Modular function development method and device of digital twin network management platform

Abstract

The embodiment of the application provides a modular function development method and device of a digital twin network management platform, accurate modeling of a network is realized through state analysis and topology mapping. A management mechanism is constructed, a reliable early warning strategy is established in combination with an intelligent model and fault correlation. Decision optimization is introduced, and continuous improvement of management is ensured through scheme verification and feedback adjustment. The method effectively solves the deficiencies of traditional technologies in network modeling, fault early warning and scheme decision, and provides technical support for network management.

Description

Technical Field

[0001] This application relates to the field of data processing, specifically to a modular function development method and apparatus for a digital twin network management platform. Background Technology

[0002] Existing digital twin network management platform development methods have significant shortcomings. Traditional systems perform poorly in state analysis and topology mapping, failing to effectively achieve accurate network modeling and thus impacting management efficiency.

[0003] Furthermore, existing technologies suffer from bottlenecks in anomaly detection and fault correlation. Most systems lack robust intelligent model training mechanisms and risk warning strategies, resulting in suboptimal fault identification accuracy.

[0004] The existing system has technical shortcomings in solution decision-making. It lacks in-depth verification of disposal solutions, making it difficult to achieve efficient policy adjustments through feedback optimization, thus impacting operational effectiveness. Solving these problems is crucial for improving network management capabilities. Summary of the Invention

[0005] To address the problems in existing technologies, this application provides a modular functional development method and apparatus for a digital twin network management platform, which can effectively solve the shortcomings of traditional technologies in network modeling, fault early warning, and solution decision-making, and provide technical support for network management.

[0006] To solve at least one of the above problems, this application provides the following technical solution: Firstly, this application provides a modular functional development method for a digital twin network management platform, including: Based on the communication node status data, extract the node operation parameter set and network topology information set, perform time series correlation analysis on the operation parameter set and the network topology information set to obtain the state feature vector, generate the network structure model through topology mapping processing of the state feature vector, perform hierarchical parsing on the network structure model to obtain the functional component library, and establish a twin object mapping table based on the functional component library. The twin object mapping table is divided into training datasets according to a preset training strategy. Training constraint condition groups are constructed according to historical operation and maintenance rules. Constraint training is performed on the training dataset to generate a network management intelligent model library. The network management intelligent model library is applied to real-time status evaluation to obtain a performance index set. Multi-dimensional anomaly detection is performed on the performance index set to obtain a risk warning list. The risk warning list is compared with the historical fault case library to calculate the similarity and obtain a fault correlation matrix. A set of handling schemes is generated based on the fault correlation matrix. The set of handling schemes is simulated and verified to obtain a scheme scoring table. The scheme scoring table is processed by an intelligent decision engine to obtain an operation instruction sequence. The execution status of the operation instruction sequence is tracked to obtain a feedback result set. The parameters of the digital twin model are updated based on the feedback result set. The updated model parameters and the early warning threshold are adaptively adjusted to obtain an optimization configuration item. The network operation strategy is dynamically optimized according to the optimization configuration item.

[0007] Furthermore, it also includes: establishing a monitoring parameter matrix based on communication node status data, performing normalization processing on the monitoring parameter matrix to obtain a standardized parameter table, dividing the time window based on the standardized parameter table to obtain a time-series sampling set, mapping the time-series sampling set according to topological relationships to generate a node association table, and extracting features from the node association table to obtain parameter feature groups; The parameter feature group is used to generate a feature weight matrix through correlation calculation. The feature weight matrix is ​​then filtered by applying a preset threshold condition to obtain a key feature set. A time-series feature vector is constructed based on the key feature set. The time-series feature vector is then matched with topological constraint rules to generate a state feature vector.

[0008] Furthermore, it also includes: constructing a node topology relation matrix based on state feature vectors, grouping the node topology relation matrix according to hierarchical division rules to obtain a hierarchical structure table, performing connectivity analysis on the hierarchical structure table to generate a node mapping graph, converting the node mapping graph through topology mapping rules to obtain a network structure model, and parsing the network structure model according to functional attributes to obtain a functional component library; The component units in the functional component library are standardized and encoded to obtain a component description set. The component description set is classified and labeled according to a preset mapping rule to obtain a component tag library. A virtual-real mapping rule set is established based on the component tag library. The virtual-real mapping rule set is matched with the component association relationship to generate a twin object mapping table.

[0009] Furthermore, it also includes: dividing the twin object mapping table into sample sequence sets according to time windows, cleaning and labeling the sample sequence sets to obtain preprocessed data sets, extracting training features from the preprocessed data sets to obtain a feature sample library, dividing the feature sample library into training sets according to a preset ratio to obtain a training dataset, and extracting constraint parameters from historical operation and maintenance rules to obtain a constraint condition set. The initial model set is obtained by initializing the model parameters of the training dataset. The initial model set and the constraint condition group are iteratively trained to obtain the candidate model set. The candidate model set is validated and evaluated to obtain the model scoring table. Based on the model scoring table, models that meet the accuracy requirements are selected to build the network management intelligent model library.

[0010] Furthermore, it also includes: building a real-time evaluation engine based on the network management intelligent model library, inputting real-time monitoring data into the real-time evaluation engine to perform state calculation to obtain state evaluation results, extracting performance indicators from the state evaluation results according to the performance dimension to obtain a performance indicator set, comparing the performance indicator set with preset threshold rules to obtain anomaly detection results, and generating a risk warning list based on the anomaly detection results; The warning items in the risk warning list are processed by feature vectorization to obtain a warning feature set. The warning feature set is matched with the fault features in the historical fault case library to obtain a feature matching degree table. Based on the feature matching degree table, the fault association weight is calculated to obtain the fault association degree matrix.

[0011] Furthermore, it also includes: performing cluster analysis on the fault correlation matrix to obtain fault type groups, extracting key fault features based on the fault type groups to obtain a feature pattern set, performing rule matching between the feature pattern set and the disposal knowledge base to obtain a candidate solution group, performing feasibility evaluation on the candidate solution group to obtain a disposal solution set, and inputting the disposal solution set into a simulation environment for verification testing to obtain a solution scoring table. The scheme scoring table is sorted by priority according to preset decision rules to obtain a scheme decision sequence. Resource constraint analysis is performed on the scheme decision sequence to obtain an execution constraint table. Based on the execution constraint table, the schemes are scheduled and orchestrated to obtain a scheduling strategy set. The scheduling strategy set is converted into specific operation steps to generate an operation instruction sequence.

[0012] Furthermore, it also includes: parsing the sequence of operation instructions according to the execution time sequence to obtain an execution task table; constructing monitoring and collection points on the execution task table to obtain a monitoring parameter group; collecting execution process data based on the monitoring parameter group to obtain a state sequence set; performing deviation analysis between the state sequence set and the expected target to obtain an execution evaluation table; summarizing the results of the execution evaluation table to generate a feedback result set; and extracting model correction parameters based on the feedback result set to obtain a parameter update set. The parameter update set is input into the model optimizer to adjust the parameters and obtain an optimized parameter group. The optimized parameter group is then subjected to threshold calibration to obtain a threshold adjustment table. Based on the threshold adjustment table, a configuration update instruction is generated to obtain an optimized configuration item. The optimized configuration item is applied to the policy controller to adjust the policy and obtain an operating policy group. The operating policy group is then deployed and verified to complete the network operating policy optimization.

