Optical cable network natural language operation and maintenance method and system

Abstract

The application provides an optical cable network natural language operation and maintenance method and system, belongs to the technical field of optical cable communication, is applied to an optical cable network natural language operation and maintenance system, collects multi-dimensional data including optical cable vibration, temperature, environment, optical power and OTDR link attenuation through an enhanced optical fiber distribution robot of a physical perception layer, encapsulates into a multi-sensing data package MSDP, reports after data alignment relying on clock synchronization, carries out unified semantic modeling, fusion storage and semantic reasoning on the MSDP data package and network resource data, GIS geographic information and service data through a data fusion layer, constructs an optical cable operation and maintenance knowledge graph, receives natural language instructions of operation and maintenance personnel through an application interaction layer, completes intention understanding, knowledge retrieval, task planning and execution scheduling by the intelligent decision layer through multi-agent collaboration, simultaneously carries out alarm correlation analysis and health degree scoring, generates a preventive maintenance work order, and thus reduces the operation and maintenance threshold, improves the fault early warning and emergency recovery efficiency.

Description

Technical Field

[0001] This application belongs to the field of optical fiber communication technology, and specifically relates to a natural language operation and maintenance method and system for optical fiber networks. Background Technology

[0002] With the deepening of digital transformation in the communications industry, optical fiber networks have become the fundamental infrastructure for various critical services. Especially in industries such as highways, rail transit, power, and finance, optical fiber networks are massive in scale and widely distributed, making operation and maintenance increasingly difficult and complex. Taking highway optical fiber networks as an example, their total mileage is long, sites are scattered, and the environment along the route is complex. Once a fault such as a fiber optic cable break occurs, emergency repairs often take several hours or even longer, seriously affecting business continuity and social service levels. Currently, the operation and maintenance of optical fiber networks mainly suffers from the following technical deficiencies: (1) Existing fiber optic distribution robots can automatically perform fiber optic bridging and digitally manage network topology resources, but their sensing capabilities are limited to optical path signal receiving power and OTDR monitoring. They lack multi-dimensional sensing capabilities for external environmental vibration and temperature, and cannot provide a more comprehensive data foundation for fiber optic cable health assessment. (2) Distributed Vibration Monitoring (DVM) and Distributed Temperature Monitoring (DTM) are deployed as independent systems and lack deep integration with the fiber optic cable operation and maintenance scheduling platform, making it difficult to form a closed-loop process of "perception-decision-execution". (3) The operation and maintenance system interface is professional and complex to operate. Multiple subsystems are isolated from each other, and information silos lead to low emergency dispatch efficiency. (4) The large language model and physical execution layer are not deeply integrated. The intelligent agent can only provide suggestions and cannot directly drive the robot to complete physical actions such as fiber optic jumps and route switching. (5) Multi-source heterogeneous data lacks unified modeling, the decision-making basis is incomplete, and the false alarm and false alarm rates are high. Summary of the Invention

[0003] In view of this, the purpose of this application is to provide a natural language operation and maintenance method and system for optical cable networks. Based on the fusion of multi-parameter sensing optical fiber distribution robot and large model intelligent agent, it realizes a closed loop of multi-dimensional perception, data fusion, intelligent decision-making, physical execution and natural language feedback, thereby reducing the operation and maintenance threshold and improving the efficiency of fault early warning and emergency recovery.

[0004] Firstly, this application provides a natural language operation and maintenance method for optical fiber networks, applied to an optical fiber network natural language operation and maintenance system. The system includes a physical perception layer, a data fusion layer, an intelligent decision-making layer, and an application interaction layer. The method includes the following steps: The enhanced fiber optic distribution robot at the physical sensing layer collects multi-dimensional data including fiber optic cable vibration, temperature, environment, optical power, and OTDR link attenuation, and encapsulates the collected multi-dimensional data into standardized MSDP data packets and reports them to the data fusion layer. By using the data fusion layer, MSDP data packets, network resource data, GIS geographic information, and business data are subjected to unified semantic modeling, fusion storage, and semantic reasoning to construct a knowledge graph for optical cable operation and maintenance. The application interaction layer receives natural language commands from maintenance personnel. The intelligent decision layer, based on real-time collected multi-dimensional data and the constructed optical cable maintenance knowledge graph, completes intent understanding, knowledge retrieval, task planning and execution scheduling through multi-agent collaboration. At the same time, it performs alarm correlation analysis and health scoring to generate preventive maintenance work orders.

[0005] In some embodiments, the enhanced fiber optic distribution robot includes an intelligent fiber optic distribution robot, a distributed vibration sensing module (DVM), a distributed temperature sensing module (DTM), and an infrastructure environment sensing module (IEM). Among them, the intelligent fiber optic distribution robot collects optical power and OTDR link attenuation data; the DVM collects optical cable vibration data; the DTM collects optical cable temperature distribution, local hot spots and temperature change rate data based on the Raman scattering principle; the IEM collects cabinet temperature and humidity, air pressure and smoke alarm status data; and all collected data are encapsulated into JSON format data packets according to the MSDP protocol. The data packets are reported to the edge computing node via the MQTT protocol, and after verification, they are forwarded to the data fusion layer.

[0006] In some embodiments, constructing the optical cable operation and maintenance knowledge graph includes the following steps: The three-layer architecture of the optical cable network operation and maintenance data model ONOM is used for semantic modeling, defining network physical entities, sensor time-series entities, operation and maintenance event entities and their interrelationships. The system receives and parses MSDP data packets, instantiates entities according to the ONOM model, and stores them in the time-series database and graph database. Then, it performs time-series alignment of multi-source sensor data through hierarchical sliding window aggregation, generates multi-dimensional feature vectors, and stores them in a relational database. Finally, it performs semantic analysis based on OWL inference rules to identify optical cable fault risks and writes the inference results back to the knowledge graph. Perform the initial construction of the optical cable operation and maintenance knowledge graph, and update the data synchronously during system operation.

