Intelligent fusion terminal operation and maintenance management method based on cloud edge collaboration

Abstract

This invention discloses a cloud-edge collaborative intelligent converged terminal operation and maintenance management method, comprising: reading terminal operation and maintenance attribute data corresponding to the intelligent converged terminal, generating a standard operation and maintenance image and distributing it to edge operation and maintenance nodes; collecting operation and maintenance status data of the intelligent converged terminal, constructing a terminal operation and maintenance phase boundary sliding window and generating a time series matrix; inputting the terminal operation and maintenance time series matrix into a MAD-GAN model, calculating an operation and maintenance anomaly score and determining the managed status; generating an operation and maintenance status offset object set and constructing a broken chain graph based on the operation and maintenance status data and the operation and maintenance anomaly score; mapping the operation and maintenance status offset objects to the broken chain graph and performing broken chain verification on the managed causal relationship edges; generating a reversible repair strategy based on the managed broken chain root cause object, distributing the managed keep-alive capsule and performing the repair. This invention achieves intelligent converged terminal anomaly identification, broken chain location, and unbroken chain repair through the terminal operation and maintenance phase boundary sliding window, cloud-edge dual-domain broken chain graph, and managed keep-alive capsule.

Description

Technical Field

[0001] This invention relates to the field of cloud-edge collaborative operation and maintenance management technology, and in particular to a method for operation and maintenance management of intelligent converged terminals based on cloud-edge collaboration. Background Technology

[0002] With the increasing number of intelligent converged terminals in power distribution, communication, and IoT systems, these terminals typically connect to a cloud-based operations and maintenance (O&M) platform via edge O&M nodes. The cloud then centrally manages configuration, version control, status monitoring, policy distribution, and fault handling. Existing cloud-edge collaborative O&M methods usually collect data such as terminal heartbeats, link latency, resource usage, configuration versions, application versions, and alarm logs. Based on preset thresholds or individual alarm rules, they determine whether the terminal is abnormal, and then the cloud generates O&M policies such as restart, upgrade, configuration reissue, or log back.

[0003] However, in actual operation and maintenance, anomalies in intelligent converged terminals are often not caused by a single state, but rather by the combined effects of multiple states such as communication link jitter, edge agent response latency, configuration drift, version inconsistency, policy queue blocking, and abnormal receipts. Existing methods mainly rely on fixed thresholds or single alarms for judgment, making it difficult to identify the boundary between a terminal's transition from a stable managed state to a critical managed state or an unstable managed state, and also making it difficult to distinguish between common accompanying anomalies and key anomalies that truly affect remote managed links.

[0004] Existing cloud-based policy delivery methods typically lack verification of the broken links in the relationship between the cloud management domain, edge forwarding domain, and terminal execution domain regarding instruction arrival, execution acceptance, and status feedback. When multiple anomalies exist simultaneously, it is impossible to accurately determine the root cause of cloud policies failing to arrive, terminals failing to execute, or execution results failing to be fed back. This can easily lead to erroneous restarts, erroneous upgrades, duplicate policy delivery, or complete loss of terminal connectivity during the repair process.

[0005] Therefore, how to provide a cloud-edge collaborative intelligent converged terminal operation and maintenance management method 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 an intelligent converged terminal operation and maintenance management method based on cloud-edge collaboration. This invention fully utilizes cloud-edge collaborative operation and maintenance, generative adversarial networks, multi-domain management and disconnection analysis, and keep-alive reversible repair technology. By constructing a terminal operation and maintenance phase boundary sliding window to identify the boundary of change of intelligent converged terminals from stable management state to critical management state and unstable management state, and combining cloud-edge dual-domain management causal disconnection graph to locate the root cause of management and disconnection that causes cloud policies to be unreachable, terminals to be unable to execute, or execution results to be unable to be returned, the invention then uses a management keep-alive capsule to achieve segmented repair and failure rollback control under the condition of no disconnection. It has the advantages of high anomaly identification accuracy, strong management and disconnection location capability, high continuity of operation and maintenance repair, and low risk of terminal disconnection.

[0007] The cloud-edge collaborative intelligent converged terminal operation and maintenance management method according to embodiments of the present invention includes:

[0008] The cloud-based operations and maintenance platform reads the terminal operations and maintenance attribute data corresponding to the intelligent converged terminal, generates the corresponding standard operations and maintenance image, and distributes the standard operations and maintenance image to the corresponding edge operations and maintenance node.

[0009] Collect operation and maintenance status data of intelligent converged terminals, construct terminal operation and maintenance phase boundary sliding window with multiple consecutive collection cycles, and convert the operation and maintenance status data into terminal operation and maintenance time series matrix according to fixed field order;

[0010] Input the terminal operation and maintenance time series matrix into the MAD-GAN model with dual generator-multi-task discriminator structure, calculate the operation and maintenance anomaly score, and determine the stable management state, critical management state or unstable management state of the intelligent fusion terminal based on the operation and maintenance anomaly score.

[0011] Based on the operation and maintenance status data, operation and maintenance anomaly scores, and the state transition relationships between stable management status, critical management status, and unstable management status, a set of operation and maintenance status offset objects is generated, a cloud-edge dual-domain management causal chain break graph is constructed, and management causal association edges are established.

[0012] Map the set of offset objects in the operation and maintenance status to the cloud-edge dual-domain managed cause-effect chain breakage diagram, perform chain breakage verification on the managed cause-effect association edge, and identify the offset objects that cause cloud policies to be unable to reach intelligent converged terminals, intelligent converged terminals to be unable to execute policies, or execution results to be unable to be returned to the cloud as the root cause objects of the managed chain breakage.

[0013] The cloud-based operations and maintenance platform generates a reversible repair strategy based on the root cause of the chain breakage. The edge operations and maintenance nodes issue the management and survival capsule to the intelligent converged terminal. After establishing the survival heartbeat and receipt token, the reversible repair strategy is broken down into multiple action segments and executed sequentially. The execution of the next action segment is controlled or the failure rollback segment is triggered based on the operations and maintenance status data.

[0014] Optionally, the terminal operation and maintenance attribute data includes terminal identification attributes, communication access attributes, edge affiliation attributes, service operation attributes, configuration baseline attributes, version baseline attributes, component dependency attributes, and remote management attributes.

[0015] Optionally, generating the corresponding standard operation and maintenance image includes:

[0016] The mirror index key is determined by matching the terminal identification attribute with the communication access attribute, edge affiliation attribute, and service operation attribute.

[0017] Based on the configuration baseline attribute, version baseline attribute, and component dependency attribute, write the communication baseline field, configuration baseline field, version baseline field, proxy status baseline field, container status baseline field, and resource usage baseline field;

[0018] Write the policy queue baseline field and receipt status baseline field according to the remote management attributes;

[0019] Perform integrity checks on all baseline fields and generate mirror hash values;

[0020] Generate an image version number and store it as a standard operation and maintenance image.

[0021] Optionally, the step of constructing a terminal operation and maintenance phase boundary sliding window, which converts operation and maintenance status data into a terminal operation and maintenance time series matrix according to a fixed field order, includes:

[0022] Edge operation and maintenance nodes collect operation and maintenance status data of intelligent converged terminals according to a preset collection cycle, and write the operation and maintenance status data at each collection moment into the time-series buffer. The operation and maintenance status data includes communication link status, agent response status, configuration status, version status, container running status, interface call status, resource usage status, log accumulation status, policy queue status, and receipt feedback status.

[0023] Align the operation and maintenance status data of each collection time in the time-series buffer with the standard operation and maintenance image to generate a management continuity flag and a status transition flag corresponding to each collection time.

[0024] Read the state transition markers in the order of acquisition time, and determine the acquisition time when the first interruption, delay, blockage or acknowledgment mismatch of the management continuity marker is determined as the phase boundary start anchor point. The acquisition time when the instruction reception abnormality, agent response abnormality, strategy execution abnormality or result acknowledgment abnormality is confirmed in the current acquisition cycle is determined as the phase boundary confirmation anchor point.

[0025] Using the phase boundary start anchor point and the phase boundary confirmation anchor point as boundaries, the stable support data before the phase boundary start anchor point, the transition data between the phase boundary start anchor point and the phase boundary confirmation anchor point, and the confirmation data after the phase boundary confirmation anchor point are extracted from the time-series buffer and combined to generate the terminal operation and maintenance phase boundary sliding window.

[0026] Following a fixed field order of communication link status, agent response status, configuration status, version status, container running status, interface call status, resource usage status, log accumulation status, policy queue status, and receipt feedback status, the operation and maintenance status data in the terminal operation and maintenance phase window is written into the matrix row time by time to generate the terminal operation and maintenance time series matrix.

[0027] Optionally, the calculation of the operation and maintenance anomaly score, and the determination of the stable management status, critical management status, or unstable management status of the intelligent fusion terminal based on the operation and maintenance anomaly score, includes:

[0028] The terminal operation and maintenance time series matrix is ​​input into the MAD-GAN model with a dual generator-multi-task discriminator structure. The MAD-GAN model includes a reconstruction generator, a prediction generator, a multi-task discriminator, and a managed link masking layer.

[0029] The managed link masking layer marks the link fields corresponding to instruction reception, agent response, policy execution and result receipt based on the stability support data, transition data and confirmation data in the terminal operation and maintenance phase boundary sliding window, and generates the managed link masking sequence.

[0030] The reconstructor performs link-keeping reconstruction on the terminal operation and maintenance timing matrix based on the managed link masking sequence, generates a normal managed reconstruction sequence, and generates reconstruction deviation results based on the field deviation status between the normal managed reconstruction sequence and the terminal operation and maintenance timing matrix.