[0013] Secondly, this application provides a modular function development device for a digital twin network management platform, comprising: Based on the communication node status data, extract the node operation parameter set and network topology information set, perform time series correlation analysis on the operation parameter set and the network topology information set to obtain the state feature vector, generate the network structure model through topology mapping processing of the state feature vector, perform hierarchical parsing on the network structure model to obtain the functional component library, and establish a twin object mapping table based on the functional component library. The twin object mapping table is divided into training datasets according to a preset training strategy. Training constraint condition groups are constructed according to historical operation and maintenance rules. Constraint training is performed on the training dataset to generate a network management intelligent model library. The network management intelligent model library is applied to real-time status evaluation to obtain a performance index set. Multi-dimensional anomaly detection is performed on the performance index set to obtain a risk warning list. The risk warning list is compared with the historical fault case library to calculate the similarity and obtain a fault correlation matrix. A set of handling schemes is generated based on the fault correlation matrix. The set of handling schemes is simulated and verified to obtain a scheme scoring table. The scheme scoring table is processed by an intelligent decision engine to obtain an operation instruction sequence. The execution status of the operation instruction sequence is tracked to obtain a feedback result set. The parameters of the digital twin model are updated based on the feedback result set. The updated model parameters and the early warning threshold are adaptively adjusted to obtain an optimization configuration item. The network operation strategy is dynamically optimized according to the optimization configuration item.

[0014] Thirdly, this application provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the steps of the modular function development method of the digital twin network management platform.

[0015] Fourthly, this application provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the modular function development method for the digital twin network management platform.

[0016] Fifthly, this application provides a computer program product, including a computer program / instruction, which, when executed by a processor, implements the steps of the modular function development method for the digital twin network management platform.

[0017] As can be seen from the above technical solution, this application provides a modular functional development method and apparatus for a digital twin network management platform. Through state analysis and topology mapping, it achieves accurate network modeling. A management mechanism is constructed, combining intelligent models and fault correlation to establish a reliable early warning strategy. Decision optimization is introduced, and through scheme verification and feedback adjustments, continuous improvement in management is ensured. This method effectively solves the shortcomings of traditional technologies in network modeling, fault early warning, and scheme decision-making, providing technical support for network management. Attached Figure Description

[0018] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0019] Figure 1 This is a flowchart illustrating the modular function development method of the digital twin network management platform in this application embodiment; Figure 2 This is a structural diagram of the modular function development device for the digital twin network management platform in this application embodiment. Detailed Implementation

[0020] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0021] The acquisition, storage, use, and processing of data in this application comply with relevant laws and regulations.

[0022] In view of the problems existing in the prior art, this application provides a modular functional development method and apparatus for a digital twin network management platform. Through state analysis and topology mapping, it achieves accurate network modeling. A management mechanism is constructed, combining intelligent models and fault correlation to establish a reliable early warning strategy. Decision optimization is introduced, and continuous improvement in management is ensured through scheme verification and feedback adjustments. This method effectively solves the shortcomings of traditional technologies in network modeling, fault early warning, and scheme decision-making, providing technical support for network management.

[0023] To effectively address the shortcomings of traditional technologies in network modeling, fault early warning, and solution decision-making, and to provide technical support for network management, this application provides an embodiment of a modular functional development method for a digital twin network management platform. See [link to relevant documentation]. Figure 1 The modular function development method of the digital twin network management platform specifically includes the following: Step S101: Extract the node operation parameter set and network topology information set based on the communication node status data, perform time series correlation analysis on the operation parameter set and the network topology information set to obtain the state feature vector, generate the network structure model by topology mapping of the state feature vector, perform hierarchical parsing on the network structure model to obtain the functional component library, and establish a twin object mapping table based on the functional component library. First, after accessing the communication node status data, time alignment is performed, and the data is grouped according to the data source and node identifier. The node operating parameter set and network topology information set are then extracted. Time alignment is performed based on the data acquisition timescale and heartbeat packet sequence verification, correcting gaps and marking interpolation positions, producing aligned parameter fragments and topology snapshots. These two types of data are uniformly encoded into a sequence mapping with nodes as the key, serving as input for subsequent time series correlation analysis.

[0024] Based on the aligned data, the node operating parameter set is divided into fixed-length windows, and parameter neighborhoods within each window are constructed using the link direction of the topology snapshot as a constraint. To suppress occasional spikes, robust filtering is used to smooth the parameter trajectory of each window, and then the rate of change and fluctuation amplitude are calculated on a window-by-window basis to output initial time-series metrics. These initial time-series metrics and topology adjacency relationships are fed in parallel into an integrated analyzer to complete cross-node time-series correlation calculations, resulting in a preliminary set of state feature vectors.

[0025] Based on the aforementioned pre-set set, a "topological constraint correlator" is constructed. This correlator jointly determines the correlation strength and causal direction of node pairs within the same window, and applies a suppression threshold condition to weakly correlated pairs, forming state feature vectors. To ensure the stability of subsequent mappings, each vector is accompanied by a window identifier and a topological hierarchy label, and the source field from the running parameters is registered for traceability. These state feature vectors then enter the topological mapping processing stage.

[0026] Based on the state feature vectors, a network structure model is generated. The mapping step first constructs node connections based on the topology snapshot, then injects the state feature vectors as weights into the connection edges and node weights. To avoid structural distortion caused by locally high weights, intra-layer normalization and inter-layer balance are introduced. This embodiment uses a constraint objective as the mapping criterion: Ψ = μ1·Ω adjacent + μ2·Ω stable, Where Ψ is the mapping target value, μ1 and μ2 are non-negative balance coefficients, Ωneighbor represents the sum of squares of the weight differences between adjacent nodes, and Ωstable represents the deviation of the node weights from their window mean. This target is minimized by the mapper to obtain a stable network structure model.

[0027] Based on the network structure model, a hierarchical parsing process is performed to obtain a functional component library. During parsing, the graph is first split into subgraphs according to the physical layer, logical layer, and service carrying relationship. Then, within each subgraph, components are aggregated according to node roles and traffic paths to form candidate components. Boundary connectivity and internal consistency are calculated for candidate components, filtering out sets with excessively high cross-layer coupling, and producing component units with clearly defined responsibilities. The functional component library records component roles, ingress / egress ports, and dependency chains for further coding.

[0028] Based on the aforementioned functional component library, component units are standardized and described to form triples of component tags, interface constraints, and behavioral summaries. These triples are then mapped one-to-one or one-to-many to the node set in the network structure model. To support the correspondence between virtual and real objects, a "virtual-real mapping rule" is established. This rule reads component tags and node roles, provides the types of devices that can be bound, monitoring points, and control points, and outputs candidate twin anchor points.

[0029] Based on the aforementioned virtual-real mapping rules and candidate twin anchor points, a twin object mapping table is generated. This table uses the component identifier as the primary key, lists the corresponding real device identifier, monitoring parameter path, and control command entry point, and registers the source state feature vector window range to ensure traceability of subsequent updates. To facilitate subsequent training data segmentation, time segments and topological hierarchy indexes are also recorded, forming a mapping record that can be retrieved both temporally and structurally.

[0030] Finally, the twin object mapping table is read in subsequent steps: in the subsequent training dataset partitioning, the time segments and component keys of the mapping table are directly called to complete the sample sequence construction; in real-time state evaluation, the monitoring parameter paths and control command entry points of the mapping table are used for read and write binding. Through the above connection, the network structure model and functional component library produced in this step not only support the correspondence between virtual and real, but also provide a consistent index entry point for subsequent model training and online evaluation.

[0031] Step S102: Divide the twin object mapping table into training datasets according to a preset training strategy, construct training constraint condition groups according to historical operation and maintenance rules, perform constraint training on the training datasets to generate a network management intelligent model library, apply the network management intelligent model library to real-time status evaluation to obtain a performance index set, perform multi-dimensional anomaly detection on the performance index set to obtain a risk warning list, and perform similarity calculation between the risk warning list and the historical fault case library to obtain a fault correlation matrix; First, the aforementioned twin object mapping table is integrated into the sample preparation process. Based on a preset training strategy, the mapping records are segmented by time and grouped by components. Time segmentation is based on window boundaries, and component grouping is based on role labels in the functional component library; the intersection of these two forms a training slice. For each slice, the monitoring parameter path and control command entry are read to complete data extraction and field alignment. For missing test segments, backfillable interpolation is performed, and a backfill marker is registered at the sample label, generating a traceable training dataset.

