Dynamic link mapping generation method based on metadata

Abstract

The invention relates to the technical field of data processing, in particular to a metadata-based dynamic link mapping generation method, which comprises the following steps of: acquiring distributed node metadata and historical track data; analyzing the static dependency relationship to generate a basic topology, predicting risk characteristics through a long short-term memory network, and merging to generate an enhanced topology model; concurrently executing conventional scene A * search, abnormal scene Monte Carlo tree search and a high-concurrency scene genetic algorithm based on the enhanced topology model, and outputting a path effectiveness index; dynamic weight factors are generated for data types of numerical value abnormity, time sequence fluctuation and protocol change through a streaming computation engine; constructing three-layer link mapping of a main service path, a degradation path and a pressure measurement path; and verifying node coverage and dependency transitivity, dynamically generating a score feedback optimization algorithm, injecting a flow test, and comparing performance data to position a service bottleneck node. The problems of insufficient topology adaptation, multi-protocol dependence inference lag and concurrent pressure measurement distortion caused by metadata staticization are solved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of data processing, in particular to a dynamic link mapping generation method based on metadata. BACKGROUND

[0002] As information describing data attributes and relationships, such as data types, source locations and timestamps, metadata provides a basic context for information systems; further, in a computer network environment, dynamic link mapping uses such metadata to update and optimize data paths in real time, such as automatically adjusting link topology by analyzing node state information in metadata to respond to network traffic changes or failover, thereby improving transmission efficiency and reliability, this process relies on a logical inference framework to continuously infer optimal connection strategies.

[0003] The existing dynamic link mapping generation based on metadata has the following technical pain points, specifically, the static metadata model is difficult to adapt to the dynamic topology changes of the distributed system, the algorithm cannot real-time infer the complex dependency relationship between multi-protocol nodes, for example, in the cross-cloud platform service invocation scenario, the metadata of the order processing microservice does not timely reflect the response time mutation of the downstream payment node, the test link mapping misses the abnormal path, and the concurrent stress test cannot simulate the real traffic peak, causing performance bottleneck diagnosis lag and stability misjudgment, the root cause lies in that the mapping generation relies on fixed metadata and cannot cope with environmental heterogeneity, resulting in monitoring data feedback delay and coverage blind area. SUMMARY

[0004] In view of the technical problems of the prior art, the present application provides a dynamic link mapping generation method based on metadata, which solves the technical problems of the prior art that the static metadata model is difficult to adapt to the dynamic topology changes of the distributed system and the algorithm cannot real-time infer the complex dependency relationship between multi-protocol nodes, causing the link mapping to miss the abnormal path in the cross-cloud platform service invocation scenario, and the concurrent stress test to be distorted, resulting in performance bottleneck diagnosis lag and system stability misjudgment.

[0005] To solve the above technical problems, the specific content of the present application is as follows: The dynamic link mapping generation method based on metadata provided by the present application comprises: Step 1, obtaining metadata and historical trajectory data of distributed nodes, the metadata comprising static dependency relationship, and the historical trajectory data comprising performance indicators; Step 2, parsing the static dependency relationship of step 1 to generate a basic topology, and inputting the historical trajectory data of step 1 into a long short-term memory network to predict risk features, and merging the basic topology and the risk features to generate an enhanced topology model; Step 3, based on the enhanced topology model of step 2, perform in parallel: the A* search algorithm is used to calculate the optimal path in the normal scenario, the Monte Carlo tree search is used to simulate faults in the abnormal scenario, and the genetic algorithm is used to schedule nodes in the high concurrency scenario, and the path validity index is output to the streaming computing engine; Step 4, receive node state events and path validity indexes of step 3 through the streaming computing engine, trigger isolated forest anomaly detection, Holt-Winters traffic prediction and reinforcement learning topology reconstruction algorithm for three data types of numerical anomaly, time fluctuation and protocol change, and generate dynamic weight factors; Step 5, use the dynamic weight factors of step 4 to construct three-layer link mapping: the improved Dijkstra algorithm is used to generate the main business path, the Monte Carlo simulation path library is used to match the degraded path, and the generative adversarial network is used to generate the stress test path; Step 6, verify the three-layer path node coverage and dependency delivery of step 5, dynamically generate path validity score and node stability score, use the path validity score as the update weight factor of the Monte Carlo tree search simulation fault in the abnormal scenario, and input step 3; use the node stability score as the update weight factor of the isolated forest anomaly detection and Holt-Winters traffic prediction in step 4, and input step 4, restart step 3 for the failed path, inject reference flow to the main path, inject fault test to the degraded path, inject chaos disturbance to the stress test path, and compare the performance data of the three-layer path to locate the service bottleneck node.

[0006] Further, the metadata-based dynamic link mapping generation method of the application, step 1 comprises: Collecting the basic attributes of periodically captured node ID, type and dependency relationship; Collecting runtime indicators such as message queue depth and thread pool state; Slicing the historical response time delay and error rate data according to time window, and outputting the time sequence feature vector to the long short-term memory network of step 2 to predict the risk characteristics.

[0007] Further, the metadata-based dynamic link mapping generation method of the application, step 2 comprises: Analyzing the static dependency relationship of step 1 to generate an adjacency list through a syntax parser, and the adjacency list stores the node dependency relationship; Encoding and labeling the risk hot area output by the long short-term memory network of step 2 on the corresponding nodes of the adjacency list; Merge the labeled adjacency list and the risk characteristics predicted by the long short-term memory network of step 2, and output the enhanced topology model.

[0008] Further, the metadata-based dynamic link mapping generation method of the application, step 3 comprises: Based on the enhanced topology model, the conventional scene performs A search calculation of the optimal path with the node weight factor as the heuristic function; The abnormal scene injects random fault to expand the Monte Carlo tree search space, and the high concurrency scene adopts a chromosome coding scheme to perform a node sequence genetic algorithm, records the calculation results of the conventional scene, the abnormal scene and the high concurrency scene, and outputs the path validity index to step 4.

[0009] Further, the metadata-based dynamic link mapping generation method of the present application, step 4 includes: Performing isolated forest detection on numerical anomaly data, and outputting abnormal node identification; Performing Holt-Winters prediction on time series fluctuation data, and outputting adjusted traffic trend parameters; Performing reinforcement learning strategy on protocol change data, and outputting a set of topology adjustment instructions; Merging the abnormal node identification, the adjusted traffic trend parameters and the set of topology adjustment instructions to generate a dynamic weight factor.

[0010] Further, the metadata-based dynamic link mapping generation method of the present application, step 5 includes: Converting the dynamic weight factor of step 4 into an edge cost, and performing an improved Dijkstra algorithm to output a shortest path sequence; Matching the simulation path library output by the Monte Carlo tree search of step 3 to the degraded path; Generating a final stress test path by screening a candidate path set through a discriminator of a generative adversarial network; Integrating the main business path, the degraded path and the stress test path to construct a three-layer link mapping.