[0031] The prediction generator predicts the corresponding control state of the confirmation data based on the stable support data and transition data, generates a phase boundary prediction sequence, and generates a prediction deviation result based on the state deviation between the phase boundary prediction sequence and the confirmation data.

[0032] The multi-task discriminator performs reconstruction consistency discrimination on normal managed reconstruction sequences, prediction consistency discrimination on phase boundary prediction sequences, and link continuity discrimination on managed link masking sequences, and outputs reconstruction consistency results, prediction consistency results, and link continuity results.

[0033] An operation and maintenance anomaly score is generated based on the reconstruction deviation result, prediction deviation result, reconstruction consistency result, prediction consistency result, and link continuity result. The operation and maintenance anomaly score is compared with the corresponding stable management threshold, critical management threshold, and unstable management threshold in the standard operation and maintenance image. Intelligent converged terminals below the stable management threshold are identified as stable management states, intelligent converged terminals that reach the critical management threshold but do not reach the unstable management threshold are identified as critical management states, and intelligent converged terminals that reach the unstable management threshold are identified as unstable management states.

[0034] Optionally, the step of generating a set of operation and maintenance status offset objects, constructing a cloud-edge dual-domain managed causal chain break graph, and establishing managed causal association edges includes:

[0035] Read the operation and maintenance status data, operation and maintenance anomaly scores, and the state transition relationship between stable management status, critical management status, and unstable management status. The collection segment from stable management status to critical management status is determined as the critical offset segment, and the collection segment from critical management status to unstable management status is determined as the unstable offset segment.

[0036] Read the communication link status, agent response status, configuration status, version status, container running status, interface call status, resource usage status, log accumulation status, policy queue status, and receipt feedback status within the critical offset section and unstable offset section. Convert the status fields that deviate from the standard operation and maintenance image into corresponding offset objects and generate a set of operation and maintenance status offset objects.

[0037] Following the management sequence of cloud policy generation, edge policy forwarding, terminal policy execution, and execution result feedback, cloud management domain, edge forwarding domain, and terminal execution domain are constructed respectively.

[0038] According to the direction of operation and maintenance strategy distribution, the strategy generation node, edge agent node, channel scheduling node, terminal operation and maintenance agent node, configuration loading node, version adaptation node and business container node are connected in sequence to form an instruction distribution chain. According to the direction of execution result feedback, the business container node, execution receipt node, terminal operation and maintenance agent node, edge cache node, edge agent node and result confirmation node are connected in sequence to form a status feedback chain. The instruction distribution chain and status feedback chain are combined to form a cloud-edge dual-domain managed causal chain break diagram.

[0039] Each offset object in the set of operation and maintenance status offset objects is attached to the corresponding node in the cloud-edge dual-domain management causal chain diagram, and management causal association edges are established according to the influence relationship of each offset object on the instruction issuance chain, terminal local execution chain and status feedback chain.

[0040] Optionally, the offset object that causes the cloud policy to fail to reach the intelligent converged terminal, the intelligent converged terminal to fail to execute the policy, or the execution result to fail to be sent back to the cloud is identified as the root cause object of the management chain break, including:

[0041] Based on the type of offset object, the communication-type offset objects in the operation and maintenance status offset object set are mapped to the channel scheduling node and the communication component node; the agent-type offset objects are mapped to the edge agent node and the terminal operation and maintenance agent node; the configuration-type offset objects are mapped to the configuration image node and the configuration loading node; the version-type offset objects are mapped to the version baseline node and the version adaptation node; the container-type offset objects are mapped to the business container node; the policy queue-type offset objects are mapped to the policy queue node; and the receipt-type offset objects are mapped to the execution receipt node and the result confirmation node.

[0042] Based on the mapping position of each offset object in the cloud-edge dual-domain managed causal chain break diagram, read the managed causal association edges associated with each offset object, and divide the managed causal association edges into instruction arrival verification edges, execution acceptance verification edges, status return verification edges, and failure masking verification edges according to the cloud policy distribution direction and execution result return direction.

[0043] Perform chain break verification on the instruction arrival verification edge and execution acceptance verification edge. If the cloud policy cannot reach the corresponding execution node through the policy generation node, edge agent node, channel scheduling node and terminal operation and maintenance agent node, or the terminal operation and maintenance agent node cannot pass the cloud policy to the configuration loading node, version adaptation node or business container node, then mark the corresponding offset object as a candidate object for chain break management.

[0044] Perform chain break verification on the status feedback verification edge and the failure masking verification edge. If the execution result of the action segment cannot be fed back to the result confirmation node through the execution receipt node, terminal operation and maintenance agent node, edge cache node and edge agent node, or if the execution result reported by the terminal is inconsistent with the edge observation result, then mark the corresponding offset object as a feedback chain break candidate object or a failure masking candidate object.

[0045] Perform association merging on candidate objects for managed chain break, candidate objects for backhaul chain break, and candidate objects for failure masking. The offset objects that can cause cloud policies to fail to reach smart converged terminals, smart converged terminals to fail to execute policies, or execution results to fail to be backhauled to the cloud are identified as the root cause objects of managed chain break.

[0046] Optionally, the edge operation and maintenance node issues a management and keep-alive capsule to the intelligent converged terminal. After establishing a keep-alive heartbeat and receipt token, the reversible repair strategy is broken down into multiple action segments and executed sequentially, including:

[0047] Based on the object type, graph node, and associated causal relationship edge of the root cause of the chain break, a corresponding reversible repair strategy is generated. The reversible repair strategy includes communication recovery strategy, proxy reconnection strategy, configuration resending strategy, version adaptation strategy, strategy queue unblocking strategy, and receipt link recovery strategy.

[0048] The management and survival capsule is issued to the intelligent converged terminal. The management and survival capsule includes a survival heartbeat field, a receipt token field, a status snapshot field, a breakpoint resume field, and a rollback trigger field.

[0049] After establishing the minimum management heartbeat based on the keep-alive heartbeat field, establishing the action segment receipt identifier based on the receipt token field, and recording the operation and maintenance status before repair based on the status snapshot field in the intelligent fusion terminal, the management keep-alive capsule is confirmed to be established.

[0050] The reversible repair strategy is divided into a keep-alive anchoring segment, a chain break removal segment, a root cause repair segment, a state lock segment, and a failure rollback segment, and is issued and executed in the order of the keep-alive anchoring segment, the chain break removal segment, the root cause repair segment, and the state lock segment.

[0051] After each action segment is completed, read the operation and maintenance status data and the action segment execution result. When the keep-alive heartbeat, receipt token, configuration verification, version adaptation and policy queue status all meet the conditions for the next action segment execution, issue the next action segment. When any status triggers the rollback condition recorded in the rollback trigger field, stop the action segment and execute the failure rollback segment.

[0052] The beneficial effects of this invention are:

[0053] This invention constructs a terminal operation and maintenance phase boundary sliding window and inputs operation and maintenance status data into a MAD-GAN model with a dual generator-multi-task discriminator structure. This enables joint identification of communication link status, agent response status, configuration status, version status, policy queue status, and receipt feedback status. This avoids judging anomalies based solely on a single threshold or a single alarm, and improves the accuracy and timeliness of identification when intelligent fusion terminals transition from a stable management state to a critical management state or an unstable management state.

[0054] This invention constructs a cloud-edge dual-domain managed causal chain break diagram to model the policy delivery, terminal execution, and status feedback processes in the cloud management domain, edge forwarding domain, and terminal execution domain. Based on the managed causal association edge execution chain break verification, it can distinguish between ordinary accompanying anomalies and the root cause objects of managed chain break that truly cause cloud policies to be unreachable, terminals to be unable to execute, or execution results to be unable to be fed back, thereby reducing the probability of erroneous restarts, erroneous upgrades, and duplicate policy delivery.

[0055] This invention issues a managed and liveness capsule to intelligent converged terminals. After establishing a liveness heartbeat and receipt token, the reversible repair strategy is broken down into multiple action segments and executed sequentially. This allows for continuous monitoring of operation and maintenance status data during the repair process. Based on the execution results of the action segments, the next action segment can be controlled or a failure rollback segment can be triggered, reducing the risk of the terminal completely losing connection during the repair process and improving the continuity, security, and recoverability of cloud-edge collaborative operation and maintenance. Attached Figure Description

[0056] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings:

[0057] Figure 1 The flowchart shows the intelligent converged terminal operation and maintenance management method based on cloud-edge collaboration proposed in this invention.

[0058] Figure 2 This is a schematic diagram of the processing flow of the MAD-GAN model with a dual generator-multi-task discriminator structure in the cloud-edge collaborative intelligent fusion terminal operation and maintenance management method proposed in this invention. Detailed Implementation

[0059] 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.

[0060] refer to Figure 1 and Figure 2 The cloud-edge collaborative intelligent converged terminal operation and maintenance management method includes:

[0061] The cloud-based operations and maintenance platform reads the terminal operations and maintenance attribute data corresponding to the intelligent converged terminal, generates the corresponding standard operations and maintenance image, and distributes the standard operations and maintenance image to the corresponding edge operations and maintenance node.

[0062] Collect operation and maintenance status data of intelligent converged terminals, construct terminal operation and maintenance phase boundary sliding window with multiple consecutive collection cycles, and convert the operation and maintenance status data into terminal operation and maintenance time series matrix according to fixed field order;

[0063] Input the terminal operation and maintenance time series matrix into the MAD-GAN model with dual generator-multi-task discriminator structure, calculate the operation and maintenance anomaly score, and determine the stable management state, critical management state or unstable management state of the intelligent fusion terminal based on the operation and maintenance anomaly score.