[0032] Based on the training dataset, historical operation and maintenance rules are loaded and constraint entries are extracted to form a training constraint condition group. Constraint entries include three categories: safety boundaries, execution sequence relationships, and cross-component consistency, and are bound to corresponding fields. To avoid rule conflicts, multiple constraints on the same field are first checked for consistency, then given a priority order, generating a constraint structure that can be directly injected into the training process.

[0033] Based on the training constraint set and training dataset, a "network management constraint learner" is constructed to simultaneously model component-level and cross-component-level objectives. The input is a slice sequence containing monitored trajectories and control actions, and the output is a joint result of state estimation and policy proposal. The training phase employs staged optimization: first, parameters are warmed up under unconstrained conditions, then constraint terms are gradually introduced and weight adjustments are performed; gradient updates that do not meet hard boundaries are directly discarded, and penalty terms are applied to soft consistency constraints. After several iterations, a candidate model set is obtained.

[0034] Based on the candidate model set, offline validation and cross-evaluation are performed to select stable and interpretable models, which are then assembled into a network management intelligent model library using component roles as indexes. The assembly process retains the model input field mappings and constraint references to ensure read / write consistency during the online phase. The network management intelligent model library then enters the real-time status evaluation process.

[0035] Based on the aforementioned intelligent network management model library, the current monitoring stream is accessed, and the data is read using the parameter path given by the twin object mapping table. The model of the corresponding component is then invoked to perform state calculations, outputting a set of performance metrics. The metrics cover node health, link availability, and control responsiveness, all with timestamps and window identifiers for easy subsequent alignment. To avoid the impact of short-term noise, robust aggregation is performed on metrics within the same window to obtain a snapshot of the current metrics.

[0036] Based on the performance metric set, multi-dimensional anomaly detection is initiated. The detection process is divided into three branches: threshold comparison, morphological deviation, and correlation consistency, which respectively determine fixed boundaries, temporal patterns, and cross-component collaboration. The results of the three branches are merged on the same metric snapshot to form a set of entries containing anomaly type, confidence range, and trigger field, which are then summarized into a risk warning list. Duplicate entries for the same component are merged, and the earliest and most recent trigger time pairs are recorded.

[0037] Based on the risk warning list, a similarity calculation is performed with the historical fault case library. First, the warning entries are converted into structured features, covering anomaly type, trigger field combinations, and time frame. Then, the corresponding fault features are retrieved from the case library, and structural similarity and temporal similarity are calculated separately, then weighted to obtain the matching degree. To ensure traceability, each matching degree record indicates the source warning entry.

[0038] Based on the previous matching results, a fault correlation matrix is ​​generated. The rows of the matrix correspond to early warning entries, the columns to historical fault cases, and the elements are the fused correlation weights, along with trigger field references and time segment indexes. This matrix will be directly accessed in subsequent steps of generating handling solutions and simulation verification, used to locate key fault types and select candidate solutions. It also links back to the network management intelligent model library as an entry point for labeling difficult samples during online learning. Through this connection, this step forms a closed-loop data channel between training, evaluation, and anomaly correlation, providing a structured basis for subsequent decision-making and execution.

[0039] Step S103: Generate a set of handling schemes based on the fault correlation matrix, perform simulation verification on the set of handling schemes to obtain a scheme scoring table, process the scheme scoring table through an intelligent decision engine to obtain an operation instruction sequence, track the execution status of the operation instruction sequence to obtain a feedback result set, update the digital twin model parameters based on the feedback result set, adaptively adjust the updated model parameters and the early warning threshold to obtain an optimization configuration item, and dynamically optimize the network operation strategy according to the optimization configuration item.

[0040] First, after the fault correlation matrix is ​​accessed, it is aggregated row by row to extract several historical cases with the highest correlation weight to the same early warning item, and the corresponding key fault characteristics and handling action fragments are read to generate a handling candidate pool. Based on the action fragments in the candidate pool, they are spliced ​​according to component roles and link directions to form a set of handling solutions with sequential dependencies; for action pairs with mutually exclusive resources, mutual exclusion markers are registered and alternative paths are given to obtain an executable solution sequence representation.

[0041] Based on the aforementioned set of handling schemes, a network-oriented simulation environment is constructed. Simulation input is read from the twin object mapping table, along with monitoring parameter paths and control command entry points. Operations are injected step-by-step according to the scheme sequence, and the state response is calculated using the network management intelligent model library to generate the scheme's process trajectory. The trajectory records three types of quantities: indicator fallback speed, link recovery sequence, and side effect magnitude, serving as the original basis for scheme scoring. Simultaneously, test seeds and topology snapshot indexes are retained for experimental reproduction.

[0042] Based on the aforementioned trajectory, a scheme scoring table is output. The scoring primarily uses a weighted aggregation of three types of quantities, with penalties applied to schemes that trigger mutual exclusion, resulting in a comprehensive score that balances convergence speed and stability. To avoid the influence of occasional impulses, robust statistics from multiple simulation results are used as the final indicator. The scheme scoring table refers back to the original scheme sequence and its referenced case entries, ensuring that subsequent decisions are traceable.

[0043] Based on the aforementioned scheme scoring table, the intelligent decision engine is activated, outputting a sequence of operation instructions. The decision-making process first sorts the instructions by their comprehensive scores, then reads resource usage and timing windows, and separately arranges the parallel and sequential segments across components to form a time-sequential control instruction flow. Each instruction includes the target device, parameter value range, and rollback conditions, and is simultaneously written into the observation point list to provide an entry point for subsequent execution tracking.

[0044] Based on the stated sequence of operation instructions, execution status tracking is initiated. The tracking device samples corresponding monitoring points before and after the instructions are issued, recording the achievement rate, deviation direction, and anomaly reports to form an execution log. The logs are aggregated by instruction number to generate a feedback result set, which includes records of achieved goals, reasons for non-achievement, and rollback triggers. This result set is cross-referenced with the scheme scoring table to identify deviations between simulation and field conditions.

[0045] Based on the feedback result set, the parameters of the digital twin model are updated. The update process distinguishes between rapidly adjustable state parameters and slowly evolving structural parameters; the former are adjusted incrementally within a window, while the latter are revised only after consensus is reached across multiple windows. The observation points and command pairs involved in this update are bound as evidence to ensure reusability in the next round of evaluation. The updated model and warning thresholds then enter an adaptive adjustment process.

[0046] Based on the updated model parameters and historical threshold performance, threshold adaptation is performed. The adjustment strategy reads the recent false positives and false negatives occurrence segments, applies limited-amplitude corrections to the upper and lower bounds of the thresholds, while maintaining cross-component consistency constraints; for fault types that the correlation matrix stably points to, tightening of the boundaries is allowed in the short term. The output optimization configuration items cover the threshold table and instruction limit set.

[0047] Finally, based on the aforementioned optimization configuration items, the network operation strategy is dynamically optimized. The strategy controller reads the threshold and priority, and rearranges the gain and triggering order of the control loop; for segments with parallel windows, batches are re-divided based on resource consumption and link impact range. After deployment, the strategy execution results are verified in the monitoring channel, and the verification data is fed back to the feedback result set, providing new samples for the next cycle of simulation and threshold adjustment, thus achieving closed-loop update.

[0048] As described above, the modular functional development method for the digital twin network management platform provided in this application can achieve accurate network modeling through state analysis and topology mapping. It constructs a management mechanism, combining intelligent models and fault correlation to establish reliable early warning strategies. Decision optimization is introduced, and continuous improvement in management is ensured through scheme verification and feedback adjustments. This method effectively addresses the shortcomings of traditional technologies in network modeling, fault early warning, and scheme decision-making, providing technical support for network management.