[0011] Further, the metadata-based dynamic link mapping generation method of the present application, step 6 includes: Traversing the main business path, the degraded path and the stress test path node sequence constructed in step 5 by a main check flow to calculate a deep traversal hash value; Analyzing the node dependency relationship data output in step 5 by an auxiliary check flow to verify the cross-layer transmission closure relationship; When detecting a missing node coverage, triggering step 2 to perform incremental topology update.

[0012] Further, the metadata-based dynamic link mapping generation method of the present application, the dynamic scoring in step 6 includes: Calculating the difference between the path validity index of step 3 and the mapping result of step 5 to generate a path validity efficiency index; Extracting historical failure rate and recovery time data of step 1 to calculate a node stability score; The path efficiency index and node stability score are input into the step 3 feedback control loop to correct the node selection probability parameters of the step 3 genetic algorithm.

[0013] Further, the metadata-based dynamic link mapping generation method of the present application, the comparison of three-layer performance data in step 6 includes: Extract the main path, degraded path, and stress test path response delay data after injecting the reference flow, fault test, and chaos disturbance, and mark the nodes whose main path delay exceeds the degraded path by 200%; Establish the three-dimensional performance matrix of the main path, degraded path, and stress test path in terms of delay, error rate, and throughput; Perform spline interpolation calculation on the three-dimensional performance matrix to output the non-linear performance bottleneck node.

[0014] Further, the metadata-based dynamic link mapping generation method of the present application, in step 4 further includes: When protocol changes are detected in step 4, analyze the abnormal mode characteristics of the stress test path in step 5 to generate a set of topology reconstruction instructions; Update the action space of the step 4 reinforcement learning using the reconstruction instruction set to execute new topology reconstruction decisions.

[0015] Advantages of the present application; The present application solves the problem of static metadata model not being able to adapt to dynamic topology changes in distributed systems by enhancing the topology model to fuse static dependency relationships with dynamic risk features predicted by long short-term memory networks; uses a streaming computing engine to trigger isolated forest anomaly detection, Holt-Winters traffic prediction, and reinforcement learning topology reconstruction algorithms to generate dynamic weight factors according to different data types, realizes real-time inference of complex dependency relationships between nodes of multiple protocols; constructs a three-layer link mapping architecture of main business paths, degraded paths, and stress test paths to eliminate the risk of missing abnormal paths; updates the simulation weight of the Monte Carlo tree search and the node stability score to optimize the anomaly detection algorithm to form a closed loop control; combines three-layer performance comparison and spline interpolation bottleneck positioning of injected reference flow, fault test flow, and chaos disturbance flow to solve the problems of concurrent stress test distortion and diagnosis lag; when the protocol changes, the abnormal characteristics of the stress test path drive the update of the reinforcement learning action space to realize accurate judgment of system stability in cross-cloud platform service calling scenarios. BRIEF DESCRIPTION OF DRAWINGS

[0016] In order to more clearly illustrate the technical solutions of the present application, the following will briefly introduce the drawings needed in the embodiments. Obviously, for those skilled in the art, other drawings can also be obtained without creative labor based on the drawings.

[0017] Figure 1A flowchart of a metadata-based dynamic link mapping generation method provided by an embodiment of the present application is shown. DETAILED DESCRIPTION

[0018] In order to make the technical solutions of the present application clearer, the present application will be described in detail below with reference to specific embodiments of the present application and corresponding drawings. Obviously, the described embodiments are only some of the embodiments of the present application, not all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor fall within the scope of protection of the present application. The present application provided by each embodiment of the present application will be described in detail below with reference to the drawings. In order to better understand the purpose of the present application, the present application will be further described in detail below.

[0019] In a first aspect, the present application provides a metadata-based dynamic link mapping generation method, comprising: Step 1, obtaining metadata and historical trajectory data of distributed nodes, the metadata including static dependency relationships, and the historical trajectory data including performance indicators; Step 2, parsing the static dependency relationships of step 1 to generate a basic topology, and inputting the historical trajectory data of step 1 into a long short-term memory network to predict risk features, and combining the basic topology and the risk features to generate an enhanced topology model; Step 3, based on the enhanced topology model of step 2, performing in parallel: using A* search to calculate an optimal path in a regular scenario, using Monte Carlo tree search to simulate a fault in an abnormal scenario, and using a genetic algorithm to schedule nodes in a high-concurrency scenario, and outputting path validity indicators to a stream computing engine; Step 4, receiving node state events and the path validity indicators of step 3 through the stream computing engine, triggering isolated forest anomaly detection, Holt-Winters traffic prediction, and reinforcement learning topology reconstruction algorithms for three data types of numerical anomalies, time series fluctuations, and protocol changes, respectively, to generate dynamic weight factors; Step 5, using the dynamic weight factors of step 4 to construct a three-layer link mapping: using an improved Dijkstra algorithm to generate a main service path, matching a Monte Carlo simulation path library for a degraded path, and generating a test path through a generative adversarial network; Step 6, verify the three-layer path node coverage of step 5 with dependency transmission, dynamically generate path validity score and node stability score, use path validity score as an update weight factor for Monte Carlo tree search simulation failure in abnormal scenarios, and input step 3; use node stability score as an update weight factor for step 4 isolated forest anomaly detection and Holt-Winters traffic prediction, and input step 4, restart step 3 for verification failed path, inject reference traffic to main path, inject fault test to degraded path, inject chaos disturbance to stress test path, and compare three-layer path performance data to locate service bottleneck node.

[0020] Deploy lightweight monitoring agents to distributed nodes to periodically capture basic attributes such as node ID, type, and inter-service dependencies. Synchronously collect runtime performance indicators such as message queue depth and thread pool state to form a metadata set. Organize historical response latency and error rate data through time window slicing technology to generate time series feature vectors and transmit them to the long short-term memory network in step 2. This step provides a complete data source for static dependencies and dynamic performance indicators for subsequent analysis, realizing the data foundation for topology modeling and risk prediction.

[0021] The main channel uses a syntax parser to process the static dependencies of step 1, generating an adjacency list that stores node connection relationships. The auxiliary channel inputs the time series feature vectors of step 1 into the long short-term memory network to predict response timeout probability and fault correlation degree, among other risk features. Encode risk features as hot zone markers and label them on the corresponding node positions in the adjacency list. Merge the labeled adjacency list with the risk features output by the long short-term memory network to generate an enhanced topology model that includes potential fault hot zones. This step integrates static structure and dynamic risk prediction, breaking through the limitations of existing static topology.