[0064] Based on the operation and maintenance status data, operation and maintenance anomaly scores, and the state transition relationships between stable management status, critical management status, and unstable management status, a set of operation and maintenance status offset objects is generated, a cloud-edge dual-domain management causal chain break graph is constructed, and management causal association edges are established.

[0065] Map the set of offset objects in the operation and maintenance status to the cloud-edge dual-domain managed cause-effect chain breakage diagram, perform chain breakage verification on the managed cause-effect association edge, and identify the offset objects that cause cloud policies to be unable to reach intelligent converged terminals, intelligent converged terminals to be unable to execute policies, or execution results to be unable to be returned to the cloud as the root cause objects of the managed chain breakage.

[0066] The cloud-based operations and maintenance platform generates a reversible repair strategy based on the root cause of the chain breakage. The edge operations and maintenance nodes issue the management and survival capsule to the intelligent converged terminal. After establishing the survival heartbeat and receipt token, the reversible repair strategy is broken down into multiple action segments and executed sequentially. The execution of the next action segment is controlled or the failure rollback segment is triggered based on the operations and maintenance status data.

[0067] In this embodiment, the terminal operation and maintenance attribute data includes terminal identification attributes, communication access attributes, edge affiliation attributes, service operation attributes, configuration baseline attributes, version baseline attributes, component dependency attributes, and remote management attributes.

[0068] In this embodiment, generating the corresponding standard operation and maintenance image includes:

[0069] The mirror index key is determined by matching the terminal identification attribute with the communication access attribute, edge affiliation attribute, and service operation attribute.

[0070] Based on the configuration baseline attribute, version baseline attribute, and component dependency attribute, write the communication baseline field, configuration baseline field, version baseline field, proxy status baseline field, container status baseline field, and resource usage baseline field;

[0071] Write the policy queue baseline field and receipt status baseline field according to the remote management attributes;

[0072] Perform integrity checks on all baseline fields and generate mirror hash values;

[0073] Generate an image version number and store it as a standard operation and maintenance image.

[0074] In this embodiment, the step of constructing a terminal operation and maintenance phase boundary sliding window, which converts operation and maintenance status data into a terminal operation and maintenance time series matrix according to a fixed field order, includes:

[0075] Edge operation and maintenance nodes collect operation and maintenance status data of intelligent converged terminals according to a preset collection cycle, and write the operation and maintenance status data at each collection moment into the time-series buffer. The operation and maintenance status data includes communication link status, agent response status, configuration status, version status, container running status, interface call status, resource occupation status, log accumulation status, policy queue status, and receipt feedback status, wherein the preset collection cycle is 2s to 10s.

[0076] The operation and maintenance status data of each collection time in the time-series buffer are aligned with the standard operation and maintenance image to generate a management continuity flag and a state transition flag corresponding to each collection time. The management continuity flag is used to record the continuous state of instruction reception, agent response, policy execution and result receipt. The state transition flag is used to record the transition state between stable management state, critical management state and unstable management state.

[0077] Read the state transition markers in the order of acquisition time, and determine the acquisition time when the first interruption, delay, blockage or acknowledgment mismatch of the management continuity marker is determined as the phase boundary start anchor point. The acquisition time when the instruction reception abnormality, agent response abnormality, strategy execution abnormality or result acknowledgment abnormality is confirmed in the current acquisition cycle is determined as the phase boundary confirmation anchor point.

[0078] Using the phase boundary start anchor point and phase boundary confirmation anchor point as boundaries, stable support data before the phase boundary start anchor point, transition data between the phase boundary start anchor point and the phase boundary confirmation anchor point, and confirmation data after the phase boundary confirmation anchor point are extracted from the time-series buffer and combined to generate a terminal operation and maintenance phase boundary sliding window. Specifically, the generation of the terminal operation and maintenance phase boundary sliding window is as follows:

[0079] The operation and maintenance status data corresponding to the 10 consecutive collection cycles before the phase boundary starting anchor point are written into the first segment of the sliding window in chronological order as stable support data. The operation and maintenance status data corresponding to all collection cycles between the phase boundary starting anchor point and the phase boundary confirmation anchor point are read, and the data immediately following the first segment is written into the middle segment of the sliding window as transition data. The operation and maintenance status data corresponding to the 5 consecutive collection cycles after the phase boundary confirmation anchor point are then read and written into the last segment of the sliding window as confirmation data. If there are less than 10 collection cycles before the phase boundary starting anchor point, the data of the first available collection cycle is copied in reverse chronological order to make up the difference. If there are less than 5 collection cycles after the phase boundary confirmation anchor point, the data of the last available collection cycle is copied in chronological order to make up the difference. The three segments of data are continuously spliced ​​on the time axis to generate the terminal operation and maintenance phase boundary sliding window according to the fixed field order of communication link status, agent response status, configuration status, version status, container running status, interface call status, resource usage status, log accumulation status, policy queue status, and receipt feedback status.

[0080] Following a fixed field order for communication link status, proxy response status, configuration status, version status, container running status, interface call status, resource usage status, log backlog status, policy queue status, and receipt feedback status, the maintenance status data in the terminal maintenance phase window is written into the matrix rows time-by-time to generate a terminal maintenance time-series matrix. Specifically, the generation of the terminal maintenance time-series matrix is ​​as follows:

[0081] Create a row for each acquisition moment within the sliding window, arranged sequentially by time. Create a column for each of the ten maintenance status fields, with the column order remaining fixed. For each acquisition moment, first read the communication link status. If the field is an enumerated value, map it to an integer according to the order of Normal -0, Delay -1, Packet Loss -2, and Link Disconnection -3. Then read the baseline mean and standard deviation of the communication link from the standard maintenance image, normalize it by subtracting the baseline mean from the current value and dividing by the standard deviation, and write it into the first column of the current row. Repeat the enumeration mapping and normalization process for the agent response status to the receipt feedback status, filling the ten normalized values ​​into the ten columns of the current row. After writing all acquisition moments, if the number of matrix rows is less than the preset minimum sliding window length, copy the row sequence in reverse chronological order from the earliest row until the minimum length is reached. If the number of matrix rows exceeds the preset maximum sliding window length, retain the latest rows and delete earlier rows to ensure the number of rows does not exceed the maximum length. Finally, obtain a terminal maintenance time series matrix with the number of rows meeting the preset range, the number of columns fixed at ten, and all elements normalized.

[0082] In this embodiment, the calculation of the operation and maintenance anomaly score, and the determination of the stable management status, critical management status, or unstable management status of the intelligent fusion terminal based on the operation and maintenance anomaly score, include:

[0083] The terminal operation and maintenance time series matrix is ​​input into the MAD-GAN model with a dual generator-multi-task discriminator structure. The MAD-GAN model includes a reconstruction generator, a prediction generator, a multi-task discriminator, and a managed link masking layer.

[0084] The managed link masking layer marks the link fields corresponding to instruction reception, agent response, policy execution and result receipt based on the stability support data, transition data and confirmation data in the terminal operation and maintenance phase boundary sliding window, and generates the managed link masking sequence.

[0085] The refactoring generator performs link-keeping refactoring on the terminal operation and maintenance timing matrix based on the managed link masking sequence, generates a normal managed refactoring sequence, and generates a refactoring deviation result based on the field deviation status between the normal managed refactoring sequence and the terminal operation and maintenance timing matrix. Specifically, the generated refactoring deviation result is as follows:

[0086] Create a deviation record table with the same row and column size as the terminal operation and maintenance time series matrix; starting from the first row, traverse the normal management reconstruction sequence and the terminal operation and maintenance time series matrix in chronological order, and write the absolute value of the difference between two values ​​for numerical fields into the corresponding cell, and write 0 for enumerated fields if the states are consistent and 1 if the states are inconsistent; after traversal, calculate the maximum deviation, average deviation, and consecutive over-threshold sampling point count for each field in the entire sliding window, write the maximum deviation into the field-level peak vector, the average deviation into the field-level average vector, and the consecutive over-threshold count into the field-level persistence vector; finally, write the three field-level vectors and the corresponding cell-level deviation records into the reconstruction deviation result.

[0087] The prediction generator predicts the corresponding control state based on stable support data and transition data, generates a phase boundary prediction sequence, and generates a prediction deviation result based on the state deviation between the phase boundary prediction sequence and the control data. Specifically, the generation of the phase boundary prediction sequence is as follows:

[0088] First, in the stable support data, the continuous values ​​of communication link status, agent response status, policy queue status, and receipt feedback status are recorded in chronological order, and the most recent normal value of each field is written into the baseline vector. Then, in the transition data, adjacent sampling points are compared time-by-time. If the value of the same field is monotonically increasing, it is marked as an upward trend; if it is monotonically decreasing, it is marked as a downward trend. If the state of an enumerated field continues to deteriorate, it is marked as a deteriorating trend; if it continues to recover, it is marked as a recovering trend. The duration of the trend is accumulated for each field. Then, in chronological order, the baseline vector is copied row by row for the sampling time corresponding to the confirmed data. Based on the trend direction and duration of each field in the transition segment, the copied values ​​are incrementally adjusted: the upward trend field increases by the maintenance step size, the downward trend field decreases by the maintenance step size, the deteriorating trend field jumps to the next level of abnormal state, and the recovering trend field falls back to the previous level of normal state, generating a phase boundary prediction sequence with the same number of rows as the confirmed data.