[0049] In one embodiment of the modular function development method for the digital twin network management platform of this application, the following specific content may also be included: Step S201: Establish a monitoring parameter matrix based on the communication node status data, perform normalization processing on the monitoring parameter matrix to obtain a standardized parameter table, divide the time window based on the standardized parameter table to obtain a time series sampling set, map the time series sampling set according to the topological relationship to generate a node association table, and extract features from the node association table to obtain parameter feature groups. Step S202: Generate a feature weight matrix by calculating the correlation of the parameter feature group, apply a preset threshold condition to the feature weight matrix to obtain a key feature set, construct a time series feature vector based on the key feature set, and match the time series feature vector with the topological constraint rules to generate a state feature vector.

[0050] First, after accessing the status data of the communication nodes, a monitoring parameter matrix is ​​established by grouping the data according to node identifier and acquisition channel. The rows of the matrix correspond to sampling points on a unified time axis, and the columns correspond to three types of raw parameters on the node side: electrical quantities, link quantities, and protocol stack quantities. To eliminate interference from differences in dimensions and abnormal spikes, interval scaling and outlier truncation are performed on each column to form a standardized parameter table; at the same time, the mapping relationship of the original values ​​is preserved to facilitate subsequent backtracking and interpretation.

[0051] Based on the standardized parameter table, time windows are divided according to the sampling period and heartbeat alignment results to obtain a time-series sampling set. The window length and sliding step size are set with reference to the network load cycle, and elastic windows are added during holidays or peak sales periods to cover sudden fluctuations. The sampling index, which maintains inter-node alignment within each window, serves as the time reference for cross-node analysis, and the output is a sampling slice with window number and timestamp interval.

[0052] Based on the time-series sampling set, samples from each window are mapped according to topological relationships to generate a node association table. During mapping, the uplink and downlink directions and forwarding relationships of the links are read, and the corresponding samples of adjacent nodes are aligned into directed adjacency pairs, and link attributes and congestion flags are recorded. For node pairs with multi-path forwarding, weights are allocated according to routing priority and actual forwarding proportion to form an adjacency view that can be used for feature calculation.

[0053] Based on the node association table, feature extraction is performed to obtain parameter feature groups. The extraction process includes three types of measures: trend slope within the window, short-term fluctuation amplitude, and sudden recovery time. For directed adjacency pairs, upstream and downstream response lag and cooperative change intensity are additionally calculated. All measures are referenced back to the original column and window number, forming indexable feature records that serve as input for subsequent correlation calculations.

[0054] Based on the parameter feature set, the correlation strength across nodes and within the same node is calculated to generate a feature weight matrix. The correlation strength comprehensively considers linear correlation and morphological similarity, and the reliability of cross-link strength is reduced during time periods with congestion markers. The feature weight matrix uses feature pairs as indices, with elements representing fused weights, and also includes the window number and adjacency marker used in the calculation to ensure that weights from different time periods are not confused.

[0055] Based on the aforementioned feature weight matrix, a key feature set is obtained by applying preset threshold conditions. The threshold conditions are set separately for each feature type, and a stability requirement is introduced: only features that maintain high weights within a consecutive number of windows are selected. During the screening process, redundant feature pairs are compressed, retaining the one with higher information content, and the replacement relationship is recorded to prevent subsequent duplicate calculations.

[0056] Based on the key feature set, a temporal feature vector is constructed. The vector is organized in units of windows, concatenating the key features of the same node and its adjacent nodes in a fixed order, and attaching topological hierarchy and link direction labels. To suppress the influence of abnormal windows, extreme metrics within a single window are gently truncated, and missing entries across windows are minimized through imputation to ensure consistent vector dimensions and stable alignment.

[0057] Based on the temporal feature vectors, topological constraint rules are loaded to complete the matching and generation of state feature vectors. The topological constraint rules include three types of constraints, corresponding to intra-layer connectivity consistency, cross-layer dependency closure, and loop suppression, respectively. During the matching process, if a vector segment violates intra-layer consistency, its contribution is reduced; if cross-layer dependency closure is satisfied, the metric weight consistent with the dependency chain is increased, thereby obtaining a set of state feature vectors that conform to structural semantics.

[0058] Finally, the state feature vector is directly read by the topology mapping process in step S101 and used to inject the weights of nodes and edges to generate a network structure model. Simultaneously, the key feature set and temporal sampling set are reused in the subsequent training data preparation stage as the basis for sample slicing and label alignment. Through this connection, the structured features and weight information output in steps S201 and S202 can be consistently used in subsequent hierarchical parsing, constrained training, and real-time evaluation.

[0059] In one embodiment of the modular function development method for the digital twin network management platform of this application, the following specific content may also be included: Step S301: Construct a node topology relation matrix based on state feature vectors, group the node topology relation matrix according to hierarchical division rules to obtain a hierarchical structure table, perform connectivity analysis on the hierarchical structure table to generate a node mapping graph, transform the node mapping graph through topology mapping rules to obtain a network structure model, and parse the network structure model according to functional attributes to obtain a functional component library; Step S302: Standardize and encode the component units in the functional component library to obtain a component description set; classify and label the component description set according to a preset mapping rule to obtain a component tag library; establish a virtual-real mapping rule set based on the component tag library; and match the virtual-real mapping rule set with the component association relationship to generate a twin object mapping table.

[0060] First, the state feature vector is read and the weights and directions of the nodes are organized to construct a node topology matrix. The rows and columns of the matrix correspond to unified node indices, and the elements record the interaction strength and directionality. The source window number and hierarchy label are also retained for constraint reference during hierarchical layering. To avoid abnormal window perturbations, low-confidence elements are gently truncated, and adjacency consistency is used as a constraint to reduce the impact on isolated high-weight entries.

[0061] Based on the node topology matrix, a hierarchical structure table is obtained by grouping nodes according to hierarchical partitioning rules. During grouping, physical bearer, logical forwarding, and service control are assigned to different subsets, and reference pointers are retained for cross-layer edges to ensure that dependencies are not lost in subsequent mappings. A one-to-one correspondence is established between the hierarchical structure table and the aforementioned window numbers to ensure that subsequent connectivity analysis is evaluated independently on a time-slice basis.

[0062] Based on the hierarchical structure table, connectivity analysis is performed to generate a node mapping graph. The analysis uses intra-layer connectivity, cross-layer dependency closure, and loop risk as quantitative indicators, outputting a subgraph for each layer and a set of anchor points across layers. To reduce false connections, short-lived cross-layer edges are weighted less, and candidate ranges for upstream and downstream nodes of anchor points are provided, forming a mapping framework that satisfies dependency integrity. This node mapping graph is directly referenced in subsequent mapping stages.

[0063] Based on the node mapping graph, a network structure model is obtained by applying topological mapping rules. The mapping rules preserve the topological skeleton of the intra-layer subgraph using minimum cuts, insert cross-layer anchor points in the order of dependency chains, and determine the directed attributes of edges using the direction labels recorded in the state feature vectors. To suppress local weight shifts, dual constraints of intra-layer and inter-layer balancing are employed, outputting a structured model with node roles and edge attributes.

[0064] Based on the network structure model, a functional component library is obtained by parsing according to functional attributes. During parsing, three types of responsibilities—control, forwarding, and monitoring—are first identified. Node sets that satisfy cohesion and clear boundaries are aggregated into component units. Sets with multiple entry points are split according to entry point type to ensure that the external interface of each component is singular and identifiable. Each component unit records its input / output ports, dependency chains, and the layer it belongs to, serving as input for subsequent encoding.

