring dataset, a semantic event template for the test script is generated, and the current instruction echo test chain is generated according to the order of the test events. Filter historical event chains from historical normal test segments to generate instruction echo shadow chains; The Smith-Waterman algorithm is used to perform event chain comparison, sparse unit evaluation, bidirectional expansion of central anchor events, and relative sequence difference encoding on the current command echo measured chain and command echo shadow chain to generate an event alignment table and event residual data. Based on the event residual data, construct a residual entropy Bloom plot, filter the entropy trigger event window, and generate a DeepESN state update queue; A physical topology mask matrix is ​​generated based on a standardized test and monitoring dataset. The DeepESN state update queue, event alignment table, event residual data and physical topology mask matrix are input into an improved DeepESN model containing a topology perforation reservoir, a single-step inverse time echo excitation unit and an instantaneous mutual exclusion freeze unit to generate chain break identification results. Based on the set of broken links, suspected abnormal channels are extracted. Standard load bypass micro-retest is performed on the suspected abnormal channels to generate bypass micro-retest results. The initial abnormal source marker is corrected and an abnormal source marker is generated. Based on the chain break identification results and bypass micro-retest results, a chain break recursive index rope is constructed. The recurrence records of the same chain break edge are embedded into the corresponding rope nodes and a chain break tracing index is generated. The monitoring results of the electrical integrated control test equipment are generated and output according to the chain break identification results, abnormal source markers, bypass micro-retest results and chain break tracing index.

2. The monitoring method for electrical integrated control testing equipment based on edge computing according to claim 1, characterized in that, The original test monitoring data includes test scripts, test instructions, communication messages, device channel topology data, relay action data, power output data, sampling channel data, load response data, reset and recovery data, historical normal test segments, standard load configuration data, bypass switching status, safety interlock status, test task pause markers, monitoring record library, and historical chain break recursive index.

3. The monitoring method for electrical integrated control testing equipment based on edge computing according to claim 1, characterized in that, The preprocessing includes timestamp calibration, field unification, duplicate message deletion, missing event marking, unit unification, and event number association.

4. The monitoring method for electrical integrated control testing equipment based on edge computing according to claim 1, characterized in that, The process of generating a test script semantic event template based on a standardized test monitoring dataset and generating the current instruction echo test chain according to the order of test events includes: Read the standardized test monitoring dataset and extract the test instruction number, test step number, target channel number, relay number, sampling channel number, load level mark, feedback status code and event timestamp; According to the test steps in the test script, each test instruction is broken down into instruction issuance event, bus confirmation event, relay action event, power output event, load response event, sampling stabilization event, feedback return event, and reset recovery event, and the event sequence number and the test instruction number to which the event belongs are written. Write the target channel number, relay number, sampling channel number, load level mark, and feedback status code into the corresponding event to generate a test script semantic event template; Arrange the events corresponding to the current test command according to the semantic event template and event sequence number of the test script, and write the event timestamp, event duration, sampling amplitude, load response value and feedback status code to generate the current command echo test chain.

5. The monitoring method for electrical integrated control testing equipment based on edge computing according to claim 1, characterized in that, The generation instruction echo shadow chain includes: Read the test instruction number, target channel number, relay number, load level mark, and event sequence number from the historical normal test segments and the current instruction echo test chain; From historical normal test segments, select those that have the same test command number, target channel number, relay number, and load level mark as the current command echo test chain, and generate candidate historical test segments; Extract corresponding events from candidate historical test segments according to the semantic event template of the test script, and generate a historical event chain; The historical event chain is aligned according to the event sequence number. Based on the event timestamp, event duration, sampling amplitude, load response value and feedback status code corresponding to the same event sequence number in the candidate historical test segment, the field baseline value of each event is generated, and the instruction echo shadow chain is generated.

6. The monitoring method for electrical integrated control testing equipment based on edge computing according to claim 1, characterized in that, The generation of the event alignment table and event residual data includes: Read the event type, event number, test command number, target channel number, relay number, load level mark, event timestamp, event duration, sampling amplitude, load response value, and feedback status code from the current command echo test chain and command echo shadow chain; The event character code is generated according to the event type, target channel number, relay number, and load level mark, and then modulo hash processing is performed to generate the event hash code; Establish a Smith-Waterman alignment matrix, calculate the matching score for matrix cells with the same event hash code, and write empty cell markers for matrix cells with inconsistent event hash codes to complete the sparse cell evaluation; Filter events from the semantic event template of the test script and select the central anchor event, and perform bidirectional local comparison starting from the central anchor event; Read the event sequence number difference of each event relative to the central anchor event, write the event sequence number difference into the comparison symbol, generate an event alignment table based on the comparison path, and generate event residual data according to the event timestamp difference, event duration difference, sampling amplitude difference, load response value difference, and feedback status code difference between aligned events, and write it into the causal edge number.