[0022] Based on the enhanced topology model, perform path calculation by scenario: Normal scenario: use node weight factor as heuristic function to perform A* search algorithm to calculate optimal transmission path; Abnormal scenario: inject random node failure events to expand Monte Carlo tree search space to simulate fault transfer path; High concurrency scenario: use chromosome coding scheme to represent node sequence and optimize scheduling order through genetic algorithm.

[0023] Record the path validity indicators (such as latency and throughput) output in the three scenarios and transmit them to the streaming computing engine in step 4. This step simulates real environment complexity through differentiated algorithms to cover multi-dimensional performance challenges.

[0024] The streaming computing engine receives node state events and path validity indicators from step 3 in real time. Process data types separately: Numerical anomaly data: trigger isolated forest algorithm to detect abnormal nodes, output node identification; Timing fluctuation data: apply Holt-Winters model to adjust traffic trend parameters, output corrected parameters; Protocol change data: start reinforcement learning strategy to generate topology reconstruction instruction set.

[0025] Merge the abnormal node identification, traffic trend parameters and reconstruction instructions output by the three types of algorithms to generate a dynamic weight factor matrix. This step realizes real-time decision response in a multi-protocol environment and solves the problem of dynamic dependence inference lag.

[0026] The main service path converts the dynamic weight factor into edge value and executes the improved Dijkstra algorithm to output the shortest path sequence. The degraded path matches the simulation path library generated by the Monte Carlo tree search in step 3 to filter high-availability paths. The stress test path filters the candidate path set through the discriminator of the generative adversarial network to generate the extreme stress test path. The main service path, degraded path and stress test path are integrated to form a three-layer link mapping covering normal service, fault transfer and stress test. This hierarchical architecture eliminates the risk of missing abnormal paths and realizes full-scene coverage.

[0027] The main inspection flow traverses the three-layer path node sequence to calculate the depth traversal hash value to verify the integrity of the node. The auxiliary inspection flow analyzes the node dependence data to verify the cross-layer transmission closure relationship. If a node is missing, trigger the incremental topology update in step 2. Dynamically generate path efficiency score and node stability score, feed the path efficiency score back to step 3 to update the Monte Carlo tree search weight, and feed the node stability score back to step 4 to optimize the abnormal detection algorithm. Restart step 3 modeling for the verification failed path, and inject benchmark traffic, fault test traffic and chaos disturbance traffic into the three-layer path respectively. Compare the delay, error rate and throughput data of the three-layer path, and calculate the nonlinear performance bottleneck node through spline interpolation. This step forms a closed-loop control chain of simulation-verification-optimization, solving the problems of diagnosis lag and misjudgment.

[0028] When step 4 detects a protocol-level dependence change, analyze the abnormal mode characteristics of the stress test path in step 5 to generate a topology reconstruction instruction set. Update the action space of the reinforcement learning strategy using the instruction set and execute a new topology reconstruction decision. This mechanism realizes adaptive response to protocol dynamic changes and improves the system robustness in complex heterogeneous environments.

[0029] Specifically, the metadata-based dynamic link mapping generation method of the present application, step 1 includes: Collecting periodically captured node ID, type and dependence relationship basic attributes; Collecting runtime indicators such as message queue depth and thread pool state; The historical response latency and error rate data are sliced by time window, and a time sequence feature vector is output to the long short-term memory network in step 2 to predict the risk features.

[0030] In step 1, a lightweight monitoring agent is deployed to the distributed nodes, and a system application program interface is periodically called or a configuration file is read to capture node identifiers, node types such as microservices or database nodes, and inter-service dependencies. These basic attributes form the static dependency part of the metadata, providing structured data input for subsequent topology generation. The monitoring agent is configured to trigger data collection tasks at regular intervals to ensure coverage of key dimensions of node basic metadata, including upstream dependencies and downstream dependencies in the service call link.

[0031] Through log analysis tools or real-time monitoring systems, depth indicators of message queues such as queue length and backlog message quantity, and thread pool states including active thread count, idle thread count, and queue waiting time are collected. These runtime indicators constitute the performance indicator part of the historical trajectory data, reflecting the real-time load and resource utilization of the nodes. Message queue depth and thread pool state data are directly reported to the central storage through the agent, with timestamps recorded synchronously to support time series analysis and provide the original performance data set for dynamic risk assessment.

[0032] The historical response latency and error rate data are sliced by fixed or sliding time windows, and the average response latency and error rate change rate are calculated for each time window to generate a multi-dimensional time sequence feature vector. The window size is pre-defined, for example, five minutes or ten minutes, and normalization is performed on the data within the window to eliminate dimensional differences, outputting a feature vector containing latency mean, error rate fluctuation, and trend slope. This time sequence feature vector is transmitted to the long short-term memory network in step 2 as input to predict risk features such as node response timeout probability and failure correlation degree, completing the transition from raw data to prediction model.

[0033] Specifically, the metadata-based dynamic link mapping generation method of the present application, step 2 includes: Parse the static dependencies of step 1 to generate an adjacency list through a syntax parser, and the adjacency list stores node dependencies; Encode the risk hot zone output by the long short-term memory network in step 2 on the corresponding nodes of the adjacency list; Merge the annotated adjacency list with the risk features predicted by the long short-term memory network in step 2, and output an enhanced topology model.

[0034] In step 2, the static dependency data obtained in step 1 is parsed, and a syntax parser is used to process the data to generate an adjacency list structure; the adjacency list is indexed by node identifiers and stores direct dependency connection relationships between nodes to form a basic topology framework. The syntax parser is designed to process dependency relationship descriptions in a specific format, such as service call chain configuration files or API definition documents, and converts text or structured data into adjacency list representations through lexical analysis and syntax tree construction, ensuring accurate mapping and efficient storage of dependency relationships. This process provides a static structure basis for subsequent enhanced models, directly extracting key connection information from the metadata input in step 1, avoiding manual configuration errors.

[0035] The long short-term memory network receives the time series feature vector output in step 1 as input to predict dynamic risk features such as node response timeout probability and fault correlation degree; the prediction results are encoded into risk hot zone labels and marked on the corresponding node positions in the adjacency list to identify potential high-fault-risk areas. The encoding method uses binary or probability value mapping, such as assigning specific hot zone identifiers to high-risk nodes, and accurately positioning based on adjacency list node indexes during the marking process, so that static topology is integrated into dynamic risk assessment. This marking step combines historical performance data with real-time predictions to address the lack of adaptability of static models to dynamic environments, providing risk-aware input for path optimization.

[0036] The adjacency list after merging the marked risk hot zones and the risk feature data predicted by the long short-term memory network is output as an enhanced topology model; the merging operation is achieved through a data fusion algorithm, which aligns the node attributes of the marked adjacency list with the risk feature vectors and integrates them into a unified model containing static dependencies and dynamic risk weights. The enhanced topology model is output to the path calculation module in step 3 as the input basis for multi-scenario algorithm execution, realizing the transition from basic structure to risk enhancement and supporting the parallel processing of regular scenarios, abnormal scenarios, and high-concurrency scenarios. The logic relationship of the entire step 2 is to process static dependency parsing and dynamic risk prediction in parallel, and then build a comprehensive model through merging to form the input premise for step 3 algorithms.