[0089] The prediction bias results are generated as follows:

[0090] Create a prediction deviation record table with the same row and column size as the phase boundary prediction sequence, and align the phase boundary prediction sequence and the confirmed data row by row starting from the first row; write the absolute amount of the difference between the predicted value and the actual value in the numerical field, and write a Boolean flag indicating whether the predicted state is equal to the actual state in the enumeration field; after traversal, calculate the mean deviation, peak deviation, and continuous over-threshold duration for each field by column, write the three statistics into the field-level deviation summary vector, and combine the unit-level deviation record and the field-level deviation summary to form the prediction deviation result;

[0091] The multi-task discriminator performs reconstruction consistency discrimination on normal managed reconstruction sequences, prediction consistency discrimination on phase boundary prediction sequences, and link continuity discrimination on managed link masking sequences, and outputs reconstruction consistency results, prediction consistency results, and link continuity results. Specifically, the output of reconstruction consistency results, prediction consistency results, and link continuity results is as follows:

[0092] For the three types of discrimination tasks, establish unit-level consistency tables and field-level consistency tables respectively;

[0093] In the reconstruction consistency judgment, the normal management reconstruction sequence and the terminal operation and maintenance time sequence matrix are aligned row by row. For numerical fields, if the absolute value of the difference between two values ​​is less than twice the benchmark standard deviation of the corresponding field, it is considered consistent; otherwise, it is considered inconsistent. For enumerated fields, if the status is the same, it is considered consistent; otherwise, it is considered inconsistent. The consistency flags of the ten fields in each row are written into the reconstruction unit-level consistency table. If all the consistency flags in a row are consistent, it is considered row consistent. The reconstruction row consistency rate is obtained by counting row by row, and then the field consistency rate is counted by column. The row consistency rate and the field consistency rate are written together into the reconstruction field-level consistency table, and the output is the reconstruction consistency result.

[0094] In the prediction consistency judgment, the phase boundary prediction sequence and the confirmed data are aligned row by row. The same numerical threshold is used to judge and enumerate the equal value judgment rules to record the unit-level consistency mark. The row consistency rate and field consistency rate are statistically written into the prediction field-level consistency table, and the output is the prediction consistency result.

[0095] In the link continuity determination, the managed link masking sequence is read line by line. The four link channel markers of instruction reception, agent response, policy execution and result receipt are checked to see if they appear in order and without any missing ones. If they are satisfied, they are recorded as continuous and consistent links. Otherwise, they are recorded as broken and inconsistent links. The proportion of lines with simultaneous continuous and consistent links is calculated, and the proportion is written into the link continuity field and output as the link continuity result.

[0096] An operation and maintenance anomaly score is generated based on the reconstruction deviation result, prediction deviation result, reconstruction consistency result, prediction consistency result, and link continuity result. The operation and maintenance anomaly score is compared with the corresponding stable management threshold, critical management threshold, and unstable management threshold in the standard operation and maintenance image. Intelligent fusion terminals below the stable management threshold are identified as stable management status, intelligent fusion terminals that reach the critical management threshold but do not reach the unstable management threshold are identified as critical management status, and intelligent fusion terminals that reach the unstable management threshold are identified as unstable management status.

[0097] The managed link masking layer includes:

[0098] Field Role Index Unit: Receives the terminal operation and maintenance time sequence matrix, and writes the field number and field type for each field according to the fixed field order of communication link status, agent response status, configuration status, version status, container running status, interface call status, resource usage status, log accumulation status, policy queue status, and receipt feedback status;

[0099] Three-segment interval marking unit: Reads the stable support data, transition data and confirmation data in the terminal operation and maintenance phase boundary sliding window, and marks each collection moment as a stable support segment, phase boundary transition segment or abnormal confirmation segment;

[0100] The managed link role mapping unit maps communication link status and policy queue status to instruction receiving link fields, agent response status to agent response link fields, and configuration status, version status, container running status, interface call status, resource usage status, and log accumulation status to policy execution link fields.

[0101] Anomaly Trigger Flag Unit: Reads operation and maintenance status data every hour and writes fields that indicate interruption of instruction reception, delay of agent response, blocking of policy queue, or mismatch of receipt into the anomaly trigger flag.

[0102] Masking sequence output unit: According to the acquisition time order, the field number, field type, interval marker, management link role and abnormal trigger marker are combined into a management link masking sequence;

[0103] After receiving the terminal operation and maintenance time sequence matrix, the managed link masking layer first reads ten operation and maintenance status fields by column by the field role indexing unit, and writes the field number and field type for each column; then, the three-segment interval marking unit reads the collection time by row, writes the data rows belonging to the phase boundary start anchor point into the stable support segment mark, writes the data rows belonging to the phase boundary start anchor point to the phase boundary confirmation anchor point into the phase boundary transition segment mark, and writes the data rows belonging to the phase boundary confirmation anchor point into the abnormal confirmation segment mark; the managed link role mapping unit assigns the fields in each row to the instruction receiving link, agent response link, policy execution link, and result receipt link respectively; then, the abnormal triggering marking unit checks each row and writes the fields with communication interruption, agent delay, policy queue blocking, or receipt mismatch into the abnormal triggering marking; the masking sequence output unit outputs the managed link masking sequence with the same number of rows as the terminal operation and maintenance time sequence matrix in the original time order.

[0104] Refactoring generators include:

[0105] Masking Fusion Input Unit: Receives the terminal operation and maintenance time sequence matrix and the managed link masking sequence, and concatenates the operation and maintenance status field, field type, interval marker and managed link role at the same collection time to generate a fusion input sequence;

[0106] Stable support coding unit: Reads the stable support segment from the fused input sequence, extracts the continuous change records of communication link status, agent response status, policy queue status and receipt feedback status in time order, and generates a stable managed link baseline;

[0107] Link maintenance constraint unit: Reads the stable managed link baseline and managed link masking sequence, and establishes link maintenance constraints in the order of instruction receiving link, agent response link, policy execution link and result receipt link;

[0108] Transition backfill decoding unit: Reads the fused input sequence corresponding to the transition segment and the confirmation segment, performs normal management status backfilling on the abnormal field under the link maintenance constraint, and generates a normal management reconstruction sequence;

[0109] Reconstruction Deviation Summary Unit: Aligns the normal management reconstruction sequence with the terminal operation and maintenance time sequence matrix item by item according to the same collection time and the same field order to generate reconstruction deviation results;

[0110] The reconstruction generator first reads the terminal operation and maintenance time sequence matrix line by line by the mask fusion input unit, and concatenates the mask markers of the managed links corresponding to the same row to the operation and maintenance status data of that row to form a fusion input sequence; the stability support encoding unit reads each row of data in the stability support segment, records in chronological order whether the communication link status is continuous, whether the agent response has returned, whether the policy queue can receive tasks, and whether the acknowledgment feedback is valid, and generates a stable managed link baseline; the link maintenance constraint unit, in the order of instruction receiving link first, agent response link next, policy execution link last, and result acknowledgment link last, performs checks on the transition segment and confirmation. The reconstruction order of fields in the segment is restricted; under this restriction, the transition backfill decoding unit performs normal management status backfilling on the communication link status, agent response status, configuration status, version status, policy queue status, and receipt feedback status in the transition segment and confirmation segment, generating a normal management reconstruction sequence with the same row and column size as the terminal operation and maintenance time sequence matrix; the reconstruction deviation summary unit reads the normal management reconstruction sequence and the terminal operation and maintenance time sequence matrix row by row, records the deviation between the current value and the reconstruction value for the numerical field, records whether the current state and the reconstruction state are consistent for the enumerated field, and summarizes all deviation records into a reconstruction deviation result;

[0111] Prediction generators include:

[0112] Stable state memory unit: Reads the stability support data in the terminal operation and maintenance phase boundary sliding window, records the communication link status, agent response status, policy queue status and receipt feedback status before the phase boundary starting anchor point, and generates a stable management status benchmark.

[0113] Transition Trend Extraction Unit: Reads transition data between the phase boundary start anchor point and the phase boundary confirmation anchor point, compares the changes in operation and maintenance status according to adjacent collection times, and extracts the communication link change trend, agent response change trend, policy queue change trend and receipt feedback change trend.

[0114] Phase boundary propulsion prediction unit: Based on the stable pipe-to-pipe state benchmark and transition trend, it generates the pipe-to-pipe state prediction results after the phase boundary confirmation anchor point;

[0115] Confirmation data alignment unit: Read the confirmation data and align the monitoring status prediction results with the confirmation data item by item according to the collection time and field order;

[0116] Prediction Deviation Summary Unit: Generates prediction deviation results based on the field deviation status between the prediction results and the confirmed data of the management status;

[0117] The prediction generator first reads the stable support data before the phase boundary starting anchor point from the stable state memory unit, and writes the continuously normal communication link state, agent response state, policy queue state, and receipt feedback state into the stable management state benchmark. The transition trend extraction unit then reads the transition data between the phase boundary starting anchor point and the phase boundary confirmation anchor point, compares the field states of adjacent acquisition times in chronological order, and registers the link delay increase, agent response slowdown, policy queue growth, and receipt timeout increase as deterioration trends, and the link recovery, agent response shortening, policy queue reduction, and receipt recovery as recovery trends. The phase boundary advancement prediction unit generates the management state prediction result corresponding to the acquisition time of the confirmation segment based on the stable management state benchmark and the transition trend. The confirmation data alignment unit aligns the prediction result with the confirmation data line by line in chronological order, and aligns them column by column according to the fixed field order from the communication link state to the receipt feedback state. The prediction deviation summary unit checks each aligned field, writes the field where the prediction state is inconsistent with the confirmation state into the prediction deviation mark, and writes the field where it is inconsistent for multiple consecutive acquisition times into the continuous prediction deviation mark, generating the prediction deviation result.