[0007] In some embodiments, the intelligent decision-making layer includes an intent-understanding agent, a knowledge-retrieval agent, a task-planning agent, and an execution-monitoring agent, which interact through a message bus and work collaboratively in a Supervisor-Worker mode. Specifically, the intent-based intelligent agent classifies natural language instructions, extracts key entities, and performs semantic parsing based on the dialogue context; the knowledge retrieval intelligent agent uses a combination of structured query and vector similarity retrieval to retrieve relevant information from knowledge graphs and sensor databases to generate RAG context; the task planning intelligent agent combines the parsing results and retrieval information to generate a task plan containing tool calls and step dependencies through thought chain reasoning; and the execution monitoring intelligent agent schedules the execution according to the task plan and detects execution anomalies in real time and adjusts adaptively.

[0008] In some embodiments, alarm correlation analysis is performed in the following manner, including the following steps: Multiple types of alarms, including vibration, temperature, optical power, OTDR, and network management services, will be uniformly converted into standard alarm objects. Based on time and space dimensions, alarms from the same source are aggregated and compressed to generate alarm clusters, and the Drools rule engine calls the pre-built rule base for correlation reasoning; among them, alarm clusters without matching rules are combined with historical fault cases and semantic reasoning is supplemented by large language model LLM. By integrating the analysis results of the Drools rule engine and the Large Language Model (LLM), a composite alarm is output, which includes type, root cause, impact on business, handling suggestions, confidence level, and inference chain.

[0009] In some embodiments, the alarm correlation analysis further includes the following steps: The rule base is dynamically updated by combining three approaches: manual data entry and translation, fault review and summarization, and offline data mining, along with a large language model (LLM).

[0010] In some embodiments, health scores are determined by the following steps: Taking optical cable segments as units, multi-dimensional time-series statistical data corresponding to vibration, temperature, optical power, and OTDR are extracted from the time-series database; The collected multidimensional time-series data were preprocessed and feature mined. Among them, the isolated forest algorithm was used to identify and complete abnormal data, the trend components were obtained through time-series decomposition, and the vibration features were clustered and classified using the K-means algorithm. The time-series prediction model is used to perform rolling predictions on the trend components of data in each dimension; among them, the optical cable operation status is determined according to the preset health threshold and risk warnings are triggered. A weighted linear model with configurable weights is used to calculate the overall health score of the optical cable; when the overall health score of the optical cable is lower than a set threshold, a preventive maintenance work order is generated.

[0011] Secondly, this application also provides a natural language operation and maintenance system for optical fiber networks, including a physical perception layer, a data fusion layer, an intelligent decision-making layer, and an application interaction layer, for collaboratively executing the steps of the natural language operation and maintenance method for optical fiber networks described in any one of the first aspects.

[0012] Thirdly, this application also provides an electronic device, including: a processor, a memory, and a bus, wherein the memory stores machine-readable instructions executable by the processor, and when the electronic device is running, the processor communicates with the memory via the bus, and when the machine-readable instructions are executed by the processor, the steps of the optical cable network natural language operation and maintenance method described in any of the first aspects are performed.

[0013] Fourthly, this application also provides a computer-readable storage medium storing a computer program, which, when executed by a processor, performs the steps of the optical fiber network natural language operation and maintenance method described in any one of the first aspects.

[0014] The optical fiber network natural language operation and maintenance method and system described in this application enables operation and maintenance personnel to directly interact with the optical fiber network emergency dispatch and resource management system through the collaborative operation of the physical perception layer, data fusion layer, intelligent decision-making layer and application interaction layer, thereby reducing the operation threshold and improving operation and maintenance efficiency and intelligence level. Attached Figure Description

[0015] To more clearly illustrate the technical solutions of the embodiments of this application, the accompanying drawings used in the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of the present invention and should not be regarded as a limitation of the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0016] Figure 1 A flowchart of the natural language operation and maintenance method for optical fiber networks according to an embodiment of this application is shown; Figure 2 A structural block diagram of the optical fiber network natural language operation and maintenance system according to an embodiment of this application is shown; Figure 3 A schematic diagram of the structure of the electronic device described in an embodiment of this application is shown. Detailed Implementation

[0017] 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. It should be understood that the accompanying drawings in this application are for illustrative and descriptive purposes only and are not intended to limit the scope of protection of this application. Furthermore, it should be understood that the schematic drawings are not drawn to scale. The flowcharts used in this application illustrate operations implemented according to some embodiments of this application. It should be understood that the operations in the flowcharts may not be implemented in sequence, and steps without logical contextual relationships may be reversed or implemented simultaneously. In addition, those skilled in the art, guided by the content of this application, may add one or more other operations to the flowcharts, or remove one or more operations from the flowcharts.

[0018] Furthermore, the described embodiments are merely some, not all, of the embodiments of this application. The components of the embodiments of this application described and illustrated herein can typically be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of this application provided in the accompanying drawings is not intended to limit the scope of the claimed application, but merely to illustrate selected embodiments of the application. All other embodiments obtained by those skilled in the art based on the embodiments of this application without inventive effort are within the scope of protection of this application.

[0019] It should be noted that the term "comprising" will be used in the embodiments of this application to indicate the presence of the features declared thereafter, but does not exclude the addition of other features.

[0020] In view of the technical problems raised in the background, this application provides a natural language operation and maintenance method and system for optical cable networks. Based on the fusion of multi-parameter sensing optical fiber distribution robot and large model intelligent agent, it realizes a closed loop of multi-dimensional perception, data fusion, intelligent decision-making, physical execution and natural language feedback, thereby reducing the operation and maintenance threshold and improving the efficiency of fault early warning and emergency recovery.