[0037] Specifically, the dynamic link mapping generation method based on metadata of the present application, step 3 includes: Based on the enhanced topology model, the regular scenario performs A search calculation to find the optimal path using node weight factors as heuristic functions; The abnormal scenario injects random faults to expand the Monte Carlo tree search space, and the high-concurrency scenario uses a chromosome coding scheme to execute a node sequence genetic algorithm, records the calculation results of the regular scenario, the abnormal scenario, and the high-concurrency scenario, and outputs the path effectiveness indicators to step 4.

[0038] In step 3, based on the enhanced topology model output in step 2 as input, the path calculation operation of the normal scenario, the abnormal scenario and the high concurrency scenario is executed in parallel. The enhanced topology model contains static dependency relationship and dynamic risk characteristics, providing a unified input framework for the algorithms of different scenarios, and realizing the basic support of multi-environment path optimization. This step uses the data structure of the enhanced topology model to start the independent processing flow of the three scenarios, forming the initialization stage of algorithm execution.

[0039] The normal scenario adopts A search algorithm to calculate the optimal path, and uses the node weight factor as the heuristic function to guide the search direction. The node weight factor is derived from the dynamic risk characteristic weight in the enhanced topology model, and the A search process preferentially evaluates the path cost and heuristic estimate value to generate a path sequence that minimizes the transmission delay or resource consumption. The normal scenario execution output satisfies the optimal path scheme of normal service demand, providing a benchmark reference value for the subsequent path effectiveness indicators.

[0040] The abnormal scenario injects random fault events to expand the Monte Carlo tree search space, simulates the path transfer behavior under the condition of node failure or network interruption. The Monte Carlo tree search evaluates the path availability under the fault probability through random sampling and tree structure expansion, and generates a high-reliability fault transfer path library. The abnormal scenario processes the risk hot area data in the enhanced topology model, and outputs the simulation results as a candidate set of degraded paths, improving the robustness of the system in abnormal environment.

[0041] The high concurrency scenario applies genetic algorithm to schedule the node sequence, and uses chromosome coding scheme to represent the node scheduling order. The chromosome coding converts the node dependency relationship into gene sequence, and optimizes the node resource allocation under high load through selection, crossover and mutation operations. During the execution process of the genetic algorithm, the fitness function is adjusted according to the weight factor of the enhanced topology model, and the high-efficiency concurrent scheduling path is output, solving the performance bottleneck problem under peak traffic.

[0042] The calculation results of the normal scenario, the abnormal scenario and the high concurrency scenario are recorded, including path delay, throughput and error rate parameters, which are summarized as path effectiveness indicators. The path effectiveness indicators are output to the streaming computing engine of step 4 through standardized data format, as the input basis for dynamic weight factor generation. The logical relationship of the whole step 3 is embodied as from the enhanced topology model input to the parallel algorithm execution and then to the indicator output, completing the transition of path calculation to feedback loop.

[0043] Specifically, the metadata-based dynamic link mapping generation method of the present application, step 4 includes: Performing isolated forest detection on numerical abnormal data to output abnormal node identification; Performing Holt-Winters prediction on time series fluctuation data to output adjusted traffic trend parameters; The reinforcement learning strategy is performed on the protocol change data, and a topology adjustment instruction set is output. The abnormal node identifier, the adjusted traffic trend parameter, and the topology adjustment instruction set are merged to generate a dynamic weight factor.

[0044] In step 4, the stream computing engine receives the path validity indicators from step 3 and the real-time state events of the distributed nodes as input data; the path validity indicators include parameters such as path latency, throughput, and error rate, and the node state events include performance monitoring data and system event notifications. This step serves as the core of data processing, classifies and processes multiple heterogeneous data sources, provides basic input for dynamic weight factor generation, and connects the path calculation results of step 3 with the link mapping construction of step 5.

[0045] The isolated forest detection operation is performed on the numerical abnormal data, and the numerical abnormal data refers to the numerical values of node performance indicators such as message queue depth or thread pool state deviating from the normal range; the isolated forest detection algorithm analyzes these numerical points, identifies outliers by constructing a random tree structure, and outputs an abnormal node identifier list including unique identifiers of high-risk nodes. This processing process directly extracts abnormal features from the data stream of the stream computing engine, realizes rapid positioning of fault nodes, and provides abnormal signal input for subsequent weight adjustment.

[0046] The Holt-Winters prediction operation is performed on the time series fluctuation data, and the time series fluctuation data refers to continuous time series formed by historical response latency or traffic trend; the Holt-Winters prediction model decomposes data into horizontal, trend, and seasonal components, and outputs adjusted traffic trend parameters including predicted traffic values and change slopes. This prediction step uses data organized by time window slicing for modeling, adapts to dynamic traffic fluctuations, and provides forward-looking parameter input for path optimization.

[0047] The reinforcement learning strategy operation is performed on the protocol change data, and the protocol change data refers to node-to-node communication protocol or interface update events; the reinforcement learning strategy trains agents to generate decisions based on a reward mechanism, and outputs a topology adjustment instruction set including instructions to add, remove, or modify node connections. This strategy responds to real-time protocol change events, drives topology reconstruction, and ensures dependency adaptation capability in a multi-protocol environment.

[0048] The abnormal node identifier list, the adjusted traffic trend parameter, and the topology adjustment instruction set are merged to integrate the three outputs through a data fusion algorithm; the data fusion algorithm uses vector splicing or weight superposition to generate a unified dynamic weight factor matrix including node-level weight values and path adjustment coefficients. This merging operation unifies the results of abnormal detection, traffic prediction, and protocol adaptation into an operational factor, which is output to step 5 for constructing a three-layer link mapping, completing the closed-loop logic flow from data processing to weight generation.

[0049] Specifically, the metadata-based dynamic link mapping generation method of the present application, step 5 includes: The main service path conversion step 4 converts the dynamic weight factor into an edge cost, and executes the improved Dijkstra algorithm to output a shortest path sequence; The degraded path matching step 3 matches the simulation path library output by the abnormal scenario Monte Carlo tree search; The stress test path filters the candidate path set through the discriminator of the generative adversarial network to generate a final stress test path; The main service path, degraded path and stress test path are integrated to construct a three-layer link mapping.

[0050] In step 5, based on the dynamic weight factor generated in step 4 as input data, the construction process of the three-layer link mapping is started; the dynamic weight factor includes node-level weight values and path adjustment coefficients, which provide dynamic parameter basis for path calculation. This step connects the data processing results of step 4, outputs a structured mapping model through path type processing, and supports the verification operation of step 6.