[0118] The multi-task discriminator includes:

[0119] Shared temporal feature unit: Receives normal management reconstruction sequence, management status prediction result and management link masking sequence, extracts the reconstruction status, prediction status and link masking status at the same time according to the acquisition time order, and generates shared temporal features;

[0120] Reconstruction Consistency Judgment Unit: Reads the reconstruction status from the shared timing features, performs consistency judgment on the field order, segment order, and link order of the normal managed reconstruction sequence and the terminal operation and maintenance timing matrix, and generates reconstruction consistency results;

[0121] Prediction consistency discrimination unit: Reads the prediction status from the shared time series features, compares the management status prediction results with the confirmed data for consistency in collection time and field status, and generates prediction consistency results;

[0122] Link continuity determination unit: Reads the link masking status in the shared timing characteristics, and determines whether there is a link interruption, link jump, link inversion or link missing according to the order of instruction receiving link, agent response link, policy execution link and result receipt link;

[0123] Discriminant Result Aggregation Unit: This unit aggregates the reconstructed consistency results, predicted consistency results, and link continuity results according to the collection time and field order to generate a discriminant result set.

[0124] The multi-task discriminator first uses a shared timing feature unit to read the normal management reconstruction sequence, management status prediction results, and management link masking sequence in chronological order. It then merges the reconstruction status, prediction status, and link masking status at the same acquisition time into a shared timing feature. The reconstruction consistency discriminator reads the reconstruction status from the shared timing feature, checking each moment whether the normal management reconstruction sequence maintains the same field order as the terminal operation and maintenance timing matrix, the segment order from the stable support segment to the confirmation segment, and the link order from instruction reception to result receipt, and outputs the reconstruction consistency result. The prediction consistency discriminator reads the prediction status from the shared timing feature, checking each moment whether the management status prediction result is consistent with the actual acquisition status in the confirmation data, and outputs the prediction consistency result. The link continuity discriminator reads the link masking status from the shared timing feature, checking for interruptions, jumps, inversions, or missing links in the order of instruction reception, agent response, policy execution, and result receipt, and outputs the link continuity result. The discriminator result aggregation unit writes the three types of discriminator results into the discriminator result set according to the acquisition time and field order, which is used to generate an operation and maintenance anomaly score.

[0125] In this embodiment, the step of generating a set of operation and maintenance status offset objects, constructing a cloud-edge dual-domain managed causal chain break graph, and establishing managed causal association edges includes:

[0126] Read the operation and maintenance status data, operation and maintenance anomaly scores, and the state transition relationship between stable management status, critical management status, and unstable management status. The collection segment from stable management status to critical management status is determined as the critical offset segment, and the collection segment from critical management status to unstable management status is determined as the unstable offset segment.

[0127] Read the communication link status, agent response status, configuration status, version status, container running status, interface call status, resource usage status, log backlog status, policy queue status, and receipt feedback status within the critical offset section and unstable offset section. Convert the status fields that deviate from the standard operation and maintenance image into corresponding offset objects, generating a set of operation and maintenance status offset objects. Specifically, the generation of the operation and maintenance status offset object set is as follows:

[0128] First, for each sampled row in the critical offset section and the unstable offset section, read the communication link status, agent response status, configuration status, version status, container running status, interface call status, resource usage status, log accumulation status, policy queue status and receipt feedback status field by field, and compare them one by one with the baseline values ​​of the corresponding fields in the standard operation and maintenance image.

[0129] If the current value of a numeric field exceeds the baseline mean plus or minus twice the baseline standard deviation or is below the zero tolerance threshold, the field is written to the numeric deviation flag. If the current state of an enumerated field is different from the baseline state, the field is written to the enumeration deviation flag.

[0130] Offset object encapsulation is performed according to field type: the deviation between communication link status and policy queue status is written to the communication class offset object, the deviation of proxy response status is written to the proxy class offset object, the deviation of configuration status is written to the configuration class offset object, the deviation of version status is written to the version class offset object, the deviation of container running status is written to the container class offset object, the deviation of interface call status is written to the interface class offset object, the deviation of resource usage status is written to the resource class offset object, the deviation of log accumulation status is written to the log class offset object, and the deviation of receipt feedback status is written to the receipt class offset object.

[0131] Each offset object records the field name, sampling time, deviation magnitude, duration, and segment label. After traversing all sampling rows, duplicate offset objects of the same type that appear continuously are merged according to the field name and segment label. The offset objects are sorted in the order of sampling start time, deviation magnitude, and duration. The final output is a set of operation and maintenance status offset objects grouped by field category, with each group containing several specific offset instances.

[0132] Following the management sequence of cloud policy generation, edge policy forwarding, terminal policy execution, and execution result feedback, a cloud management domain, an edge forwarding domain, and a terminal execution domain are constructed respectively. The construction of the cloud management domain, edge forwarding domain, and terminal execution domain is as follows:

[0133] Create a management topology initialization table and allocate an independent node area for each domain; read the management order of cloud policy generation, edge policy forwarding, terminal policy execution, and execution result feedback, write the cloud management domain into the first area, the edge forwarding domain into the second area, and the terminal execution domain into the last area; generate four nodes in the cloud management domain area in sequence: policy generation node, configuration image node, version baseline node, and result confirmation node, and write a node number, domain identifier, and function label for each node;

[0134] Four nodes are generated sequentially in the edge forwarding domain: edge agent node, channel scheduling node, edge cache node, and policy queue node. The node number, domain identifier, and function label are also written in the same way.

[0135] In the terminal execution domain area, six nodes are generated in sequence: terminal operation and maintenance agent node, communication component node, configuration loading node, version adaptation node, business container node, and execution receipt node. The node number, domain identifier, and function label are then filled in.

[0136] After node generation is completed, a one-way sequential connection is established within the cloud management domain based on policy generation, configuration image, version baseline, and result confirmation. Within the edge forwarding domain, a one-way sequential connection is established based on edge proxy, channel scheduling, edge caching, and policy queue. Within the terminal execution domain, a one-way sequential connection is established based on terminal operation and maintenance agent, communication component, configuration loading, version adaptation, business container, and execution receipt.

[0137] Write cross-domain exit and entry tags in the order of management between the three domains: connect the policy generation node exit to the edge agent node entry, connect the policy queue node exit to the terminal operation and maintenance agent node entry, and connect the execution receipt node exit to the result confirmation node entry.

[0138] The cloud management domain includes policy generation nodes, configuration image nodes, version baseline nodes, and result confirmation nodes; the edge forwarding domain includes edge proxy nodes, channel scheduling nodes, edge cache nodes, and policy queue nodes; and the terminal execution domain includes terminal operation and maintenance proxy nodes, communication component nodes, configuration loading nodes, version adaptation nodes, business container nodes, and execution receipt nodes.

[0139] According to the direction of operation and maintenance strategy distribution, the strategy generation node, edge agent node, channel scheduling node, terminal operation and maintenance agent node, configuration loading node, version adaptation node and business container node are connected in sequence to form an instruction distribution chain. According to the direction of execution result feedback, the business container node, execution receipt node, terminal operation and maintenance agent node, edge cache node, edge agent node and result confirmation node are connected in sequence to form a status feedback chain. The instruction distribution chain and status feedback chain are combined to form a cloud-edge dual-domain managed causal chain break diagram.

[0140] Each offset object in the set of operation and maintenance status offset objects is attached to the corresponding node in the cloud-edge dual-domain managed causal chain diagram. Based on the influence relationship of each offset object on the instruction issuance chain, terminal local execution chain and status feedback chain, managed causal association edges are established. The managed causal association edges include instruction reachable edges for recording policy arrival status, execution acceptance edges for recording terminal local execution transmission status, status feedback edges for recording execution result feedback status, and failure masking edges for recording anomaly masking relationships.

[0141] In this embodiment, the offset object that causes the cloud policy to fail to reach the intelligent converged terminal, the intelligent converged terminal to fail to execute the policy, or the execution result to fail to be sent back to the cloud is identified as the root cause object of the management chain break, including:

[0142] Based on the type of offset object, the communication-type offset objects in the operation and maintenance status offset object set are mapped to the channel scheduling node and the communication component node; the agent-type offset objects are mapped to the edge agent node and the terminal operation and maintenance agent node; the configuration-type offset objects are mapped to the configuration image node and the configuration loading node; the version-type offset objects are mapped to the version baseline node and the version adaptation node; the container-type offset objects are mapped to the business container node; the policy queue-type offset objects are mapped to the policy queue node; and the receipt-type offset objects are mapped to the execution receipt node and the result confirmation node.

[0143] Based on the mapping position of each offset object in the cloud-edge dual-domain managed causal chain break diagram, the managed causal association edges associated with each offset object are read. Then, according to the cloud policy's distribution direction and execution result feedback direction, the managed causal association edges are divided into instruction arrival verification edges, execution acceptance verification edges, status feedback verification edges, and failure masking verification edges. Specifically, the division of managed causal association edges into instruction arrival verification edges, execution acceptance verification edges, status feedback verification edges, and failure masking verification edges is as follows:

[0144] In the edge table of the broken link graph, write the starting point number, ending point number, starting point domain identifier, ending point domain identifier, and link direction label for each managed causal relationship edge; scan the edge table, write edges with link directions labeled as cloud or edge into temporary set A, edges with link directions labeled as edge or terminal into temporary set B, edges with link directions labeled as terminal or edge into temporary set C, and edges with link directions labeled as edge or cloud into temporary set D; then subdivide the four temporary sets according to node function labels: in set A, retain edges whose starting point is a policy generation node and whose ending point is an edge proxy node, and record them as instruction arrival verification edges; In set B, edges whose starting point is an edge proxy node or channel scheduling node and whose ending point is a terminal operation and maintenance proxy node, configuration loading node, version adaptation node, or business container node are retained and denoted as execution acceptance verification edges. In set C, edges whose starting point is an execution receipt node and whose ending point is an edge cache node or edge proxy node are retained and then merged with edges in set D whose starting point is an edge proxy node and whose ending point is a result confirmation node. The merged parts are denoted as status feedback verification edges. Any remaining edges whose starting point or ending point contains an abnormal occlusion marker or whose link direction is reversed from the expected direction and causes the information flow to be covered are uniformly classified as failure masking verification edges.