[0021] See the instruction manual appendix Figure 1 Included with instruction manual Figure 2 This application provides a natural language operation and maintenance method for optical fiber networks, applied to an optical fiber network natural language operation and maintenance system. The system includes a physical perception layer, a data fusion layer, an intelligent decision-making layer, and an application interaction layer. The method includes the following steps: S1. The enhanced fiber optic distribution robot through the physical sensing layer collects multi-dimensional data including fiber optic cable vibration, temperature, environment, optical power and OTDR link attenuation, and encapsulates the collected multi-dimensional data into standardized MSDP data packets and reports them to the data fusion layer. S2. Through the data fusion layer, MSDP data packets, network resource data, GIS geographic information, and business data are uniformly semantically modeled, fused, stored, and semantically reasoned to construct a knowledge graph for optical cable operation and maintenance. S3. The application interaction layer receives natural language commands from maintenance personnel. The intelligent decision layer, based on real-time collected multi-dimensional data and the constructed optical cable maintenance knowledge graph, completes intent understanding, knowledge retrieval, task planning and execution scheduling through multi-agent collaboration. At the same time, it performs alarm correlation analysis and health scoring to generate preventive maintenance work orders.

[0022] Step S1 mainly addresses the problem of the single sensing dimension of existing fiber optic distribution robots. This application adopts an enhanced fiber optic distribution robot that integrates a distributed vibration sensing module (DVM), a distributed temperature sensing module (DTM), and an infrastructure environment monitoring module (IEM). Together with the original optical power monitoring and OTDR detection, it forms a multi-dimensional sensing system responsible for collecting data on the physical state of the optical cable and encapsulating it into a multi-sensor data packet (MSDP) format for transmission to the data fusion layer.

[0023] Specifically, the Distributed Vibration Sensing Module (DVM) is an integrated subsystem within the fiber optic cabling robot. Based on the principle of phase-sensitive optical time-domain reflectometry, it emits narrow-linewidth pulsed laser light into the fiber optic cable under test. By analyzing the phase change of the backscattered Rayleigh light, it achieves continuous vibration state sensing along the cable. The collected data includes: vibration intensity distribution along the cable (physical quantity: acceleration, unit: dB; data format: a one-dimensional array with a length of L / ΔR for each sample, where L represents the total length of the cable and ΔR represents the spatial resolution); extraction of the time-frequency matrix of the corresponding frequency band through STFT transformation to generate the time-frequency characteristics of the vibration event; the physical coordinates of the vibration event; and vibration waveform segments (the original time-domain waveform within a set time period before and after the event trigger, used for pattern recognition).

[0024] The Distributed Temperature Sensing Module (DTM) is based on the Raman scattering principle and acquires temperature information by measuring the intensity ratio of the anti-Stokes light to the Stokes light. Specific data collected includes: temperature distribution along the optical cable (physical quantity: degrees Celsius, data format: a one-dimensional temperature array with a length of L / ΔR for each sample); local hotspot detection results (detection results of temperature anomaly segments exceeding preset thresholds and their peak values); and temperature change rate (temperature difference between adjacent sampling cycles, used for detecting rapid heating / cooling events).

[0025] Infrastructure Environmental Sensing Module (IEM): Installed in the fiber optic distribution robot cabinet and along the fiber optic cable splice boxes, it includes temperature and humidity sensors, air pressure sensors, and smoke detectors. Specific data collected includes: internal temperature and humidity, air pressure, and smoke / fire alarm status.

[0026] Furthermore, to achieve unified transmission and upper-layer system parsing of sensor data, this application defines the Multi-Sensor Data Packet (MSDP) protocol. Each MSDP packet is encapsulated in JSON (JavaScript Object Notation) format and contains the following field hierarchy.

[0027] MSDP packet structure diagram (JSON format): { "packet_id": "UUID-v4 unique identifier", "device_id": "Enhanced-AFS device ID", "cable_id": "Cable segment number", "timestamp_utc": "UTC timestamp of the time of collection", "sync_ref": "PTP network synchronization reference clock identifier", "sensors": { "dvm": { "enabled": true, "sample_rate_hz": 1000, "spatial_resolution_m": 2, "fiber_length_m": 40000, "intensity_array":[0.12,0.13,...], / / The number of samples is the fiber length / spatial resolution, i.e., fiber_length_m / spatial_resolution_m "events": [ { "event_id": "EVT-20260326-001", "location_m": 12500, "peak_intensity_db": 3.8, "duration_ms": 1200, "freq_spectrum": [[0,0.02],[10,0.15],...], / / [Hz, amplitude] "waveform_raw": "base64 encoded raw waveform", "event_type": "vibration|impact|unknown" } ] }, "dtm": { "enabled": true, "sample_interval_s": 10, "spatial_resolution_m": 1, "temp_array": [22.5, 22.6, ...], / / Unit: °C "hotspots": [ {"location_m": 8200, "temp_c": 65.3, "delta_rate": 2.1} ] }, "iem": { "cabinet_temp_c": 32.1, "cabinet_humidity_pct": 68.0, "pressure_hpa": 1013.2, "accel_xyz_g": [0.01, -0.02, 0.98], "smoke_alarm": false, "image_b64": null }, "opm": { "channels": [ {"fiber_id": "F01", "power_dbm": -18.3, "loss_db": 2.1} ] } }, "checksum": "SHA256 checksum" } MSDP data packets are reported to edge computing nodes in real time via the MQTT (Message Queuing Telemetry Transport) protocol (QoS=1), and the edge computing nodes verify the data before forwarding it to the data fusion layer. For large amounts of information such as the original waveforms of vibration events, an event-triggered reporting mechanism is adopted, uploading the complete waveform only when a valid event is detected, and uploading only statistical feature summaries on a daily basis to reduce network bandwidth consumption.

[0028] In step S2, the data fusion layer adopts a multi-source heterogeneous data fusion engine, including the optical network operation and maintenance data model ONOM (Optical Network Operations Model), unified data storage (time series database + graph database + relational database), and optical network operation and maintenance knowledge graph. It is responsible for unified modeling, fusion storage, and correlation analysis of the multi-dimensional sensor data reported by the physical sensing layer with network resource data, GIS geographic information, business data, etc.