[0065] Based on the functional component library, standardized coding is performed to generate a component description set. The coding includes three parts: interface fields, temporal behavior summaries, and operational constraints, and synonyms are eliminated using a unified dictionary. Subsequently, the components are classified and labeled according to preset mapping rules to obtain a component tag library; the labeling process provides explicit labels for roles, levels, and observable parameters, and records the reference relationships with the node sets in the network structure model.

[0066] Based on the component tag library, a virtual-physical mapping rule set is established. This rule set reads component tags and operational constraints, provides the types of devices that can be bound, monitoring points, and control points, and defines pre-binding verifications, such as device firmware version and protocol stack compatibility. Rule items and component units form a one-to-one or one-to-many optional binding set, facilitating reuse in different deployment scenarios.

[0067] Based on the virtual-real mapping rule set, a twin object mapping table is generated by matching component associations. During matching, the continuity of dependency chains is prioritized, followed by order decisions based on monitoring point coverage. For bindings with multiple candidates, the primary binding and backup references are retained and written back to the component description set for retrieval. The twin object mapping table uses the component identifier as the primary key, lists the real device identifier, parameter reading path, and control entry point, and includes a time window and hierarchical index.

[0068] Finally, the twin object mapping table and functional component library are directly accessed during subsequent training and evaluation phases: on the one hand, they provide the correspondence between components and devices for training data slices; on the other hand, they provide parameter paths and control channels for real-time state evaluation. Through this connection, the structural semantics of the network structure model remain consistent across the data, model, and control stages.

[0069] In one embodiment of the modular function development method for the digital twin network management platform of this application, the following specific content may also be included: Step S401: The twin object mapping table is divided into sample sequence sets according to the time window. The sample sequence sets are cleaned and labeled to obtain preprocessed data sets. Training features are extracted from the preprocessed data sets to obtain a feature sample library. The feature sample library is divided into training sets according to a preset ratio to obtain a training dataset. Constraint parameters are extracted from historical operation and maintenance rules to obtain a constraint condition set. Step S402: Initialize the model parameters of the training dataset to obtain an initial model set, iteratively train the initial model set and the constraint condition group to obtain a candidate model set, perform verification and evaluation on the candidate model set to obtain a model scoring table, and select models that meet the accuracy requirements based on the model scoring table to build a network management intelligent model library.

[0070] First, the twin object mapping table is read and segmented by time window to obtain a sample sequence set covering component identifiers, device identifiers, and monitoring paths. During segmentation, the continuity of component dependency chains is maintained, and boundary events across windows are marked as transitional segments to avoid breakpoints during training. For each sequence, control entry points and corresponding observation points are simultaneously extracted to form alignable action-response pairs, providing a reference for subsequent cleaning.

[0071] Based on the sample sequence set, data cleaning and labeling are performed to generate a preprocessed data set. The cleaning process involves bounded imputation of missing test segments, source verification of abrupt transitions, and marking them as anomaly candidates; rollback markers are added to operation segments that have undergone rollback to ensure that non-monotonic responses can be identified during training. The labeling process adds state labels and load segment labels according to component roles and records the hierarchical index inherited from the mapping table, enabling subsequent feature extraction to distinguish signals at different levels.

[0072] Based on the preprocessed data set, training features are extracted and aggregated into a feature sample library. Features cover three categories of metrics within the coverage window: trend, fluctuation, and hysteresis. Trigger delay and recovery path length are calculated on action-response pairs. For cross-component links, cascaded feature fragments are aggregated according to the dependency chain order. A one-to-one reference is established between all features and the original path, facilitating the location of anomalies and backtracking in case of training failure.

[0073] Based on the aforementioned feature sample library, the dataset is divided into a training set, a validation set, and a reserved test set according to a preset ratio to obtain the training dataset. The division process maintains the chronological order and performs stratified sampling on the same event cluster to prevent a certain type of fault from being overly concentrated in a single set. Subsequently, constraint parameters are extracted from historical operation and maintenance rules and organized into constraint condition groups, including three categories: safety boundaries, execution order, and cross-component consistency, and these are bound to the feature fields.

[0074] Based on the training dataset and constraint set, the model parameters are initialized and the initial model set is assembled. The model is divided into sub-models according to component roles, and a unified input dictionary is used to adapt to different feature subsets. An unbiased distribution is used during initialization to preserve space for subsequent constraint injection. During the training phase, a "network management constraint learner" is constructed, updating parameters iteratively with mini-batch sequences. Hard boundaries are treated as inviolable conditions, and consistency is used as a soft penalty term, outputting a candidate model set.

[0075] Validation and evaluation are performed on the candidate model group to obtain a model scoring table. The evaluation calculates state estimation error, action suggestion reachability, and constraint violation rate for the three task types, and verifies generalization performance on a reserved test slice. For models with violations, the violation type and triggering fragment are recorded. The scoring table summarizes scores and constraint event counts using component roles as the primary key, supporting horizontal comparison and vertical tracing.

[0076] Based on the model scoring table, models that meet the accuracy and constraint compliance requirements are selected to construct a network management intelligent model library. Input mappings, feature selection masks, and constraint reference indexes are retained during library entry to ensure consistency between online inference and offline training. This network management intelligent model library will be directly invoked by subsequent real-time status evaluations, reading data according to the parameter paths in the twin object mapping table and outputting performance indicators. Simultaneously, the violation types recorded during verification are written back as alarm preconditions for online monitoring, forming a closed loop of training-evaluation-deployment.

[0077] In one embodiment of the modular function development method for the digital twin network management platform of this application, the following specific content may also be included: Step S501: Construct a real-time evaluation engine based on the network management intelligent model library, input real-time monitoring data into the real-time evaluation engine to perform state calculation and obtain state evaluation results, extract performance indicators from the state evaluation results according to the performance dimension to obtain a performance indicator set, compare the performance indicator set with the preset threshold rules to obtain anomaly detection results, and generate a risk warning list based on the anomaly detection results. Step S502: Perform feature vectorization processing on the warning items in the risk warning list to obtain a warning feature set, perform similarity matching between the warning feature set and the fault features in the historical fault case library to obtain a feature matching degree table, and calculate the fault association weight based on the feature matching degree table to obtain a fault association degree matrix.

[0078] First, the network management intelligent model library is read and assembled into a real-time evaluation engine, completing the binding of input interfaces and model indexes. The binding is based on the parameter paths in the twin object mapping table, mapping each monitoring stream to the model input of the corresponding component. For inputs with cross-component dependencies, time markers are aligned within the same time slice, and necessary preceding observations are supplemented. After the real-time monitoring data enters, it undergoes de-jittering and missing measurement marking processing, and is then organized into a sliding sequence according to the required window length for the model, serving as the input for state calculation.

[0079] Based on the real-time evaluation engine, window-by-window inference is performed on the sliding sequence to output the state evaluation results. During the inference phase, component-level models are executed in parallel, and models with cascading dependencies are calculated sequentially according to the dependency chain order to avoid invalid reads. Each calculation returns three types of state variables: node health, link availability, and control responsiveness, along with a time slice number and component identifier for downstream metric extraction and location.

[0080] Based on the state assessment results, performance metrics are extracted to obtain a performance metric set. The extraction transforms state variables into interval mean, fluctuation amplitude, and recovery latency; for time slices containing control actions, the response slope after the action is additionally calculated to reflect the execution arrival status. Metric entries inherit the time slice number and component identifier, and register the source model index to ensure traceability for anomaly comparisons.

[0081] Based on the performance metric set, a comparison is performed with preset threshold rules, and anomaly detection results are output. The threshold rules are derived from the aforementioned adaptive configuration and include fixed boundaries, trend deviations, and cross-component consistency expressions. During the comparison, a buffer is set for short-term out-of-bounds errors to reduce false triggers; for consistency checks, relevant metrics are collected from the same dependency chain for intra-group judgment, generating a record set with anomaly type and trigger field.