[0051] The main service path executes the improved Dijkstra algorithm to calculate the shortest path sequence, first converts the dynamic weight factor output in step 4 into an edge cost, and the edge cost represents the transmission cost between nodes such as delay or resource consumption; the improved Dijkstra algorithm optimizes the traditional method, introduces a priority queue to manage node access order, preferentially processes low-cost edges, and outputs a main service path sequence that minimizes the total cost. This processing targets regular service requirements, uses dynamic weight factors to adapt to real-time environmental changes, and provides an efficient transmission basis for the system.

[0052] The degraded path matching step 3 matches the simulation path library output by the abnormal scenario Monte Carlo tree search, and the simulation path library stores a high-availability path set generated under simulated fault events; the matching process compares the current topology with the path library entries through a path similarity algorithm, and selects the degraded path with the highest fault tolerance, such as selecting the optimal match based on node availability scores. This step uses the simulation results of step 3 to enhance the robustness of the system in abnormal environments and ensure the reliability of the fault transfer path.

[0053] The stress test path generates a final path through a generative adversarial network, which includes generator and discriminator components; the generator outputs a candidate path set based on an enhanced topology model, the discriminator evaluates the authenticity and stress bearing capacity of the path, and filters out the extreme test path; the final stress test path is designed to simulate high concurrency or extreme conditions, and the generation process iteratively optimizes the path distribution. This operation solves the problem of concurrent stress test distortion, provides real traffic simulation capability, and perfects the stress test dimension of the three-layer architecture.

[0054] The three-layer link mapping is constructed by integrating the main service path, the degraded path and the stress test path. The three-layer path sequence is aligned by using data fusion technology in the integration operation to form a structured mapping model. The model includes node sequence and dependency data, and is output to step 6 for verification. This construction process completes the transition from sub-path calculation to unified model, covering the full scene of normal service, fault transfer and stress test. The logical relationship of the whole step 5 is embodied from dynamic weight factor input to sub-path independent processing and then to integrated output, realizing the complete construction of link mapping and providing basic input for closed-loop verification.

[0055] Specifically, the metadata-based dynamic link mapping generation method of the present application, step 6 includes: The main inspection flow traverses the node sequence of the main service path, the degraded path and the stress test path constructed in step 5, and calculates the depth traversal hash value; The auxiliary inspection flow analyzes the node dependency data output by step 5 to verify the cross-layer transmission closure relationship; When the node coverage is detected to be missing, step 2 is triggered to perform incremental topology update.

[0056] In step 6, the main inspection flow traverses the node sequence of the main service path, the degraded path and the stress test path constructed in step 5, and uses a depth-first traversal algorithm to access the nodes of each layer path. A unique identifier of the node sequence is generated during the traversal process, and a depth traversal hash value is calculated by a hash function. The hash value serves as a check code for the integrity of the path. The main inspection flow outputs the path node coverage state to provide basic data for subsequent verification, and realizes the structural integrity test of the three-layer link mapping.

[0057] The auxiliary inspection flow analyzes the node dependency data output by step 5 to establish a cross-layer transmission closure relationship model. The transmission closure relationship model describes the node dependency transmission rules between the main service path, the degraded path and the stress test path, and verifies the cross-layer node connectivity through matrix operation. The auxiliary inspection flow detects the logical consistency of the node dependency link and outputs the cross-layer dependency verification result to supplement the structural verification dimension of the main inspection flow, forming a complete topology logical verification system.

[0058] When the main inspection flow or the auxiliary inspection flow detects that the node coverage is missing, step 2 is triggered to perform incremental topology update. The incremental topology update mechanism only reanalyzes the static dependency relationship for the missing nodes, and predicts the local risk features through a long short-term memory network to generate a supplementary topology segment. The updated topology segment is merged into the original enhanced topology model, and the path calculation process of steps 3 to 5 is restarted to construct the corrected three-layer link mapping.

[0059] The path effectiveness score is dynamically generated by comparing the path effectiveness index in step 3 with the difference degree of the mapping result in step 5. The node stability score is calculated based on the historical failure rate and recovery time data in step 1. The score results are respectively used as feedback signals: the path effectiveness score is input into step 3 to update the failure simulation weight factor of the Monte Carlo tree search; and the node stability score is input into step 4 to optimize the model parameters of the isolated forest anomaly detection and Holt-Winters traffic prediction.

[0060] The reference flow is injected into the main service path to simulate normal requests, the failure test flow is injected into the degraded path to simulate node failure, and the chaotic disturbance flow is injected into the stress test path to simulate extreme conditions. The response delay, error rate and throughput data of the three-layer path are collected, and a three-dimensional performance matrix of delay / error rate / throughput is established. The three-dimensional performance matrix is subjected to spline interpolation calculation, the abnormal node whose delay exceeds 200% of the degraded path is identified, a nonlinear performance bottleneck node positioning report is output, and the service bottleneck diagnosis closed loop is completed.

[0061] Specifically, the dynamic link mapping generation method based on metadata of the application comprises the following steps: The difference degree of the path effectiveness index in step 3 and the mapping result in step 5 is calculated to generate a path effectiveness index; The historical failure rate and recovery time data in step 1 are extracted to calculate a node stability score; The path effectiveness index and the node stability score are input into the feedback control loop in step 3 to correct the node selection probability parameters of the genetic algorithm in step 3.

[0062] In the dynamic scoring process of step 6, the path effectiveness index is generated by difference degree calculation. The specific execution mode is as follows: the path effectiveness index output by step 3 is compared with the actual result data of the three-layer link mapping in step 5. The path effectiveness index includes path delay, throughput and other parameters, and the mapping result indicates the node sequence state of the main service path, the degraded path and the stress test path. The difference degree calculation adopts a dispersion analysis method to measure the deviation degree of the theoretical path performance and the actual path performance, and generates a quantitative path effectiveness index value. The index reflects the degree of coincidence between path planning and actual deployment, and provides a data basis for feedback control.

[0063] The node stability score is calculated based on the historical trajectory data collected in step 1. The historical trajectory data includes periodic records of node failure rate and recovery time. The calculation process extracts the failure frequency and average recovery time within a specific time window, generates a node stability score value through a weighted fusion formula, and the higher the score value, the stronger the node reliability. The score is directly related to the historical operation performance of the node, and provides a historical basis for anomaly detection.