[0029] Specifically, this application proposes an ONOM (Optical Network Operation and Maintenance) data model for optical cable operation and maintenance scenarios. ONOM performs unified semantic modeling of physical entities, logical resources, sensor data, and operation and maintenance events in the optical cable network. It maps multi-parameter sensor data, GIS topology data, historical work order data, equipment ledgers, and other multi-source data to ONOM, storing them in structured JSON format in both the InfluxDB time-series database and the Neo4j graph database, thus enabling the organization and management of optical cable operation and maintenance data. The ONOM model comprises a physical ontology layer, a sensor data layer, and an operation and maintenance event layer.

[0030] The physical ontology layer defines the core entity classes, properties, and relationships of the optical fiber network physical layer. Major entity classes include: Cable Segment (properties include number, start and end points, length, material, laying method, and GIS coordinate sequence), Fiber Core (properties include core sequence, type, and attenuation coefficient), Optical Cable Splice Box (OCC), Optical Distribution Frame (ODF), Alternating Field Distribution System (AFS), and Repeater Section. Relationships between entities include semantic predicates such as containOf, connectsTo, and monitoredBy, forming a complete physical topology ontology graph.

[0031] The sensor data layer maps various sensor measurements from the MSDP data packet to temporal attributes associated with physical entities in the ONOM model. These include: vibration events sensed by the DVM are instantiated as VibrationEvent entities, associated with the corresponding CableSegment entity via the occurAt relationship, and accompanied by attributes such as timestamp, intensity, and spectrum. DTM temperature arrays are instantiated as TemperatureDistribution entities and attached to the corresponding optical cable segment via the associatedWith relationship. OPM data are instantiated as OpticalPowerMeasurement entities and associated with the FiberCore entity via the associatedWith relationship. All types of sensor entities inherit from the abstract base class SensorObservation.

[0032] The operations and maintenance event layer uses ontology semantics to describe operations and maintenance events such as alarms, faults, and maintenance work orders, and establishes associations with physical entities and sensor data. The alarm entity AlarmEvent points to the SensorObservation entity that triggered the alarm through the triggeredBy relationship, points to the affected business entity ServicePath through the affectedSessionsService relationship, and is associated with the corresponding CableSegment through the locatedAt relationship, forming a complete semantic link of "sensor data alarm event physical location affected business".

[0033] ONOM subscribes to the MQTT message queue, receives MSDP data packets reported by Enhanced-AFS devices, and performs JSON parsing, field validation, and data quality assessment (including missing value detection, outlier removal, and clock drift correction). Then, it instantiates the parsed MSDP data according to the ONOM model, generating various sensor observation entities (VibrationEvent, TemperatureDistribution, OpticalPowerMeasurement, etc.). Using the mapping relationship between device IDs and fiber optic cable IDs, it automatically mounts the newly generated instances to the corresponding physical entity nodes and writes them to the Neo4j graph database and the InfluxDB time-series database.

[0034] Among them, the sampling rates of vibration, temperature, and optical power sensor data differ. The fusion engine adopts a "hierarchical sliding window aggregation" strategy: using a preset time interval as the base time window, it calculates the maximum vibration intensity, event frequency, and dominant frequency distribution within the window for vibration data; calculates the mean and standard deviation for temperature data; and calculates the mean and variation for optical power data. This generates a multi-dimensional feature vector with a unified time resolution, which is stored in a PostgreSQL relational database as input for alarm correlation analysis and prediction models. Furthermore, it calls the OWL (Web Ontology Language) inference rules pre-configured based on the Apache Jena framework to perform semantic inference on the instantiated ONOM graph, completing the following correlation derivation: when a certain optical cable segment simultaneously has a vibration event entity (VibrationEvent) with an intensity exceeding a set value and an optical power attenuation exceeding a set value within the same time window, it automatically determines that there is a composite event with external force damage risk; when the temperature of a certain optical cable segment shows a continuous upward trend, and the daily rate of increase reaches a set threshold, it automatically infers that there is an aging risk warning. The inference results are written back to the graph as upper-layer maintenance event entities, triggering the multi-modal alarm correlation analysis process.

[0035] Furthermore, a knowledge graph for optical cable operation and maintenance is constructed, encompassing multi-dimensional entities and relationships such as optical cable routing, fiber core status, equipment nodes, alarm history, maintenance records, and environmental awareness, with Neo4j graph database as the underlying storage. Entities in the knowledge graph include optical cable segments, repeater nodes, equipment rooms, fiber optic distribution robots, service circuits, and maintenance personnel; relationships include physical connections, bearer relationships, alarm associations, and historical repairs. The knowledge graph is updated in real time and, combined with a large model, enhances RAG (Retrieval-Augmented Generation) retrieval generation, improving the accuracy of the large model's professional knowledge in operation and maintenance scenarios. The knowledge graph construction consists of two processes: initial construction and incremental updates.

[0036] (1) Initialize data acquisition and cleaning, and convert structured data to a map. Obtain full resource information (optical cable segment list, fiber core usage status, service routes), optical cable route geographic coordinate data, equipment list, topology connection relationships, and operation and maintenance event records from the optical cable resource management system, GIS system, and network management system through data interfaces or data import methods. Data cleaning includes deduplication, format standardization, completion of missing fields using topology inference, and filtering of erroneous data by verifying topology connectivity through a rule engine. Based on the Python py2neo library's ETL (Extract-Transform-Load) module, the cleaned structured data is converted into knowledge graph triples (subject-predicate-object) according to the ONOM ontology model and then imported into the Neo4j database. For example, a record of an optical cable segment obtained from the optical cable resource management system is converted into the triple: (CableSegment:CABLE-G318-K325)-[:connectsTo]->(OCC:OCC-K325), which means that the optical cable segment numbered CABLE-G318-K325 is connected to the optical distribution box equipment numbered OCC-K325.