[0082] Based on the anomaly detection results, a risk warning list is compiled. Similar anomalies in adjacent time slices are merged into consecutive events, and the first and most recent trigger times are recorded. For cross-component consistency anomalies, they are aggregated into composite entries according to the dependency chain, clearly defining the order of upstream triggering and downstream response. Each warning refers to a set of performance metrics and status assessment results, with a clear source.

[0083] Based on the risk warning list, the warning items are vectorized to obtain a warning feature set. Vectorization encodes the anomaly type, trigger field combination, time frame, and dependency location. For composite entries, segmentation and concatenation are used to preserve the relative order of upstream and downstream components. To reduce noise, weak segments that appear only once and last for less than a window are removed, retaining structurally stable feature entries.

[0084] Based on the aforementioned early warning feature set, similarity matching is performed with the fault features in the historical fault case database to obtain a feature matching table. Matching is performed in two ways: structural coding similarity and temporal morphological similarity, and these are merged at the entry level. This embodiment provides a fusion criterion, using the following formula to calculate the final weight: R = a·S + b·T − c·C.

[0085] In the formula, R is the association weight after fusion, S is the structural similarity score, T is the temporal morphological similarity score, and C is the degree of conflict with known mutually exclusive patterns; a, b, and c are non-negative coefficients, and their sum is limited to a finite interval to stabilize the scale.

[0086] Based on the feature matching table, the R-values ​​of each warning item and historical cases are calculated and filled into the corresponding row and column positions, outputting a fault correlation matrix. This matrix retains the decomposition records of S, T, and C and the source index, facilitating subsequent interpretation and playback. Subsequently, the fault correlation matrix is ​​directly read by the disposal plan generation module to select highly correlated cases in a row-first manner and extract key fault features; at the same time, the risk warning list serves as the seed input for initial condition playback in the simulation verification stage, ensuring the consistency and reproducibility of the data path from assessment to solution decision.

[0087] In one embodiment of the modular function development method for the digital twin network management platform of this application, the following specific content may also be included: Step S601: Perform cluster analysis on the fault correlation matrix to obtain fault type groups, extract key fault features based on the fault type groups to obtain a feature pattern set, perform rule matching between the feature pattern set and the disposal knowledge base to obtain a candidate solution group, perform feasibility evaluation on the candidate solution group to obtain a disposal solution set, and input the disposal solution set into the simulation environment for verification testing to obtain a solution scoring table. Step S602: Prioritize the scheme scoring table according to preset decision rules to obtain a scheme decision sequence; perform resource constraint analysis on the scheme decision sequence to obtain an execution constraint table; schedule and orchestrate the schemes based on the execution constraint table to obtain a scheduling strategy set; and convert the scheduling strategy set into specific operation steps to generate an operation instruction sequence.

[0088] First, the fault correlation matrix is ​​read and standardized within the same time slice. Cluster analysis is then performed on the row vectors to obtain fault type groups. Before clustering, noise reduction is performed based on the sparse distribution and weight distribution of the matrix rows, and low-contribution columns are compressed to avoid interference from tail cases. During clustering, the center vector and member index of each cluster are retained, and the time slice and dependency chain identifier are recorded as references for feature extraction.

[0089] Based on the fault type groups, key fault features are extracted to form a feature pattern set. During extraction, subthreshold suppression is performed on the center vector of each class, retaining only entries with stable contributions and consistency across the window. Then, combined with the common triggering fields and temporal patterns of cluster members, a combined expression of structural and temporal encoding is generated. The feature pattern set corresponds one-to-one with the fault type groups and carries an index of the source warning entries for easy playback.

[0090] Based on the aforementioned feature pattern set, rule matching of the knowledge base is initiated and processed to obtain a candidate solution group. The matching process is carried out in two paths: structural encoding and temporal encoding, requiring simultaneous satisfaction of three constraints: device boundary, link direction, and dependency chain closure. For cases where the same pattern matches multiple knowledge rules, the dependency chain coverage is used as the decision key, retaining the rule entry with more complete coverage. Candidate solutions are represented in the form of action sequences, accompanied by a list of preconditions.

[0091] Based on the candidate solution group, a feasibility assessment is conducted, resulting in a set of disposal solutions. The assessment reads current resource usage, control channel reachability, and security boundaries. Actions requiring parallel execution are marked with mutual exclusion relationships, and alternative entry points are provided for actions with potentially limited access. For cross-component cascading solutions, the preceding actions must have observable and verifiable completion markers; if these are not met, the solution is downgraded to an alternative. After screening, a set of disposal solutions that can be deployed is formed.

[0092] Based on the aforementioned set of handling schemes, a simulation environment was used for verification testing to obtain a scheme scoring table. The simulation used the monitoring path and control entry point provided by the twin object mapping table as the interface, replaying actions according to the scheme sequence and calling the network management intelligent model library to calculate the state response. The scoring dimensions included recovery speed, stability, and side effects. Simultaneously, test seeds, topology snapshots, and observation point evidence were recorded to ensure consistency between the reproduced experiments and the interpretation of results.

[0093] Based on the proposed solution scoring table, a priority ranking is performed according to preset decision-making rules, outputting a solution decision sequence. The ranking is primarily based on the comprehensive score, with conflict resolution for mutually exclusive solutions, and resource consumption and execution window fit as secondary ranking factors. For solutions with similar scores, the achievement rate from historical execution logs is used as a weighting factor to elevate the position of more reliable solutions. The ranking results retain references to the source scoring entries.

[0094] Based on the proposed decision sequence, resource constraint analysis is performed to obtain an execution constraint table. The analysis considers three aspects: device concurrency capacity, link maintenance time slots, and operational mutual exclusion. Each step is labeled with a set of possible parallelism and a set of mandatory sequential operations, and the allowed number of retries and rollback conditions are given. The execution constraint table is aligned with the proposed decision sequence, providing hard constraints for subsequent orchestration.

[0095] Based on the execution constraint table, the scheme is scheduled and orchestrated to form a scheduling strategy set. The orchestration batches parallelizable sets within the same window and queues mandatory sequential sets according to their dependency order; synchronization points are inserted for cross-domain actions to ensure that subsequent steps are only advanced after observations are received. For steps with alternative entry points, degradation paths and switching trigger points are included in the strategy to ensure robustness in field implementation.

[0096] Based on the aforementioned scheduling strategy set, instruction conversion is completed, generating an operation instruction sequence. During conversion, each strategy unit is mapped to a specific target device, parameter value range, and verification rule, and observation points and rollback entry points are bound together to form a control flow that can be directly issued. The operation instruction sequence is used as the main index in subsequent execution tracking, and its number will maintain a consistent reference relationship with simulation scores, resource constraints, and fault type groups, facilitating closed-loop verification and continuous iteration.

[0097] In one embodiment of the modular function development method for the digital twin network management platform of this application, the following specific content may also be included: Step S701: Parse the operation instruction sequence according to the execution time sequence to obtain the execution task table, construct monitoring and collection points for the execution task table to obtain the monitoring parameter group, collect execution process data based on the monitoring parameter group to obtain the state sequence set, perform deviation analysis between the state sequence set and the expected target to obtain the execution evaluation table, summarize the results of the execution evaluation table to generate a feedback result set, and extract model correction parameters based on the feedback result set to obtain the parameter update set; Step S702: Input the parameter update set into the model optimizer to adjust the parameters and obtain an optimized parameter group. Perform threshold calibration on the optimized parameter group to obtain a threshold adjustment table. Generate a configuration update instruction based on the threshold adjustment table to obtain an optimized configuration item. Apply the optimized configuration item to the policy controller to adjust the policy and obtain a running policy group. Deploy and verify the running policy group to complete the network running policy optimization.

[0098] First, the sequence of operation instructions is expanded by number and timestamp, and parsed into an execution task table. The parsing process breaks down each instruction into four fields: target device, action link, parameter range, and rollback condition, and labels the predecessors and successors according to dependencies. To ensure that subsequent observations can be effectively implemented, the required observation granularity and sampling window length are registered within each task item, forming a task list that can directly drive data acquisition.