[0145] Perform chain break verification on the instruction arrival verification edge and execution acceptance verification edge. If the cloud policy cannot reach the corresponding execution node through the policy generation node, edge agent node, channel scheduling node and terminal operation and maintenance agent node, or the terminal operation and maintenance agent node cannot pass the cloud policy to the configuration loading node, version adaptation node or business container node, then mark the corresponding offset object as a candidate object for chain break management.

[0146] Perform chain break verification on the status feedback verification edge and the failure masking verification edge. If the execution result of the action segment cannot be fed back to the result confirmation node through the execution receipt node, terminal operation and maintenance agent node, edge cache node and edge agent node, or if the execution result reported by the terminal is inconsistent with the edge observation result, then mark the corresponding offset object as a feedback chain break candidate object or a failure masking candidate object.

[0147] Perform association merging on candidate objects for managed chain break, candidate objects for backhaul chain break, and candidate objects for failure masking. The offset objects that can cause cloud policies to fail to reach smart converged terminals, smart converged terminals to fail to execute policies, or execution results to fail to be backhauled to the cloud are identified as the root cause objects of managed chain break.

[0148] In this embodiment, the edge operation and maintenance node issues a management and keep-alive capsule to the intelligent converged terminal. After establishing the keep-alive heartbeat and receipt token, the reversible repair strategy is broken down into multiple action segments and executed sequentially, including:

[0149] Based on the object type, graph node, and associated causal edges of the managed broken link root cause object, a corresponding reversible repair strategy is generated. This reversible repair strategy includes communication recovery strategy, proxy reconnection strategy, configuration resending strategy, version adaptation strategy, strategy queue unblocking strategy, and receipt link recovery strategy. Specifically, the generation of the corresponding reversible repair strategy is as follows:

[0150] The node number and function tag corresponding to the managed link break root cause object are read from the link break diagram node table. If the function tag is a communication component or channel scheduling, it is written to the communication recovery policy queue; if it is an edge agent or terminal operation and maintenance agent, it is written to the agent reconnection policy queue; if it is a configuration loading or configuration image, it is written to the configuration reissue policy queue; if it is a version adaptation or version baseline, it is written to the version adaptation policy queue; if it is a policy queue node, it is written to the policy queue unblocking policy queue; if it is an execution receipt or result confirmation node, it is written to the receipt link recovery policy queue. Subsequently, an action list is generated for each policy queue. The communication recovery action list sequentially includes link parameter reset, channel switching, heartbeat cycle shortening, and link log capture. The agent reconnection action list sequentially includes... The action list includes: agent process keep-alive, connection reconstruction, agent receipt verification, and reconnection count. The action list for configuration resending includes: configuration version verification, difference resending, configuration loading verification, and configuration rollback point generation. The action list for version adaptation includes: dependency component verification, missing package retransmission, canary upgrade, and version receipt verification. The action list for policy queue unblocking includes: mutual exclusion task suspension, blocked task isolation, queue reordering, and queue capacity reset. The action list for receipt link recovery includes: receipt token reconstruction, receipt timeout reset, edge cache transfer, and receipt integrity verification. Finally, the action lists are concatenated and written into the reversible repair policy in the order of communication recovery, agent reconnection, configuration resending, version adaptation, policy queue unblocking, and receipt link recovery.

[0151] A managed keep-alive capsule is issued to the intelligent converged terminal. The managed keep-alive capsule includes a keep-alive heartbeat field, a receipt token field, a status snapshot field, a breakpoint resume field, and a rollback trigger field. The generation of the managed keep-alive capsule is specifically as follows:

[0152] Create a capsule data frame and write a unique capsule number; write the minimum heartbeat interval, link detection flag, and maximum packet loss count in the keep-alive heartbeat field; generate a 128-bit random token in the receipt token field and record the token expiration time; copy the current communication link status, agent response status, configuration version, application version, container running status, policy queue status, and receipt status field by field in the status snapshot field; write action segment index 0 and completed flag 0 in the breakpoint resume field; write the communication interruption threshold, agent no receipt threshold, configuration verification failure flag, version incompatibility flag, container continuous exit threshold, and queue blocking threshold in the rollback trigger field; finally, calculate the entire frame verification value and encapsulate it into a managed keep-alive capsule;

[0153] After establishing the minimum management heartbeat based on the keep-alive heartbeat field, establishing the action segment receipt identifier based on the receipt token field, and recording the pre-repair maintenance status based on the status snapshot field on the intelligent converged terminal, the management keep-alive capsule is confirmed to be established. Specifically, establishing the minimum management heartbeat based on the keep-alive heartbeat field on the intelligent converged terminal is as follows:

[0154] Parse the keep-alive heartbeat field, read the minimum heartbeat interval, link detection flag, and maximum packet loss count, and write these three parameters into the terminal operation and maintenance agent; enable the heartbeat timer, set the timer period to the minimum heartbeat interval, and generate a heartbeat message containing the current capsule number and sending sequence number each time it is triggered, and send it to the edge operation and maintenance node through the edge agent; after receiving the heartbeat message, the edge operation and maintenance node returns a heartbeat acknowledgment message, and the terminal operation and maintenance agent clears the packet loss count upon receiving it; if the link detection flag is one and no acknowledgment message is received consecutively, the packet loss count is incremented; when the packet loss count reaches the maximum packet loss count threshold, a link interruption flag is written and a failure fallback segment is triggered; if the threshold is not reached, heartbeat messages continue to be sent; thus, the minimum managed heartbeat is established and maintained.

[0155] The reversible repair strategy is broken down into a survival anchoring segment, a chain break removal segment, a root cause repair segment, a state locking segment, and a failure rollback segment, and is executed in the following order: survival anchoring segment, chain break removal segment, root cause repair segment, and state locking segment. Specifically, the reversible repair strategy is broken down into the following segments: survival anchoring segment, chain break removal segment, root cause repair segment, state locking segment, and failure rollback segment.

[0156] First, read the action list within the reversible repair strategy and generate a segment index table according to action type and execution order. Write the keep-alive anchor segment in the first line of the segment index table and load the three actions from the communication recovery action: link parameter reset, channel switching, and heartbeat interval shortening. Write the link breakage removal segment in the second line and load the two actions from the proxy reconnection action: process keep-alive, connection reconstruction, and queue clearing action: mutual exclusion task suspension and blocking isolation. Write the root cause repair segment in the third line and load the main repair action from the root cause corresponding strategy. Write the state lock segment in the fourth line and load the loading verification, rollback point generation, and version receipt verification from the configuration reissue action. Write the failure rollback segment in the fifth line and load the rollback command set for all actions. After filling the segment index table, output the action instruction sequence in the order of keep-alive anchor segment, link breakage removal segment, root cause repair segment, and state lock segment, and write the failure rollback segment separately into the rollback queue.

[0157] After each action segment is completed, read the operation and maintenance status data and the action segment execution result. When the keep-alive heartbeat, receipt token, configuration verification, version adaptation and policy queue status all meet the conditions for the next action segment execution, issue the next action segment. When any status triggers the rollback condition recorded in the rollback trigger field, stop the action segment and execute the failure rollback segment.

[0158] Example 1: In a continuous cloud-edge collaborative operation and maintenance cycle, the cloud operation and maintenance platform manages 6 edge operation and maintenance nodes, each of which connects to 40 intelligent converged terminals, for a total of 240 terminals. The system first reads the terminal operation and maintenance attribute data of this batch of terminals, including terminal identification attributes, communication access attributes, edge affiliation attributes, service operation attributes, configuration baseline attributes, version baseline attributes, component dependency attributes, and remote management attributes, and generates a standard operation and maintenance image for each terminal. Taking one of the terminals as an example, its standard heartbeat interval is 5s, normal link latency is 20-90ms, edge agent response time is 40-180ms, policy queue length is 0-3 entries, acknowledgment timeout count is 0, configuration digest is "C8F21A", application version is 3.2.6, operation and maintenance agent version is 2.8.1, and communication component version is 1.9.4.

[0159] During the model training phase, the cloud extracted 96,000 sample data points from historical stable operation and maintenance records, including 78,600 stable managed samples, 11,400 critical managed samples, and 6,000 unstable managed samples. Each sample was arranged in a fixed order according to the following fields: communication link status, agent response status, configuration status, version status, container running status, interface call status, resource usage status, log backlog status, policy queue status, and receipt feedback status. In the stable managed samples, the average heartbeat interval was 5.2s, the average link latency was 54ms, the average agent response time was 93ms, and the average policy queue length was 1.1 entries. In the unstable managed samples, the average heartbeat interval was 9.8s, the average link latency was 211ms, the average agent response time was 742ms, the average policy queue length was 8.4 entries, and the average number of receipt timeouts was 3.7. After training is complete, the cloud will distribute the parameters of the MAD-GAN model with a dual generator-multi-task discriminator structure to the edge operation and maintenance nodes.