[0037] (2) Continuous incremental updates to the knowledge graph. The knowledge graph is continuously expanded during system operation: real-time sensor data is automatically written into the graph after being fused by ONOM; after each operation and maintenance operation is completed, the structured information such as the location of the fault, the handling plan, and the recovery time is automatically written into the graph to form fault case knowledge nodes; the graph quality inspection is performed regularly, and the PageRank graph algorithm is used to identify data quality problems such as "isolated nodes" and "dangling edges" and automatically repair them.

[0038] In step S3, the intelligent decision-making layer adopts a multi-agent collaborative engine, which includes four types of professional intelligent agents: intent understanding agent, knowledge retrieval agent, task planning agent, and execution monitoring agent. They collaborate in a Supervisor-Worker mode, and drive Enhanced-AFS to complete operations such as fiber optic physical patching through interface calls to physical execution tools, thereby realizing the autonomous planning and execution of operation and maintenance tasks.

[0039] Specifically, the intent-understanding agent is responsible for classifying the natural language commands input by maintenance personnel into four main categories: fault emergency dispatch, resource query, predictive maintenance, and configuration management. Entity extraction includes fiber optic cable segment names, fault locations, service names, and time ranges. Contextual semantic understanding is also achieved by parsing linguistic phenomena such as pronouns and ellipses from historical dialogues. The intent classification employs a hybrid scheme of Few-shot Prompt and fine-tuning classification to improve accuracy in intent recognition tasks related to fiber optic cable maintenance.

[0040] The knowledge retrieval agent is responsible for retrieving contextual information relevant to the current task from the optical cable maintenance knowledge graph and real-time sensor database. A hybrid retrieval strategy is employed. For structured queries such as "query the current optical power of segment K325," an InfluxQL query statement is directly generated and executed. For unstructured knowledge retrieval such as "retrieve which external force vibration events have occurred in the history of this segment," vector similarity retrieval is used. The knowledge graph text description is vectorized based on the BGE-M3 embedding model and stored in the Qdrant vector database. The Top-K retrieval results are concatenated into RAG context and provided to subsequent agents.

[0041] The task planning agent generates a step-by-step task execution plan based on intent and knowledge context using thought chain reasoning. Each task step corresponds to a specific tool call (such as: calling OTDR test call_otdr_scan, querying device list query_device_list, calculating backup route calculate_bypass_route, generating Enhanced-AFS fiber jump instruction generate_afs_instruction, etc.), and defines the dependencies between steps and the conditions for parallel execution to generate the task plan.

[0042] The execution monitoring agent schedules and executes threads according to the task plan, collects execution results in real time, detects timeouts, execution failures, or structural anomalies, and triggers adaptive adjustments when anomalies occur. If the Enhanced-AFS physical operation fails, the task planning agent is called back, and alternative routes are replanned based on the current state without manual intervention. Simultaneously, the current execution progress is output in natural language or text format to maintain visualization of the execution process.

[0043] The four agents communicate via a message bus, with each agent independently maintaining its task state and context memory. To address the complexity of the maintenance tasks, a Supervisor-Worker collaborative model is adopted. The task planning agent acts as the supervisor, coordinating the execution agents. The execution agents provide feedback on execution results and anomalies, and the supervisor dynamically adjusts the task plan based on this feedback, forming a closed-loop process of "planning-execution-feedback-adjustment".

[0044] Furthermore, spatiotemporal correlation analysis is performed on various alarms, including abnormal vibration, abnormal temperature, decreased optical power, and service interruption. Within a configurable default time window, if vibration alarms, temperature alarms, and optical power alarms on the same optical cable segment are triggered simultaneously, they are aggregated into a high-confidence "optical cable physical fault" composite alarm; if only a vibration alarm is triggered while the optical path is normal, it is identified as an "external interference warning." The alarm correlation rules are executed collaboratively by the rule engine and the large model and are dynamically updated.

[0045] Among them, the multimodal alarm correlation analysis takes the three-dimensional aggregation of "time-space-semantics" as the core, and the specific process is as follows.

[0046] (1) Normalization of raw alarms: Five types of raw alarms from DVM vibration sensing, DTM temperature sensing, OPM optical power monitoring, OTDR measurement, and network management service layer are uniformly converted into standard alarm objects, which include the fields: {alarm_id, source_type, cable_id, location_m, timestamp_utc, severity[], raw_params, status}. The fields are defined as follows: alarm_id is a unique alarm identifier, generated by combining a UTC timestamp and device number, and is globally unique; source_type is an enumeration of alarm source types, including five categories: dvm (vibration sensing), dtm (temperature sensing), opm (optical power monitoring), otdr (optical time domain reflectance measurement), and nms (network management service layer); cable_id is the optical cable segment number to which the alarm belongs, corresponding one-to-one with the CableSegment entity in the ONOM model; location_m is the physical location of the alarm on the optical cable, in meters, calculated cumulatively along the route from the starting splice point of the optical cable; timestamp_utc is the UTC timestamp of the alarm occurrence time, accurate to milliseconds, and multi-source sensor data is processed via PTP (Precision Time Protocol). After clock synchronization, the Protocol is aligned to the same time base; severity is an array of alarm severity levels, which is automatically determined by the normalization module based on the original alarm threshold and context information, with values ​​of critical, major, minor, and warning; raw_params is a snapshot of the parameters of the original alarm, which retains the original data fields of the source system in the form of a JSON object without losing any information, for subsequent in-depth analysis; status is the current processing status of the alarm, with values ​​including open (alarm in progress), acknowledged (acknowledged), and resolved (resolved).