[0064] The path efficiency indicator and the node stability score are jointly input into the feedback control loop of step 3, and the feedback control loop adopts a closed-loop adjustment mechanism: the path efficiency indicator is used as a correction factor to adjust the node selection probability parameter of the genetic algorithm in step 3 in real time. The adjustment process dynamically scales the probability distribution of the selection operator according to the value of the path efficiency indicator, and when the value is lower than a threshold, the selection weight of the high-stability node is increased to optimize the node scheduling strategy in the high-concurrency scenario. At the same time, the node stability score is synchronously input into step 4 to update the sample weight of the isolation forest anomaly detection and the seasonal coefficient of the Holt-Winters traffic prediction, forming a cross-step cooperative optimization mechanism.

[0065] The entire dynamic scoring process establishes a closed-loop logic of data verification and algorithm optimization: path difference analysis reveals mapping deviation, historical data analysis quantifies node reliability, and two-dimensional scoring jointly drives genetic algorithm parameter updating and detection model adjustment. This process solves the diagnosis lag problem and enables continuous evolution of path calculation and anomaly detection.

[0066] Specifically, the metadata-based dynamic link mapping generation method of the present application includes the following steps in step 6: Extract the response delay data of the main path, the degraded path, and the stress test path after injecting the reference traffic, the fault test, and the chaos disturbance, and mark the nodes whose main path delay exceeds the degraded path by 200%; Establish a three-dimensional performance matrix of the main path, the degraded path, and the stress test path in terms of latency, error rate, and throughput; Perform spline interpolation calculation on the three-dimensional performance matrix to output the non-linear performance bottleneck node.

[0067] The operation of comparing three-layer performance data in step 6 is started by first extracting the response delay data after injecting reference traffic to simulate normal requests to the main service path, injecting fault test traffic to simulate node failure to the degraded path, and injecting chaos disturbance traffic to simulate extreme conditions to the stress test path. The response delay data includes the end-to-end latency value in the path transmission process, which is captured in real time by a distributed monitoring agent, and the data source is the actual running output of the three-layer link mapping constructed in step 5. This data extraction step provides the original input for performance analysis, connects the traffic injection action with the subsequent calculation logic, and forms the initial stage of performance evaluation.

[0068] Then, mark the abnormal nodes whose main service path delay exceeds the degraded path by 200%, and the marking process is based on the comparison algorithm of the response delay data: calculate the ratio of the node latency of the main service path to the corresponding node latency of the degraded path, and when the ratio exceeds a predefined threshold, add a high-delay identifier to the node. This marking operation identifies performance degradation points and reveals the weaknesses of the main path under high load or fault scenarios, providing key abnormal input data for the establishment of the three-dimensional performance matrix and linking the data extraction and matrix construction steps.

[0069] Then a three-dimensional performance matrix of the main service path, the degraded path and the stress test path is established, and the three-dimensional performance matrix dimension includes delay, error rate and throughput indicators. The establishment process adopts a data aggregation technology to organize the extracted response delay data, error rate records and throughput measurement values into a structured matrix form, and the matrix row represents the path node and the list represents the performance indicator type. This matrix construction step integrates multi-dimensional performance data, provides structured input for nonlinear bottleneck analysis, and supports the execution of spline interpolation calculation.

[0070] Finally, spline interpolation calculation is performed on the three-dimensional performance matrix to output the nonlinear performance bottleneck node; the spline interpolation calculation uses a piecewise polynomial function to fit the matrix data, identifies the mutation area of the delay, error rate and throughput indicators, and locates the position of the nonlinear performance bottleneck node. The output result is a list of node identifiers, which includes key bottleneck points affecting system stability. This calculation step solves the diagnosis lag problem, provides accurate positioning basis for service optimization, and completes the closed-loop logic flow from performance data to bottleneck output. The comparison operation of the entire step 6 is represented as a sequence of data extraction, abnormality marking, matrix construction and interpolation calculation, which enhances the integrity and reliability of the three-layer path performance analysis.

[0071] Specifically, the metadata-based dynamic link mapping generation method of the application further includes the following steps in step 4: When detecting a protocol change in step 4, the abnormal mode characteristics of the stress test path in step 5 are analyzed, and a topology reconstruction instruction set is generated; The action space of the reinforcement learning in step 4 is updated using the reconstruction instruction set, and a new topology reconstruction decision is executed.

[0072] In step 4, when detecting protocol change data, the protocol change data refers to an update event of the communication protocol or interface definition between distributed nodes, and the change signal is captured in real time through an event listening module in the stream computing engine. This detection operation is started based on the protocol change data type processed in step 4, and connects the input of the reinforcement learning strategy in step 4 to provide a trigger condition for subsequent feature analysis, forming the starting point of the dynamic response mechanism.

[0073] After detecting the protocol change, the abnormal mode characteristics of the stress test path constructed in step 5 are analyzed; the abnormal mode characteristics refer to the response delay mutation or error rate anomaly of the stress test path under chaotic disturbance flow. The analysis process uses a feature extraction algorithm to analyze the stress test path data sequence, identifies high-frequency abnormal points or mode shifts, and outputs a structured feature description, which converts the stress test results of step 5 into quantifiable feature input to support the reconstruction decision.

[0074] Based on the parsed abnormal pattern features, a topology reconstruction instruction set is generated; the instruction set includes specific operation commands such as adding node connection, removing redundant link or modifying weight factor. The generation process maps the features to the reconstruction action through the decision tree algorithm, outputs the instruction set as the basic input of reinforcement learning optimization, realizes the transition from feature analysis to instruction generation, and solves the problem of dependency relationship failure caused by protocol change.

[0075] The generated reconstruction instruction set is used to update the action space of step 4 reinforcement learning; the action space refers to the set of topology adjustment operations that can be selected in the reinforcement learning strategy. The update operation adds new instructions as legal actions to the original space, optimizes the strategy selection by adjusting the action probability distribution, and ensures that the reinforcement learning model adapts to the environmental changes after the protocol change. This update step feeds back the step 5 stress test results to step 4 algorithm, forming a cross-step cooperative optimization mechanism.

[0076] Finally, the new topology reconstruction decision is executed; the execution process is based on the updated action space to run the reinforcement learning strategy, and outputs specific topology adjustment decisions such as node connection update instructions. The decision result is directly applied to generate dynamic weight factors, and input to step 5 to build link mapping, completing the adaptive closed-loop control under protocol change. The entire additional process logic reflects the detection of change trigger feature analysis, feature-driven instruction generation, instruction update action space, new decision execution reconstruction, and the robustness of the reinforcement system in a dynamic protocol environment.

[0077] The application solves the adaptability problem of static model by dynamically fusing static metadata and historical performance data to construct an enhanced topology model. Specifically, a syntax parser processes node static dependencies to generate a basic adjacency list, while a long short-term memory network analyzes historical trajectory data to predict response timeout probability and fault correlation degree, and encodes and labels the risk hot area to the adjacency list to form an enhanced topology model. This model breaks through the limitations of static metadata and reflects node state changes in real time.