[0160] In actual simulation, edge operation and maintenance nodes continuously collected operation and maintenance status data from a smart converged terminal at a 5-second collection cycle. For the first 10 collection cycles, the terminal remained stable, with heartbeat intervals ranging from 5.0 to 5.4 seconds, link latency from 46 to 68 milliseconds, agent response times from 82 to 119 milliseconds, policy queue lengths of 1 to 2 entries, and zero acknowledgment timeouts. However, starting from the 11th collection cycle, abnormal changes occurred: the heartbeat interval increased from 5.4 seconds to 6.9 seconds, link latency from 68 milliseconds to 137 milliseconds, agent response time from 119 milliseconds to 386 milliseconds, and policy queue length from 2 to 5 entries. In the 16th collection cycle, agent response time reached 721 milliseconds, policy queue length reached 8 entries, and the cumulative number of acknowledgment timeouts reached 3. Simultaneously, the configuration digest changed from C8F21A to C8F29D, but the version dependencies remained consistent.

[0161] The edge maintenance node determines the 11th acquisition cycle as the phase boundary start anchor point and the 16th acquisition cycle as the phase boundary confirmation anchor point. It extracts 10 acquisition cycles forward as stable support data, extracts acquisition cycles 11 to 16 as transition data, and then extracts 5 acquisition cycles backward as confirmation data, combining these to obtain a terminal maintenance phase boundary sliding window containing 21 rows of data. Subsequently, the system generates a terminal maintenance time series matrix with 21 rows and 10 columns, based on a fixed field order. For the communication link status field, 0 is written for normal operation, 1 for delay, 2 for packet loss, and 3 for link failure. Numerical fields such as proxy response, link latency, queue length, and receipt timeout are all normalized according to the baseline mean and standard deviation in the standard maintenance image. After processing, the normalized mean of the agent response status column was 0.18 in the stable support segment, rising to 2.46 in the transition segment and 3.21 in the confirmation segment; the strategy queue status column was 0.11 in the stable support segment, rising to 2.88 in the transition segment and 3.64 in the confirmation segment.

[0162] After the terminal operation and maintenance time series matrix is ​​input into the MAD-GAN model, the managed link masking layer first marks the communication link status and policy queue status as instruction receiving link fields, the proxy response status as proxy response link fields, the configuration status, version status, container running status, interface call status, resource usage status, and log accumulation status as policy execution link fields, and the receipt feedback status and communication link status as result receipt link fields. The refactoring generator generates a normal managed refactoring sequence based on the stability support data. The average value of the refactored proxy response status column is 0.24, while the average value of the corresponding column in the original confirmation segment is 3.21, with an average refactoring deviation of 2.97. The average value of the refactored policy queue status column is 0.17, while the average value of the original confirmation segment is 3.64, with an average refactoring deviation of 3.47. The prediction generator generates a phase boundary prediction sequence based on the stability support data and transition data. Its average prediction deviation for the confirmation segment proxy response status is 0.42, for the policy queue status is 0.39, and for the receipt feedback status is 0.51. The multi-task discriminator outputs a reconstruction consistency result of 0.31, a prediction consistency result of 0.68, a link continuity result of 0.27, and a comprehensive maintenance anomaly score of 0.86. Since the instability management threshold is set to 0.80, the terminal is identified as being in an unstable management state.

[0163] The system then reads the field status in the critical offset section and the unstable offset section, compares it field by field with the standard operation and maintenance image, and generates a set of operation and maintenance status offset objects. Among them, the agent-type offset objects record that the agent response time increased from 119ms to 721ms, lasting for 6 collection cycles; the policy queue-type offset objects record that the queue length increased from 2 to 8, lasting for 7 collection cycles; the receipt-type offset objects record that the receipt timeout count increased from 0 to 3; and the configuration-type offset objects record that the configuration summary is inconsistent, lasting for 2 collection cycles. After the system constructs the cloud-edge dual-domain managed cause-effect chain diagram, it attaches the agent-type offset objects to the edge agent node and the terminal operation and maintenance agent node, attaches the policy queue-type offset objects to the policy queue node, attaches the receipt-type offset objects to the execution receipt node and the result confirmation node, and attaches the configuration-type offset objects to the configuration image node and the configuration loading node.

[0164] During the link failure verification, a configuration verification policy was issued from the cloud. The policy generation node took 12ms to generate, and the edge agent node took 48ms to reach it. However, after entering the policy queue node, the waiting time reached 136 seconds, failing to reach the terminal operation and maintenance agent node within the preset execution window. The system then read the receipt link and found two pending receipts at the terminal execution receipt node and four backlogged receipt records at the edge cache node, confirming that the node did not receive the complete execution result. The configuration-type offset object was verified not to have caused the communication components and operation and maintenance agent to fail, and was therefore marked as an accompanying abnormal object. Ultimately, the agent-type offset object, the policy queue-type offset object, and the receipt-type offset object were identified as the root cause objects of the managed link failure.

[0165] The cloud generates a reversible repair strategy based on the root cause object. The action list includes agent process keep-alive, connection reconstruction, mutual exclusion task suspension, blocking task isolation, queue reordering, receipt token reconstruction, and receipt integrity verification. Edge maintenance nodes generate a managed keep-alive capsule, writing the capsule number, minimum heartbeat interval, maximum packet loss count, receipt token, status snapshot, breakpoint continuation field, and rollback trigger field. After receiving the capsule, the terminal maintenance agent sends six consecutive heartbeat messages, all of which receive edge confirmation messages, with an average confirmation latency of 76ms, indicating successful establishment of the minimum managed heartbeat. Subsequently, the keep-alive anchoring segment resets the link parameters, and the disconnection stripping segment suspends two mutual exclusion tasks and moves one blocking task to the isolation queue, reducing the policy queue length from eight to three. After the root cause repair segment completes agent connection reconstruction, the agent response time decreases from 721ms to 146ms, and the number of receipt timeouts decreases from three to zero. During the state-locked segment, the queue, receipt, and agent status were re-verified. The terminal maintained an inner beat interval of 5.1–5.5s and a link latency of 52–79ms for the next 8 collection cycles. The operation and maintenance anomaly score dropped to 0.32, and the terminal was re-determined to be in a stable managed state.

[0166] In the comparative experiment, the system selected 1,200 abnormal sliding windows from the same batch, including 780 concurrent sliding windows with multiple abnormalities, and processed them using the traditional fixed threshold operation and maintenance method and the method of the present invention, respectively.

[0167] The accuracy rate of state recognition using traditional methods is 78.4%, while that of this invention is 93.6%.

[0168] The accuracy rate of traditional methods in locating the root cause of chain breakage is 64.2%, while that of this invention is 91.8%.

[0169] Traditional methods take an average of 145 seconds to detect anomalies, while this invention takes only 48 seconds.

[0170] Traditional methods trigger 86 false restarts, while this invention only triggers 17.

[0171] The traditional method repeats the strategy 132 times, while this invention repeats it 24 times.

[0172] Traditional methods experienced 39 instances of brief disconnection during the repair process, while this invention only experienced 5.

[0173] The success rate of the first repair using traditional methods is 71.5%, while that of this invention is 89.7%.

[0174] Traditional methods take an average of 312 seconds to recover to a stable state, while this invention takes 168 seconds.

[0175] 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 cloud-edge collaborative intelligent converged terminal operation and maintenance management method, characterized in that, include: The cloud-based operations and maintenance platform reads the terminal operations and maintenance attribute data corresponding to the intelligent converged terminal, generates the corresponding standard operations and maintenance image, and distributes the standard operations and maintenance image to the corresponding edge operations and maintenance node. Collect operation and maintenance status data of intelligent converged terminals, construct terminal operation and maintenance phase boundary sliding window with multiple consecutive collection cycles, and convert the operation and maintenance status data into terminal operation and maintenance time series matrix according to fixed field order; Input the terminal operation and maintenance time series matrix into the MAD-GAN model with dual generator-multi-task discriminator structure, calculate the operation and maintenance anomaly score, and determine the stable management state, critical management state or unstable management state of the intelligent fusion terminal based on the operation and maintenance anomaly score. Based on the operation and maintenance status data, operation and maintenance anomaly scores, and the state transition relationships between stable management status, critical management status, and unstable management status, a set of operation and maintenance status offset objects is generated, a cloud-edge dual-domain management causal chain break graph is constructed, and management causal association edges are established. Map the set of offset objects in the operation and maintenance status to the cloud-edge dual-domain managed cause-effect chain breakage diagram, perform chain breakage verification on the managed cause-effect association edge, and identify the offset objects that cause cloud policies to be unable to reach intelligent converged terminals, intelligent converged terminals to be unable to execute policies, or execution results to be unable to be returned to the cloud as the root cause objects of the managed chain breakage. The cloud-based operations and maintenance platform generates a reversible repair strategy based on the root cause of the chain breakage. The edge operations and maintenance nodes issue the management and survival capsule to the intelligent converged terminal. After establishing the survival heartbeat and receipt token, the reversible repair strategy is broken down into multiple action segments and executed sequentially. The execution of the next action segment is controlled or the failure rollback segment is triggered based on the operations and maintenance status data.

2. The intelligent converged terminal operation and maintenance management method based on cloud-edge collaboration according to claim 1, characterized in that, The terminal operation and maintenance attribute data includes terminal identification attributes, communication access attributes, edge affiliation attributes, service operation attributes, configuration baseline attributes, version baseline attributes, component dependency attributes, and remote management attributes.

3. The intelligent converged terminal operation and maintenance management method based on cloud-edge collaboration according to claim 1, characterized in that, The generation of the corresponding standard operation and maintenance image includes: The mirror index key is determined by matching the terminal identification attribute with the communication access attribute, edge affiliation attribute, and service operation attribute. Based on the configuration baseline attribute, version baseline attribute, and component dependency attribute, write the communication baseline field, configuration baseline field, version baseline field, proxy status baseline field, container status baseline field, and resource usage baseline field; Write the policy queue baseline field and receipt status baseline field according to the remote management attributes; Perform integrity checks on all baseline fields and generate mirror hash values; Generate an image version number and store it as a standard operation and maintenance image.