[0047] (2) Spatiotemporal aggregation and compression: Multiple alarms from the same source generated within the same optical cable segment and the same time window are aggregated and compressed to avoid alarm storms. After aggregation, alarm clusters are output, recording the first occurrence time, duration, geographical range, and alarm list of each source.

[0048] (3) Association Rule Matching (Rule Engine): A pre-built alarm association rule library is used. The rules are described in Drools rule language and stored in the Rule Repository. Forward chain reasoning is performed through the Drools rule engine. Example rule: IF Vibration alarm (intensity > 2dB, duration > 500ms) AND Optical power alarm (loss > 0.3dB) AND Position deviation < 100m AND Time deviation < 5min THEN Association is a composite alarm of "external disturbance risk" (confidence = 0.9).

[0049] (4) LLM Semantic Association Supplement: For alarm combinations that the rule engine fails to hit, the alarm cluster information is formatted into a natural language description and input into the large language model LLM along with historical fault cases in the same location retrieved from the knowledge graph. The LLM performs semantic reasoning to determine the possible causal relationship between alarms and generates LOW / MEDIUM / HIGH confidence estimates and reasoning basis explanations as a semantic supplement to the rule engine.

[0050] (5) Combined output of composite alarm: Based on the analysis results of the rule engine and LLM, the final composite alarm is output, including: {alarm_id,compound_type, involved_alarms[],root_cause_hypothesis,affected_services[],recommended_action,confidence,reasoning_trace}.

[0051] The fields are defined as follows: alarm_id is a unique identifier for the composite alarm, generated globally by the system; compound_type is the composite alarm type, with values ​​including preset types such as external_force_damage (risk of external force damage), cable_aging (risk of optical cable aging), environmental_anomaly (environmental anomaly warning), and multi_source_fault (multi-source composite fault). When neither the rule engine nor LLM can classify it into an existing type, it is marked as unknown and awaits manual classification by operations and maintenance experts; involved_alarms is an array of original standard alarm IDs involved in this association, recording which original alarms the composite alarm was aggregated from, supporting tracing back to the original sensor data; root_cause_hypothesis is the root cause hypothesis, a conclusion inferred from the fault root cause by rule engine matching or LLM semantic reasoning, described in natural language, such as "the sealing failure of the junction box at K327+350m caused water ingress, leading to local temperature rise and..." "Increased fiber optic attenuation"; affected_services is an array of affected service circuit IDs, automatically derived from the mapping relationship between CableSegment and ServicePath entities in the knowledge graph; recommended_action is the system's suggested handling measures, automatically generated by LLM in combination with historical records of similar fault handling and maintenance procedures in the knowledge graph, such as "immediately arrange on-site inspection and replace the junction box sealing strip, and prepare a backup route switching plan"; confidence is the overall confidence score of the composite alarm, ranging from 0.0 to 1.0, calculated by weighted fusion of the confidence scores of the rules hit by the rule engine and the confidence scores output by the LLM. The part hit by the rule engine takes the pre-set confidence score, and the part output by the LLM takes the confidence score it is labeled; reasoning_trace is the reasoning link tracing record, which completely records the generation process of this composite alarm in the form of an ordered step array, including the reasoning source (rule number or LLM call), matching conditions, intermediate conclusions and final conclusions of each step, for post-event auditing and rule optimization.

[0052] In addition, the dynamic update of alarm association rules is achieved through the "rule engine and big model collaboration" mechanism, which is divided into the following three update paths. (1) Manual update: Operation and maintenance personnel submit new rules or modify existing rules to the system through the rule management interface that supports natural language input rule descriptions. The system translates the natural language rule descriptions into the standard Drools DRL format through LLM. After syntax verification and logical conflict detection, it takes effect in real time without system restart through the rule engine hot deployment mechanism. (2) Automatic summary of post-event review: After each fault handling is completed, the system automatically executes the post-event review process, inputs the alarm sequence (alarm timeline, location, intensity, etc.) and the finally confirmed root cause and handling plan into LLM. LLM analyzes the correlation between alarm mode and fault root cause. Through similarity retrieval with the existing rule library vector, if the similarity is less than the preset judgment parameter, it is judged as a new candidate rule. If a new association mode that does not exist in the current rule library is identified, a rule candidate plan is automatically generated and pushed to the operation and maintenance experts for review. After confirmation, it is included in the rule library. (3) Batch statistical mining: Regularly perform offline rule mining tasks: Extract alarm data from the historical alarm database for the past 3 months, use the Apriori association rule mining algorithm to analyze the frequent co-occurrence patterns of alarm events, generate candidate association rules, filter out rules without physical meaning after LLM semantic verification, and then update the rule library in batches after expert review.

[0053] In addition to alarm correlation analysis, this application further establishes a preventive maintenance and trend prediction mechanism, conducting in-depth analysis of the optical cable's operating status based on multi-dimensional time-series data to achieve proactive fault prevention. The specific processing flow is as follows: (1) Preparation of multidimensional time series data. Taking the optical cable segment as the analysis granularity, multidimensional sensing time series data within the analysis period of the optical cable segment are extracted from the InfluxDB time series library, including DVM vibration intensity statistics (daily maximum vibration intensity, number of vibration events, average main frequency), DTM temperature statistics (daily average temperature, maximum temperature, temperature standard deviation), OPM optical power statistics (daily minimum optical power, optical power change ΔP), and OTDR loss statistics (weekly OTDR measurement results, attenuation coefficient trend).

[0054] (2) Feature extraction and anomaly detection. For each dimension of time series data, the following were performed: single-dimensional anomaly detection based on Isolation Forest, anomaly sampling points were marked and interpolated to complete; trend-seasonal-residual decomposition based on STL (Seasonal-Trend decomposition using Loess) was used to remove interference such as seasonal and periodic effects of temperature and extract medium- and long-term trend components; K-means clustering was performed on the time-frequency features of vibration events to classify vibration events into categories such as "construction vibration", "traffic vibration", "natural disturbance" and "abnormal impact", and the frequency of each category was counted as feature input.