[0078] To solve the problem of multi-protocol dependency inference lag, the stream computing engine triggers special algorithms according to data types: numerical anomaly data triggers isolated forest to detect abnormal nodes, time series fluctuation data predicts traffic trend through Holt-Winters model, and protocol change data drives reinforcement learning to generate topology reconstruction instructions. The outputs of the three algorithms are combined into dynamic weight factors to realize real-time decision-making in a multi-protocol environment.

[0079] The three-layer link mapping architecture eliminates the omission of abnormal paths: the main service path executes improved Dijkstra algorithm based on dynamic weight factors to generate the optimal path; the degraded path matches the high availability path library simulated by Monte Carlo tree search; and the stress test path is filtered by the generative adversarial network to obtain the extreme test path. The hierarchical design covers normal, fault and high-pressure scenarios.

[0080] The closed-loop verification mechanism solves the problems of pressure test distortion and diagnosis lag: the path efficiency score is fed back to the Monte Carlo tree search to update the fault simulation weight, and the node stability score is fed back to optimize the abnormal detection algorithm. After injecting reference flow, fault test flow and chaotic disturbance flow into the three-layer path, a three-dimensional matrix of delay / error rate / throughput is constructed, and a spline interpolation is used to locate the nonlinear bottleneck node. The verification failed path triggers the topology incremental update, the abnormal characteristics of the pressure test path drive reinforcement learning to reconstruct the decision when the protocol changes, and the dynamic topology adaptation across cloud environments is realized.

[0081] In the cross-cloud platform service invocation scenario, the metadata model of the distributed system is difficult to adapt to dynamic topology changes, and the algorithm cannot real-time infer the complex dependency relationship between multi-protocol nodes, resulting in missing abnormal paths in link mapping, concurrent pressure test distortion, performance bottleneck diagnosis lag and system stability misjudgment.

[0082] A lightweight monitoring agent is deployed to the distributed node, and the node identifier, node type such as microservice or database node and inter-service dependency are periodically captured in the cross-cloud platform environment to form the static dependency part of the metadata. At the same time, the message queue depth indicators such as queue length and backlog message quantity and thread pool state such as active thread number and idle thread number are collected to constitute the performance indicators of the historical trajectory data. The historical response delay and error rate data are sliced according to the pre-defined time window, for example, five-minute window, the average response delay and error rate change rate are calculated, and a multi-dimensional time series feature vector is generated to provide input for subsequent risk prediction.

[0083] The static dependency relationship data obtained in step 1 is parsed, and a syntax parser is used to process the service invocation chain configuration file or API definition document to generate an adjacency list through lexical analysis and syntax tree construction. The adjacency list stores the node dependency connection relationship indexed by the node identifier to form the basic topology framework. In parallel, the time series feature vector output by step 1 is input into the long short-term memory network to predict risk features such as node response timeout probability and fault correlation degree; the prediction result is encoded as a risk hot area label and marked on the corresponding node position of the adjacency list to identify the high-risk area. The annotated adjacency list and risk feature data are merged, and the node attributes are aligned through a data fusion algorithm to output an enhanced topology model containing static dependency and dynamic risk weight.

[0084] Based on the enhanced topology model, the path calculation of the normal scenario, the abnormal scenario and the high concurrency scenario is performed in parallel. The normal scenario takes the node weight factor as the heuristic function, performs the A* search algorithm to calculate the optimal transmission path, and outputs the path sequence that minimizes the transmission delay or resource consumption. The abnormal scenario injects random node failure events, expands the Monte Carlo tree search space to simulate fault transfer paths, evaluates the path availability and generates a high-reliability path library. The high concurrency scenario adopts a chromosome coding scheme to represent the node scheduling order, and uses genetic algorithm optimization selection, crossover and mutation operations, adjusts the fitness function according to the weight factor, and outputs the high-efficiency concurrent scheduling path. Record the path delay, throughput and error rate parameters of the three scenarios, and aggregate them into path effectiveness indicators, and output them to the streaming computing engine.

[0085] The streaming computing engine receives the path effectiveness indicators of step 3 and the real-time state events of the distributed nodes such as performance monitoring data, and processes the data types separately. For numerical anomaly data such as message queue depth or thread pool state values deviating from the normal range, perform the Isolation Forest detection algorithm to build a random tree structure to identify outliers, and output an abnormal node identification list. For time series fluctuation data such as continuous sequences formed by historical response delays, perform the Holt-Winters prediction model to decompose the level, trend and seasonal components, and output the adjusted traffic trend parameters such as predicted traffic values and change slopes. For protocol change data such as node intercommunication protocol update events, perform the reinforcement learning strategy based on the reward mechanism to train the agent, and output the topology adjustment instruction set such as adding or removing node connection instructions. Merge the abnormal node identification list, the adjusted traffic trend parameters and the topology adjustment instruction set, and use vector splicing to generate a dynamic weight factor matrix containing node-level weight values and path adjustment coefficients.

[0086] Use the dynamic weight factors output in step 4 to construct a three-layer link mapping. The main business path converts the dynamic weight factors into edge value such as delay cost, performs the improved Dijkstra algorithm to introduce a priority queue to optimize the node access order, and outputs the shortest path sequence. The degraded path matches the simulation path library output by the Monte Carlo tree search of step 3 abnormal scenario, and filters the high-availability path based on the node availability score through the path similarity algorithm. The stress test path outputs a candidate path set through the generator of the generative adversarial network, and the discriminator evaluates the path authenticity and stress bearing capacity to filter the limit test path. Integrate the main business path, the degraded path and the stress test path, align the node sequence using data fusion technology, and form a three-layer link mapping model covering normal service, fault transfer and stress test.

[0087] The three-layer link mapping node coverage constructed in step 5 is verified for dependency transmission. The main inspection flow uses a depth-first traversal algorithm to access the node sequence of the main service path, the degraded path and the stress test path, calculates a depth traversal hash value as a path integrity check code. The auxiliary inspection flow parses the node dependency relationship data, establishes a transmission closure relationship model to verify the connectivity of cross-layer nodes, and outputs the dependency verification result. When the node coverage is missing, step 2 is triggered to perform incremental topology update, reparse the static dependency relationship and predict the local risk characteristics only for the missing nodes, generate a supplementary topology segment, and restart steps 3 to 5. The path validity score and the node stability score are dynamically generated; the path validity score is calculated by comparing the path validity indicators in step 3 and the mapping results in step 5, and a quantitative indicator is generated; the node stability score is calculated based on the historical failure rate and recovery time data in step 1, and the score value reflects the node reliability. The path validity score is input into step 3 to update the Monte Carlo tree search fault simulation weight factor, and the node stability score is input into step 4 to optimize the isolated forest detection sample weight and the Holt-Winters prediction seasonal coefficient. The baseline flow is injected into the main service path to simulate normal requests, the fault test flow is injected into the degraded path to simulate node failure, and the chaotic disturbance flow is injected into the stress test path to simulate extreme conditions, and the response delay data is extracted to mark the nodes whose main path delay exceeds 200% of the degraded path. A three-dimensional performance matrix of delay / error rate / throughput is established, spline interpolation calculation is performed on the matrix to identify the mutation area, and the nonlinear performance bottleneck node is output.