[0099] Based on the execution task table, monitoring and acquisition points are constructed and aggregated into monitoring parameter groups. The selection of acquisition points is constrained by the monitoring paths in the twin object mapping table, mapping the device-side sensor items and link-side counter items to specific interfaces respectively; for windows with parallel tasks, synchronization markers are added to ensure observation time alignment. The monitoring parameter groups correspond one-to-one with the task items, including the observation frequency and fault tolerance upper limit, for on-site acquisition and control.

[0100] Based on the monitoring parameter set, execution process data is collected and organized into a state sequence set. During data collection, jitter and short-term packet loss are marked, and no strong interpolation is performed to avoid masking the true fluctuations. The state sequence set uses the task number as the primary key and records three segments of data: the baseline before execution, the trajectory during execution, and the recovery segment after execution. It also includes event markers for rollback triggers and anomaly reporting for subsequent deviation analysis.

[0101] Based on the state sequence set, a deviation analysis is performed to obtain an execution evaluation table. The deviation analysis is based on the parameter range and achievement threshold conditions of the task items, calculating three types of indicators: arrival time, steady-state deviation, and residual fluctuations. For tasks with rollback conditions, the difference in the recovered trajectory after rollback is additionally calculated to distinguish between reversible disturbances and structural problems. Evaluation items are referenced back to their corresponding observation points and time windows to ensure accurate positioning.

[0102] Based on the execution evaluation form, the results are summarized to generate a feedback result set. Multiple tasks in the same batch are aggregated according to their task chains, and the achievement status, reasons for non-achievement, and scope of impact are output. For task groups with mutual conflicts, a conflict sequence is provided. The feedback result set retains a comparison index with the simulation score to compare the consistency differences between the actual situation and the simulation.

[0103] Based on the feedback result set, model correction parameters are extracted to obtain a parameter update set. The extracted actions are differentiated into two paths: fast-state types and slow-structure types. The former estimates response gain and hysteresis correction from arrival time and steady-state deviation, while the latter estimates link capacity correction and dependency strength adjustment from consistent residual deviations across multiple batches. The parameter update set records the applicable components, applicable windows, and evidence entry indexes to ensure that the updates are replayable and auditable.

[0104] Based on the parameter update set, it is input into the model optimizer for parameter adjustment, resulting in an optimized parameter set. The optimizer uses incremental updates for fast state parameters and a sliding window consistency threshold for structural parameters to avoid overfitting to single field data. After offline consistency verification, the optimized parameters generate a threshold adjustment table through a threshold calibration process. The calibration logic references the distribution of recent false alarms and false negatives to limit the magnitude of single adjustments and maintain cross-component consistency.

[0105] Based on the threshold adjustment table, configuration update instructions are generated, forming optimized configuration items. These instructions cover model weights, effective windows, and threshold configurations for each performance dimension, and include rollback strategies and effectiveness detection hooks to ensure rapid verification and, if necessary, retraction after issuance. Optimized configuration items are bound to control entry points in the twin object mapping table, facilitating direct reading and writing by the policy controller.

[0106] Based on the aforementioned optimization configuration items, they are applied to the policy controller for policy adjustment, resulting in a running policy group. The adjustments include control loop priority, batch partitioning of parallel windows, and instruction limiting strategies; synchronization points are inserted for cross-domain dependent segments to ensure that subsequent actions are only initiated after observations are achieved. The running policy group undergoes a dry run verification before being deployed to the control plane to check compatibility and mutual exclusion with existing network resources.

[0107] Finally, the deployment verification of the aforementioned operation strategy group is performed to complete the network operation strategy optimization. During the verification phase, coverage is gradually expanded in a small-scale, gray-scale manner, and comparative data before and after deployment is collected on the monitoring channel and backfilled into the feedback result set and state sequence set. Through this closed loop, the actual performance of the optimized parameter group and threshold adjustment table is continuously tracked, and new deviations will trigger the extraction of the parameter update set again, achieving a stable cycle from execution, evaluation, to re-optimization.

[0108] To effectively address the shortcomings of traditional technologies in network modeling, fault early warning, and solution decision-making, and to provide technical support for network management, this application provides an embodiment of a modular function development apparatus for a digital twin network management platform, used to implement all or part of the modular function development method for the aforementioned digital twin network management platform. See [link to relevant documentation]. Figure 2 The modular function development device of the digital twin network management platform specifically includes the following components: The object mapping module 10 is used to extract the node operation parameter set and network topology information set based on the communication node status data, perform time-series correlation analysis on the operation parameter set and the network topology information set to obtain a state feature vector, generate a network structure model by processing the state feature vector through topology mapping, perform hierarchical parsing on the network structure model to obtain a functional component library, and establish a twin object mapping table based on the functional component library. The fault prediction module 20 is used to divide the twin object mapping table into training datasets according to a preset training strategy, construct training constraint condition groups according to historical operation and maintenance rules, perform constraint training on the training dataset to generate a network management intelligent model library, apply the network management intelligent model library to real-time status evaluation to obtain a performance index set, perform multi-dimensional anomaly detection on the performance index set to obtain a risk warning list, and perform similarity calculation between the risk warning list and the historical fault case library to obtain a fault correlation matrix. The model optimization module 30 is used to generate a set of disposal schemes based on the fault correlation matrix, perform simulation verification on the set of disposal schemes to obtain a scheme scoring table, process the scheme scoring table through an intelligent decision engine to obtain an operation instruction sequence, track the execution status of the operation instruction sequence to obtain a feedback result set, update the digital twin model parameters based on the feedback result set, adaptively adjust the updated model parameters and the early warning threshold to obtain an optimization configuration item, and dynamically optimize the network operation strategy according to the optimization configuration item.

[0109] As described above, the modular functional development apparatus for the digital twin network management platform provided in this application embodiment can achieve accurate network modeling through state analysis and topology mapping. It constructs a management mechanism, combining intelligent models and fault correlation to establish reliable early warning strategies. Decision optimization is introduced, and continuous improvement in management is ensured through scheme verification and feedback adjustments. This method effectively addresses the shortcomings of traditional technologies in network modeling, fault early warning, and scheme decision-making, providing technical support for network management.

[0110] This invention also provides a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the modular function development method of the digital twin network management platform.

[0111] This invention also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the modular function development method of the digital twin network management platform described above.

[0112] This invention also provides a computer program product, which includes a computer program that, when executed by a processor, implements the modular function development method of the digital twin network management platform described above.

[0113] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0114] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0115] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0116] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0117] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above descriptions are merely specific embodiments of the present invention and are not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A modular functional development method for a digital twin network management platform, characterized in that, The method includes: Based on the communication node status data, extract the node operation parameter set and network topology information set, perform time series correlation analysis on the operation parameter set and the network topology information set to obtain the state feature vector, generate the network structure model through topology mapping processing of the state feature vector, perform hierarchical parsing on the network structure model to obtain the functional component library, and establish a twin object mapping table based on the functional component library. The twin object mapping table is divided into training datasets according to a preset training strategy. Training constraint condition groups are constructed according to historical operation and maintenance rules. Constraint training is performed on the training dataset to generate a network management intelligent model library. The network management intelligent model library is applied to real-time status evaluation to obtain a performance index set. Multi-dimensional anomaly detection is performed on the performance index set to obtain a risk warning list. The risk warning list is compared with the historical fault case library to calculate the similarity and obtain a fault correlation matrix. A set of handling schemes is generated based on the fault correlation matrix. The set of handling schemes is simulated and verified to obtain a scheme scoring table. The scheme scoring table is processed by an intelligent decision engine to obtain an operation instruction sequence. The execution status of the operation instruction sequence is tracked to obtain a feedback result set. The parameters of the digital twin model are updated based on the feedback result set. The updated model parameters and the early warning threshold are adaptively adjusted to obtain an optimization configuration item. The network operation strategy is dynamically optimized according to the optimization configuration item.