[0055] (3) Trend prediction modeling. The Prophet time series prediction model is used to make rolling predictions of the trend components of each dimension of data over a set time period and output the prediction confidence interval. The Prophet model is trained and stored independently for each optical cable segment, supports incremental updates when new data is generated, and automatically executes batch update tasks at set times. When the predicted value exceeds the preset health threshold within the confidence interval, a "health deterioration warning" is triggered and a preventive maintenance work order suggestion is generated in advance.

[0056] (4) Multidimensional health score. The overall health score (Cable Health Score, CHS, 0~100) of the optical cable segment is calculated by combining the current status and trends of the four sensing dimensions. The scoring model adopts a combination of weighted linear model and semantic correction of large model.

[0057]

[0058] The scores for each sub-item are determined by the degree of deviation of the current state of the corresponding sensor data from the healthy baseline and the predicted trend of future deterioration risk. The weights W1 to W4 sum to 1, and can be configured according to different optical cable scenarios such as overhead / pipeline / submarine cables. The weighted scores are semantically corrected by LLM. The intermediate scores and anomaly descriptions of each dimension are input into the large model. LLM combines the historical fault records of the optical cable segment in the knowledge graph and the horizontal comparison of similar optical cable segments to semantically correct the scores and output a natural language health report.

[0059] (5) Preventive maintenance work order generation. When the CHS score of the optical cable segment is below 60, the system automatically triggers the preventive maintenance work order generation process. The LLM generates maintenance recommendations based on the health score report, main cause analysis, and historical records of similar treatments (e.g., "It is recommended to conduct on-site inspection of the CABLE-G318-K325 segment within 15 days, focusing on checking the sealing of the splice box at K327+350m. In the past 30 days, the DTM at this location has continuously shown local hot spots 8°C higher than the average ambient temperature. Based on historical records, it is judged that there is a risk of water ingress into the splice box leading to increased thermal resistance"). The work order is pushed to the local maintenance team through the operation and maintenance management platform. The team's confirmation and response to the alarm are written back to the knowledge graph in real time for subsequent model evaluation and calibration.

[0060] As can be seen, the natural language operation and maintenance method for optical fiber networks provided in this application relies on the collaborative operation of the physical sensing layer, data fusion layer, intelligent decision-making layer, and application interaction layer. It achieves comprehensive collection of optical fiber operating status through multiple types of sensing modules, completes multi-source data fusion analysis by combining data modeling and knowledge graphs, and automatically executes operation and maintenance operations by leveraging multi-agent collaborative parsing of natural language commands. Simultaneously, it links with a rule engine and large-scale models to achieve intelligent alarm analysis, trend prediction, and health assessment, and can automatically generate preventative maintenance work orders. This reduces the workload of manual operation and maintenance, improves the automation and intelligence level of optical fiber network operation and maintenance and the efficiency of fault handling, enables early prediction of operational risks, and ensures the stable and reliable operation of the optical fiber network.

[0061] Based on the same concept of the present invention, as shown in the appendix to the specification. Figure 3 As shown in the figure, an embodiment of this application provides the structure of an electronic device 300, which includes: at least one processor 301, at least one network interface 304 or other user interface 303, memory 305, and at least one communication bus 302. The communication bus 302 is used to realize the connection and communication between these components. The electronic device 300 may optionally include a user interface 303, including a display (e.g., touch screen, LCD, CRT, holographic imaging, or projector, etc.), a keyboard, or a clicking device (e.g., mouse, trackball, touchpad, or touch screen, etc.).

[0062] Memory 305 may include read-only memory and random access memory, and provides instructions and data to processor 301. A portion of memory 305 may also include non-volatile random access memory (NVRAM).

[0063] In some implementations, memory 305 stores executable modules or data structures, or subsets thereof, or extended sets thereof: The 3051 operating system contains various system programs used to implement various basic business functions and handle hardware-based tasks. Application module 3052 contains various applications, such as desktop (launcher), media player (MediaPlayer), browser (Browser), etc., to implement various application services.

[0064] In this embodiment of the application, the processor 301 executes steps such as a natural language operation and maintenance method for an optical fiber network by calling programs or instructions stored in the memory 305.

[0065] This application also provides a computer-readable storage medium storing a computer program that, when executed by a processor, performs steps such as those in a natural language operation and maintenance method for an optical fiber network.

[0066] Specifically, the storage medium can be a general-purpose storage medium, such as a portable disk or hard drive. When the computer program on the storage medium is run, it can lower the threshold for operation and maintenance and improve the efficiency of fault early warning and emergency recovery.

[0067] In the embodiments provided in this application, it should be understood that the disclosed apparatus and methods can be implemented in other ways. The apparatus embodiments described above are merely illustrative. For example, the division of units is only a logical functional division, and there may be other division methods in actual implementation. Furthermore, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Additionally, the mutual coupling or direct coupling or communication connection shown or discussed may be through some communication interface, and the indirect coupling or communication connection of the apparatus or units may be electrical, mechanical, or other forms.

[0068] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0069] In addition, the functional units in the embodiments provided in this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit.

[0070] If a function is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0071] Finally, it should be noted that the above embodiments are merely specific implementations of this application, used to illustrate the technical solutions of this application, and not to limit them. The protection scope of this application is not limited thereto. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that any person skilled in the art can still modify or easily conceive of changes to the technical solutions described in the foregoing embodiments, or make equivalent substitutions for some of the technical features, within the scope of the technology disclosed in this application; and these modifications, changes, or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application. All should be covered within the protection scope of this application. Therefore, the protection scope of this application should be determined by the protection scope of the claims.