7. The monitoring method for electrical integrated control testing equipment based on edge computing according to claim 1, characterized in that, The generation of the DeepESN state update queue includes: Read the event number, residual type, event timestamp, causal edge number, event timestamp residual, event duration residual, sampling amplitude residual, load response value residual, and feedback status code difference from the event residual data; Read the historical event residuals corresponding to the same test instruction number, target channel number, relay number and load level mark in the historical normal test segment, and generate the positive judgment boundary and the negative judgment boundary of the residual; The event residual data is converted into a symbolic residual sequence, and the corresponding symbols are written according to the comparison results between the residual and the positive decision boundary of the residual and the negative decision boundary of the residual. The symbolic residual sequence is divided according to the event timestamp order, and the window residual entropy value is generated based on the proportion of the corresponding symbols in each event window. Based on the distribution of window residual entropy values ​​corresponding to historical normal test segments, an entropy decision boundary is generated. Event window numbers whose window residual entropy values ​​exceed the entropy decision boundary are written into the entropy trigger blotting bucket in the residual entropy blotting diagram. The event window number, causal edge number, and residual type are associated and written into the residual entropy blotting diagram. Read the event window number in the entropy trigger cloning bucket, perform forward and reverse cumulative detection on the corresponding event residuals, generate residual cumulative judgment boundary based on the residual cumulative result distribution corresponding to the historical normal test segments, and write the event number, residual type, residual direction and causal edge number that exceed the residual cumulative judgment boundary into the DeepESN state update queue.

8. The monitoring method for electrical integrated control testing equipment based on edge computing according to claim 1, characterized in that, The generation of broken chain identification results includes: Read the device channel topology data from the standardized test and monitoring dataset, extract the connection relationships between communication interfaces, control loops, relay nodes, power output nodes, test channels, load nodes, sampling loops, and feedback message nodes, and generate a physical topology mask matrix; An improved DeepESN model is constructed, which includes a topological perforated reservoir, a single-step reverse-time echo excitation unit, an instantaneous mutual exclusion freezing unit, and a chain break readout layer. The topology perforation reservoir reads the physical topology mask matrix, writes empty connection markers to the positions in the reservoir connection matrix that do not belong to the connection relationship, and writes the DeepESN state update queue, event alignment table and event residual data to the corresponding reservoir node according to the event number and causal edge number to generate the topology perforation reservoir state. The single-step reverse-time echo excitation unit reads the state of the topological perforated reservoir, performs a reverse-time echo calculation based on the transpose of the reservoir connection matrix, generates a reverse-time echo state, and splices the state of the topological perforated reservoir with the reverse-time echo state to generate an echo splicing state. The instantaneous mutual exclusion freeze unit reads the missing event markers in the event alignment table and the trigger event numbers in the DeepESN state update queue, performs sign inversion and one event step freeze on the reserve pool node corresponding to the missing event marker, and generates a freeze pulse state. The broken chain readout layer reads the echo splicing state and the frozen pulse state, and reads the causal edge number and abnormal channel number from the physical topology mask matrix and the DeepESN state update queue to generate the broken chain identification result. The broken chain identification result includes the broken chain edge set, the broken chain event number, the abnormal channel number, and the initial marker of the abnormal source.

9. A monitoring method for electrical integrated control testing equipment based on edge computing according to claim 1, characterized in that, The generation of anomaly source markers includes: Read the chain break identification results and extract the abnormal channel numbers corresponding to the chain break edges pointing to relay nodes, power output nodes, test channels, load nodes and sampling circuits in the chain break edge set as suspected abnormal channels. Read the standard load configuration data, bypass switching status, safety interlock status and test task pause flag corresponding to the suspected abnormal channel, and write the suspected abnormal channel that meets the requirements of valid bypass switching status, allowed safety interlock status and valid test task pause flag into the bypass micro retest channel set. Perform standard load bypass micro-retest on suspected abnormal channels in the bypass micro-retest channel set to generate bypass command echo chain; Compare the bypass instruction echo chain with the instruction echo shadow chain execution event chain to generate bypass micro-retest results; When the bypass micro-retest result has a broken chain edge that is the same as the broken chain edge set, the initial mark of the abnormal source is corrected to the internal fault mark of the electrical integrated control test equipment. When the bypass micro-retest result does not have a broken chain edge that is the same as the broken chain edge set, the initial mark of the abnormal source is corrected to the non-equipment internal abnormal mark, and an abnormal source mark is generated.

10. A monitoring method for electrical integrated control testing equipment based on edge computing according to claim 1, characterized in that, The generation and output of monitoring results from the electrical integrated control test equipment includes: Read the broken chain edge set, broken chain event number and abnormal channel number from the broken chain identification results; read the abnormal source marker, bypass micro-retest results and historical broken chain recursive index rope in the monitoring record library. Establish the top-level node of the recursive index rope of the broken chain according to the broken chain edge number in the broken chain edge set, and write the broken chain edge number, broken chain event number, abnormal channel number, abnormal source marker and bypass micro-retest result to the top-level node; Search for the top-level rope node that matches the broken chain edge number in the historical broken chain recursive index rope, and embed the current broken chain event number, abnormal channel number, abnormal source marker and bypass micro-retest result into the sub-rope node of the consistent top-level rope node. Generate the number of chain break recurrences, the first chain break event number, the most recent chain break event number, and the chain break recurrence event sequence according to the hierarchical relationship between the top-level rope node and the sub-rope nodes; A chain break tracing index is generated based on the chain break edge number, chain break event number, abnormal channel number, and chain break recurrence event sequence. Monitoring results of electrical integrated control test equipment are generated according to the chain break identification results, abnormal source markers, bypass micro-retest results, chain break recursion index rope, and chain break tracing index.

Images