4. The intelligent converged terminal operation and maintenance management method based on cloud-edge collaboration according to claim 1, characterized in that, The construction of the terminal operation and maintenance phase boundary sliding window, which converts the operation and maintenance status data into a terminal operation and maintenance time series matrix according to a fixed field order, includes: Edge operation and maintenance nodes collect operation and maintenance status data of intelligent converged terminals according to a preset collection cycle, and write the operation and maintenance status data at each collection moment into the time-series buffer. The operation and maintenance status data includes communication link status, agent response status, configuration status, version status, container running status, interface call status, resource usage status, log accumulation status, policy queue status, and receipt feedback status. Align the operation and maintenance status data of each collection time in the time-series buffer with the standard operation and maintenance image to generate a management continuity flag and a status transition flag corresponding to each collection time. Read the state transition markers in the order of acquisition time, and determine the acquisition time when the first interruption, delay, blockage or acknowledgment mismatch of the management continuity marker is determined as the phase boundary start anchor point. The acquisition time when the instruction reception abnormality, agent response abnormality, strategy execution abnormality or result acknowledgment abnormality is confirmed in the current acquisition cycle is determined as the phase boundary confirmation anchor point. Using the phase boundary start anchor point and the phase boundary confirmation anchor point as boundaries, the stable support data before the phase boundary start anchor point, the transition data between the phase boundary start anchor point and the phase boundary confirmation anchor point, and the confirmation data after the phase boundary confirmation anchor point are extracted from the time-series buffer and combined to generate the terminal operation and maintenance phase boundary sliding window. Following a fixed field order of communication link status, agent response status, configuration status, version status, container running status, interface call status, resource usage status, log accumulation status, policy queue status, and receipt feedback status, the operation and maintenance status data in the terminal operation and maintenance phase window is written into the matrix row time by time to generate the terminal operation and maintenance time series matrix.

5. The intelligent converged terminal operation and maintenance management method based on cloud-edge collaboration according to claim 1, characterized in that, The calculation of the operation and maintenance anomaly score determines the stable management status, critical management status, or unstable management status of the intelligent fusion terminal based on the operation and maintenance anomaly score, including: The terminal operation and maintenance time series matrix is ​​input into the MAD-GAN model with a dual generator-multi-task discriminator structure. The MAD-GAN model includes a reconstruction generator, a prediction generator, a multi-task discriminator, and a managed link masking layer. The managed link masking layer marks the link fields corresponding to instruction reception, agent response, policy execution and result receipt based on the stability support data, transition data and confirmation data in the terminal operation and maintenance phase boundary sliding window, and generates the managed link masking sequence. The reconstructor performs link-keeping reconstruction on the terminal operation and maintenance timing matrix based on the managed link masking sequence, generates a normal managed reconstruction sequence, and generates reconstruction deviation results based on the field deviation status between the normal managed reconstruction sequence and the terminal operation and maintenance timing matrix. The prediction generator predicts the corresponding control state of the confirmation data based on the stable support data and transition data, generates a phase boundary prediction sequence, and generates a prediction deviation result based on the state deviation between the phase boundary prediction sequence and the confirmation data. The multi-task discriminator performs reconstruction consistency discrimination on normal managed reconstruction sequences, prediction consistency discrimination on phase boundary prediction sequences, and link continuity discrimination on managed link masking sequences, and outputs reconstruction consistency results, prediction consistency results, and link continuity results. An operation and maintenance anomaly score is generated based on the reconstruction deviation result, prediction deviation result, reconstruction consistency result, prediction consistency result, and link continuity result. The operation and maintenance anomaly score is compared with the corresponding stable management threshold, critical management threshold, and unstable management threshold in the standard operation and maintenance image. Intelligent converged terminals below the stable management threshold are identified as stable management states, intelligent converged terminals that reach the critical management threshold but do not reach the unstable management threshold are identified as critical management states, and intelligent converged terminals that reach the unstable management threshold are identified as unstable management states.

6. The intelligent converged terminal operation and maintenance management method based on cloud-edge collaboration according to claim 1, characterized in that, The generated set of operation and maintenance status offset objects is used to construct a cloud-edge dual-domain managed causal chain break graph and establish managed causal association edges, including: Read the operation and maintenance status data, operation and maintenance anomaly scores, and the state transition relationship between stable management status, critical management status, and unstable management status. The collection segment from stable management status to critical management status is determined as the critical offset segment, and the collection segment from critical management status to unstable management status is determined as the unstable offset segment. Read the communication link status, agent response status, configuration status, version status, container running status, interface call status, resource usage status, log accumulation status, policy queue status, and receipt feedback status within the critical offset section and unstable offset section. Convert the status fields that deviate from the standard operation and maintenance image into corresponding offset objects and generate a set of operation and maintenance status offset objects. Following the management sequence of cloud policy generation, edge policy forwarding, terminal policy execution, and execution result feedback, cloud management domain, edge forwarding domain, and terminal execution domain are constructed respectively. According to the direction of operation and maintenance strategy distribution, the strategy generation node, edge agent node, channel scheduling node, terminal operation and maintenance agent node, configuration loading node, version adaptation node and business container node are connected in sequence to form an instruction distribution chain. According to the direction of execution result feedback, the business container node, execution receipt node, terminal operation and maintenance agent node, edge cache node, edge agent node and result confirmation node are connected in sequence to form a status feedback chain. The instruction distribution chain and status feedback chain are combined to form a cloud-edge dual-domain managed causal chain break diagram. Each offset object in the set of operation and maintenance status offset objects is attached to the corresponding node in the cloud-edge dual-domain management causal chain diagram, and management causal association edges are established according to the influence relationship of each offset object on the instruction issuance chain, terminal local execution chain and status feedback chain.

7. The intelligent converged terminal operation and maintenance management method based on cloud-edge collaboration according to claim 1, characterized in that, The offset objects that cause cloud policies to fail to reach intelligent converged terminals, intelligent converged terminals to fail to execute policies, or execution results to fail to be returned to the cloud are identified as root cause objects of management chain breakage, including: Based on the type of offset object, the communication-type offset objects in the operation and maintenance status offset object set are mapped to the channel scheduling node and the communication component node; the agent-type offset objects are mapped to the edge agent node and the terminal operation and maintenance agent node; the configuration-type offset objects are mapped to the configuration image node and the configuration loading node; the version-type offset objects are mapped to the version baseline node and the version adaptation node; the container-type offset objects are mapped to the business container node; the policy queue-type offset objects are mapped to the policy queue node; and the receipt-type offset objects are mapped to the execution receipt node and the result confirmation node. Based on the mapping position of each offset object in the cloud-edge dual-domain managed causal chain break diagram, read the managed causal association edges associated with each offset object, and divide the managed causal association edges into instruction arrival verification edges, execution acceptance verification edges, status return verification edges, and failure masking verification edges according to the cloud policy distribution direction and execution result return direction. Perform chain break verification on the instruction arrival verification edge and execution acceptance verification edge. If the cloud policy cannot reach the corresponding execution node through the policy generation node, edge agent node, channel scheduling node and terminal operation and maintenance agent node, or the terminal operation and maintenance agent node cannot pass the cloud policy to the configuration loading node, version adaptation node or business container node, then mark the corresponding offset object as a candidate object for chain break management. Perform chain break verification on the status feedback verification edge and the failure masking verification edge. If the execution result of the action segment cannot be fed back to the result confirmation node through the execution receipt node, terminal operation and maintenance agent node, edge cache node and edge agent node, or if the execution result reported by the terminal is inconsistent with the edge observation result, then mark the corresponding offset object as a feedback chain break candidate object or a failure masking candidate object. Perform association merging on candidate objects for managed chain break, candidate objects for backhaul chain break, and candidate objects for failure masking. The offset objects that can cause cloud policies to fail to reach smart converged terminals, smart converged terminals to fail to execute policies, or execution results to fail to be backhauled to the cloud are identified as the root cause objects of managed chain break.

8. The intelligent converged terminal operation and maintenance management method based on cloud-edge collaboration according to claim 1, characterized in that, The edge operation and maintenance node issues a management and keep-alive capsule to the intelligent converged terminal. After establishing a keep-alive heartbeat and receipt token, the reversible repair strategy is broken down into multiple action segments and executed sequentially, including: Based on the object type, graph node, and associated causal relationship edge of the root cause of the chain break, a corresponding reversible repair strategy is generated. The reversible repair strategy includes communication recovery strategy, proxy reconnection strategy, configuration resending strategy, version adaptation strategy, strategy queue unblocking strategy, and receipt link recovery strategy. The management and survival capsule is issued to the intelligent converged terminal. The management and survival capsule includes a survival heartbeat field, a receipt token field, a status snapshot field, a breakpoint resume field, and a rollback trigger field. After establishing the minimum management heartbeat based on the keep-alive heartbeat field, establishing the action segment receipt identifier based on the receipt token field, and recording the operation and maintenance status before repair based on the status snapshot field in the intelligent fusion terminal, the management keep-alive capsule is confirmed to be established. The reversible repair strategy is divided into a keep-alive anchoring segment, a chain break removal segment, a root cause repair segment, a state lock segment, and a failure rollback segment, and is issued and executed in the order of the keep-alive anchoring segment, the chain break removal segment, the root cause repair segment, and the state lock segment. After each action segment is completed, read the operation and maintenance status data and the action segment execution result. When the keep-alive heartbeat, receipt token, configuration verification, version adaptation and policy queue status all meet the conditions for the next action segment execution, issue the next action segment. When any status triggers the rollback condition recorded in the rollback trigger field, stop the action segment and execute the failure rollback segment.

Images