2. The modular functional development method for the digital twin network management platform according to claim 1, characterized in that, The process involves extracting a set of node operating parameters and a set of network topology information based on communication node state data, and performing time-series correlation analysis on the set of operating parameters and the set of network topology information to obtain a state feature vector, including: A monitoring parameter matrix is ​​established based on the status data of communication nodes. The monitoring parameter matrix is ​​normalized to obtain a standardized parameter table. A time window is divided based on the standardized parameter table to obtain a time series sampling set. The time series sampling set is mapped according to the topological relationship to generate a node association table. The node association table is feature extracted to obtain parameter feature groups. The parameter feature group is used to generate a feature weight matrix through correlation calculation. The feature weight matrix is ​​then filtered by applying a preset threshold condition to obtain a key feature set. A time-series feature vector is constructed based on the key feature set. The time-series feature vector is then matched with topological constraint rules to generate a state feature vector.

3. The modular functional development method for the digital twin network management platform according to claim 1, characterized in that, The process of generating a network structure model from the state feature vector through topological mapping, performing hierarchical parsing on the network structure model to obtain a functional component library, and establishing a twin object mapping table based on the functional component library includes: A node topology relation matrix is ​​constructed based on state feature vectors. The node topology relation matrix is ​​grouped according to hierarchical division rules to obtain a hierarchical structure table. The connectivity of the hierarchical structure table is analyzed to generate a node mapping graph. The node mapping graph is transformed through topology mapping rules to obtain a network structure model. The network structure model is parsed according to functional attributes to obtain a functional component library. The component units in the functional component library are standardized and encoded to obtain a component description set. The component description set is classified and labeled according to a preset mapping rule to obtain a component tag library. A virtual-real mapping rule set is established based on the component tag library. The virtual-real mapping rule set is matched with the component association relationship to generate a twin object mapping table.

4. The modular functional development method for the digital twin network management platform according to claim 1, characterized in that, The step of dividing the twin object mapping table into training datasets according to a preset training strategy, constructing training constraint condition groups based on historical operation and maintenance rules, and performing constraint training on the training datasets to generate a network management intelligent model library includes: The twin object mapping table is divided into sample sequence sets according to time windows. The sample sequence sets are cleaned and labeled to obtain preprocessed data sets. Training features are extracted from the preprocessed data sets to obtain feature sample libraries. The feature sample libraries are divided into training sets according to preset ratios to obtain training datasets. Constraint parameters are extracted from historical operation and maintenance rules to obtain constraint condition sets. The initial model set is obtained by initializing the model parameters of the training dataset. The initial model set and the constraint condition group are iteratively trained to obtain the candidate model set. The candidate model set is validated and evaluated to obtain the model scoring table. Based on the model scoring table, models that meet the accuracy requirements are selected to build the network management intelligent model library.

5. The modular functional development method for the digital twin network management platform according to claim 1, characterized in that, The process involves applying the network management intelligent model library to real-time status assessment to obtain a performance indicator set, performing multi-dimensional anomaly detection on the performance indicator set to obtain a risk warning list, and calculating the similarity between the risk warning list and a historical fault case library to obtain a fault correlation matrix, including: A real-time evaluation engine is built based on the network management intelligent model library. Real-time monitoring data is input into the real-time evaluation engine to calculate the state and obtain the state evaluation result. The state evaluation result is then used to extract indicators according to the performance dimension to obtain a performance indicator set. The performance indicator set is compared with a preset threshold rule to obtain the anomaly detection result. A risk warning list is generated based on the anomaly detection result. The warning items in the risk warning list are processed by feature vectorization to obtain a warning feature set. The warning feature set is matched with the fault features in the historical fault case library to obtain a feature matching degree table. Based on the feature matching degree table, the fault association weight is calculated to obtain the fault association degree matrix.

6. The modular functional development method for the digital twin network management platform according to claim 1, characterized in that, The process involves generating a set of remedial solutions based on the fault correlation matrix, performing simulation verification on the set of solutions to obtain a solution scoring table, and processing the solution scoring table through an intelligent decision engine to obtain a sequence of operation instructions, including: Cluster analysis is performed on the fault correlation matrix to obtain fault type groups. Key fault features are extracted based on the fault type groups to obtain a feature pattern set. The feature pattern set is matched with the disposal knowledge base to obtain a candidate solution group. The feasibility of the candidate solution group is evaluated to obtain a disposal solution set. The disposal solution set is input into the simulation environment for verification testing to obtain a solution scoring table. The scheme scoring table is sorted by priority according to preset decision rules to obtain a scheme decision sequence. Resource constraint analysis is performed on the scheme decision sequence to obtain an execution constraint table. Based on the execution constraint table, the schemes are scheduled and orchestrated to obtain a scheduling strategy set. The scheduling strategy set is converted into specific operation steps to generate an operation instruction sequence.

7. The modular functional development method for the digital twin network management platform according to claim 1, characterized in that, The execution status tracking of the operation instruction sequence yields a feedback result set. Based on this feedback result set, the digital twin model parameters are updated. The updated model parameters are adaptively adjusted with the warning threshold to obtain an optimized configuration item. The network operation strategy is then dynamically optimized based on this optimized configuration item, including: The operation instruction sequence is parsed according to the execution time sequence to obtain the execution task table. Monitoring and collection points are constructed on the execution task table to obtain the monitoring parameter group. Execution process data is collected based on the monitoring parameter group to obtain the state sequence set. Deviation analysis is performed between the state sequence set and the expected target to obtain the execution evaluation table. The results of the execution evaluation table are summarized to generate a feedback result set. Model correction parameters are extracted based on the feedback result set to obtain the parameter update set. The parameter update set is input into the model optimizer to adjust the parameters and obtain an optimized parameter group. The optimized parameter group is then subjected to threshold calibration to obtain a threshold adjustment table. Based on the threshold adjustment table, a configuration update instruction is generated to obtain an optimized configuration item. The optimized configuration item is applied to the policy controller to adjust the policy and obtain an operating policy group. The operating policy group is then deployed and verified to complete the network operating policy optimization.

8. A modular functional development device for a digital twin network management platform, characterized in that, The device includes: The object mapping module is used to extract the node operation parameter set and network topology information set based on the communication node status data, perform time-series correlation analysis on the operation parameter set and the network topology information set to obtain a state feature vector, generate a network structure model by processing the state feature vector through topology mapping, perform hierarchical parsing on the network structure model to obtain a functional component library, and establish a twin object mapping table based on the functional component library. The fault prediction module is used to divide the twin object mapping table into training datasets according to a preset training strategy, construct training constraint condition groups according to historical operation and maintenance rules, perform constraint training on the training dataset to generate a network management intelligent model library, apply the network management intelligent model library to real-time status evaluation to obtain a performance index set, perform multi-dimensional anomaly detection on the performance index set to obtain a risk warning list, and perform similarity calculation between the risk warning list and the historical fault case library to obtain a fault correlation matrix. The model optimization module is used to generate a set of disposal schemes based on the fault correlation matrix, perform simulation verification on the set of disposal schemes to obtain a scheme scoring table, process the scheme scoring table through an intelligent decision engine to obtain an operation instruction sequence, track the execution status of the operation instruction sequence to obtain a feedback result set, update the digital twin model parameters based on the feedback result set, adaptively adjust the updated model parameters and the early warning threshold to obtain an optimization configuration item, and dynamically optimize the network operation strategy according to the optimization configuration item.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the steps of the modular function development method of the digital twin network management platform according to any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the modular function development method of the digital twin network management platform according to any one of claims 1 to 7.

Images