[0088] When the protocol changes, step 4 detects protocol change data such as interface definition update events, parses the abnormal mode characteristics of step 5 stress test path such as response delay mutation, and generates a topology reconstruction instruction set. The instruction set is used to update the action space of the reinforcement learning strategy in step 4, add new topology adjustment operations, execute new topology reconstruction decisions such as node connection update instructions, and apply to dynamic weight factor generation to complete the adaptive closed-loop control reinforcement system robustness. The entire implementation enhances the topology model, processes data types by the stream computing engine, constructs three-layer link mapping, and realizes dynamic topology adaptation and performance optimization in cross-cloud platform scenarios through the closed-loop verification mechanism.

Claims

1. A method for generating dynamic link mapping based on metadata, characterized in that, Comprise: Step 1, obtain the metadata of distributed nodes and historical trajectory data, the metadata includes static dependency relationship, and the historical trajectory data includes performance index; Step 2, parse the static dependency relationship of step 1 to generate a basic topology, and input the historical trajectory data of step 1 into a long short-term memory network to predict risk features, and combine the basic topology and the risk features to generate an enhanced topology model; Step 3, based on the enhanced topology model of step 2, perform in parallel: in a normal scenario, use A* search to calculate the optimal path; in an abnormal scenario, use Monte Carlo tree search to simulate failure; in a high concurrency scenario, use genetic algorithm to schedule nodes; output path validity index to a streaming computing engine; Step 4, receive node state events and path validity index of step 3 through the streaming computing engine, trigger isolated forest anomaly detection, Holt-Winters traffic prediction and reinforcement learning topology reconstruction algorithm for numerical anomaly, time series fluctuation and protocol change respectively, and generate dynamic weight factors; Step 5, use the dynamic weight factors of step 4 to build a three-layer link mapping: the main business path is generated by improved Dijkstra algorithm, the degraded path matches the Monte Carlo simulation path library, and the stress test path is generated by generative adversarial network; Step 6, verify the three-layer path node coverage and dependency transitivity of step 5, dynamically generate path validity score and node stability score, use the path validity score as the update weight factor of Monte Carlo tree search simulation failure in abnormal scenario, and input it into step 3; use the node stability score as the update weight factor of isolated forest anomaly detection and Holt-Winters traffic prediction in step 4, and input it into step 4; restart step 3 for the failed path, inject benchmark traffic to the main path, inject fault test to the degraded path, and inject chaos disturbance to the stress test path; compare the performance data of the three-layer path to locate the service bottleneck node.

2. The metadata-based dynamic link mapping generation method according to claim 1, wherein, Step 1 includes: Collecting periodically captured node ID, type and dependency relationship basic attributes; Collecting runtime indicators such as message queue depth and thread pool state; Slice the historical response time and error rate data by time window, and output the time series feature vector to the long short-term memory network of step 2 to predict risk features.

3. The metadata-based dynamic link mapping generation method according to claim 1, wherein, Step 2 includes: Parse the static dependency relationship of step 1 to generate an adjacency list through a syntax parser, and store the node dependency relationship in the adjacency list; Encode the risk hot zone output by the long short-term memory network of step 2 and label it on the corresponding node of the adjacency list; Merge the labeled adjacency list and the risk features predicted by the long short-term memory network of step 2 to output the enhanced topology model.

4. The metadata-based dynamic link mapping generation method according to claim 1, wherein, Step 3 includes: Based on the enhanced topology model, use node weight factor as heuristic function to perform A search to calculate the optimal path in normal scenario; In abnormal scenario, expand Monte Carlo tree search space by injecting random faults, and in high concurrency scenario, use chromosome coding scheme to perform node sequence genetic algorithm, record the calculation results of normal scenario, abnormal scenario and high concurrency scenario, and output path validity index to step 4.

5. The metadata-based dynamic link mapping generation method according to claim 1, wherein, Step 4 includes: Perform isolated forest detection on numerical anomaly data to output abnormal node identification; Performing Holt-Winters prediction on time series fluctuation data, outputting adjusted traffic trend parameters; Performing reinforcement learning strategy on protocol change data, outputting topology adjustment instruction set; Merging abnormal node identification, adjusted traffic trend parameters and topology adjustment instruction set to generate dynamic weight factor.

6. The metadata-based dynamic link mapping generation method according to claim 1, wherein, Step 5 includes: Step 4 dynamic weight factor is converted to edge cost, and improved Dijkstra algorithm is executed to output shortest path sequence; The simulation path library output by the abnormal scene Monte Carlo tree search in step 3 is matched with the degraded path; The stress test path is filtered from the candidate path set by the discriminator of the generative adversarial network to generate the final stress test path; The main business path, degraded path and stress test path are integrated to construct a three-layer link mapping.

7. The metadata-based dynamic link mapping generation method according to claim 1, wherein, Step 6 includes: The main inspection flow traverses the node sequence of the main business path, degraded path and stress test path constructed in step 5, and calculates the depth traversal hash value; The auxiliary inspection flow analyzes the node dependency data output in step 5 to verify the cross-layer transmission closure relationship; When node coverage is missing, trigger step 2 to perform incremental topology update.

8. The metadata-based dynamic link mapping generation method according to claim 7, wherein, Dynamic scoring in step 6 includes: Calculate the difference between the path effectiveness index in step 3 and the mapping result in step 5 to generate a path efficiency index; Extract historical failure rate and recovery time data in step 1 to calculate node stability score; Input the path efficiency index and node stability score into the feedback control loop in step 3 to correct the node selection probability parameter of the genetic algorithm in step 3.

9. The metadata-based dynamic link mapping generation method according to claim 1, wherein, Comparing three-layer performance data in step 6 includes: Extract the response delay data of the main path, degraded path and stress test path after injecting the reference traffic, fault test and chaos disturbance in step 6, and mark the nodes whose main path delay exceeds the degraded path by 200%; Establish a three-dimensional performance matrix of the main path, degraded path and stress test path in terms of delay, error rate and throughput; Perform spline interpolation calculation on the three-dimensional performance matrix to output non-linear performance bottleneck nodes.

10. The metadata-based dynamic link mapping generation method according to claim 1, wherein, In step 4, it also includes: When detecting protocol changes in step 4, analyze the abnormal mode characteristics of the stress test path in step 5 to generate a topology reconstruction instruction set; Update the action space of step 4 reinforcement learning using the reconstruction instruction set to execute new topology reconstruction decisions.

Images