Claims

1. A natural language operation and maintenance method for optical fiber networks, characterized in that, A natural language operation and maintenance system for optical fiber networks is applied, comprising a physical perception layer, a data fusion layer, an intelligent decision-making layer, and an application interaction layer; the method includes the following steps: The enhanced fiber optic distribution robot at the physical sensing layer collects multi-dimensional data including fiber optic cable vibration, temperature, environment, optical power, and OTDR link attenuation, and encapsulates the collected multi-dimensional data into standardized MSDP data packets and reports them to the data fusion layer. By using the data fusion layer, MSDP data packets, network resource data, GIS geographic information, and business data are subjected to unified semantic modeling, fusion storage, and semantic reasoning to construct a knowledge graph for optical cable operation and maintenance. The application interaction layer receives natural language commands from maintenance personnel. The intelligent decision layer, based on real-time collected multi-dimensional data and the constructed optical cable maintenance knowledge graph, completes intent understanding, knowledge retrieval, task planning and execution scheduling through multi-agent collaboration. At the same time, it performs alarm correlation analysis and health scoring to generate preventive maintenance work orders.

2. The optical fiber network natural language operation and maintenance method according to claim 1, characterized in that, The enhanced fiber optic distribution robot includes an intelligent fiber optic distribution robot, a distributed vibration sensing module (DVM), a distributed temperature sensing module (DTM), and an infrastructure environment perception module (IEM). Among them, the intelligent fiber optic distribution robot collects optical power and OTDR link attenuation data; the DVM collects optical cable vibration data based on the phase-sensitive optical time-domain reflectometry principle; the DTM collects optical cable temperature distribution, local hot spots and temperature change rate data based on the Raman scattering principle; the IEM collects cabinet temperature and humidity, air pressure and smoke alarm status data; and all collected data are encapsulated into JSON format data packets according to the MSDP protocol, and the data packets are reported to the edge computing node via the MQTT protocol, and forwarded to the data fusion layer after verification.

3. The optical fiber network natural language operation and maintenance method according to claim 2, characterized in that, The construction of the optical cable operation and maintenance knowledge graph includes the following steps: The three-layer architecture of the optical cable network operation and maintenance data model ONOM is used for semantic modeling, defining network physical entities, sensor time-series entities, operation and maintenance event entities and their interrelationships. The system receives and parses MSDP data packets, instantiates entities according to the ONOM model, and stores them in the time-series database and graph database. Then, it performs time-series alignment of multi-source sensor data through hierarchical sliding window aggregation, generates multi-dimensional feature vectors, and stores them in a relational database. Finally, it performs semantic analysis based on OWL inference rules to identify optical cable fault risks and writes the inference results back to the knowledge graph. Perform the initial construction of the optical cable operation and maintenance knowledge graph, and update the data synchronously during system operation.

4. The optical fiber network natural language operation and maintenance method according to claim 3, characterized in that, The intelligent decision-making layer includes an intent-understanding agent, a knowledge-retrieval agent, a task-planning agent, and an execution-monitoring agent, which interact through a message bus and work collaboratively in a Supervisor-Worker mode. Specifically, the intent-based intelligent agent classifies natural language instructions, extracts key entities, and performs semantic parsing based on the dialogue context; the knowledge retrieval intelligent agent uses a combination of structured query and vector similarity retrieval to retrieve relevant information from knowledge graphs and sensor databases to generate RAG context; the task planning intelligent agent combines the parsing results and retrieval information to generate a task plan containing tool calls and step dependencies through thought chain reasoning; and the execution monitoring intelligent agent schedules the execution according to the task plan and detects execution anomalies in real time and adjusts adaptively.

5. The optical fiber network natural language operation and maintenance method according to claim 4, characterized in that, The alarm correlation analysis is performed using the following methods, including the following steps: Multiple types of alarms, including vibration, temperature, optical power, OTDR, and network management services, will be uniformly converted into standard alarm objects. Based on time and space dimensions, alarms from the same source are aggregated and compressed to generate alarm clusters, and the Drools rule engine calls the pre-built rule base for correlation reasoning; among them, alarm clusters without matching rules are combined with historical fault cases and semantic reasoning is supplemented by large language model LLM. By integrating the analysis results of the Drools rule engine and the Large Language Model (LLM), a composite alarm is output, which includes type, root cause, impact on business, handling suggestions, confidence level, and inference chain.

6. The optical fiber network natural language operation and maintenance method according to claim 5, characterized in that, The alarm correlation analysis also includes the following steps: The rule base is dynamically updated by combining three approaches: manual data entry and translation, fault review and summarization, and offline data mining, along with a large language model (LLM).

7. The optical fiber network natural language operation and maintenance method according to claim 6, characterized in that, The health score is determined using the following method, including the following steps: Taking optical cable segments as units, multi-dimensional time-series statistical data corresponding to vibration, temperature, optical power, and OTDR are extracted from the time-series database; The collected multidimensional time-series data were preprocessed and feature mined. Among them, the isolated forest algorithm was used to identify and complete abnormal data, the trend components were obtained through time-series decomposition, and the vibration features were clustered and classified using the K-means algorithm. The time-series prediction model is used to perform rolling predictions on the trend components of data in each dimension; among them, the optical cable operation status is determined according to the preset health threshold and risk warnings are triggered. A weighted linear model with configurable weights is used to calculate the overall health score of the optical cable; when the overall health score of the optical cable is lower than a set threshold, a preventive maintenance work order is generated.

8. A natural language operation and maintenance system for optical fiber networks, characterized in that, It includes a physical perception layer, a data fusion layer, an intelligent decision-making layer, and an application interaction layer, which are used to collaboratively execute the steps of a natural language operation and maintenance method for optical fiber networks as described in any one of claims 1 to 7.

9. An electronic device, characterized in that, include: The device includes a processor, a memory, and a bus. The memory stores machine-readable instructions executable by the processor. When the electronic device is running, the processor communicates with the memory via the bus. When the machine-readable instructions are executed by the processor, they perform the steps of a natural language operation and maintenance method for an optical fiber network as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, performs the steps of a natural language operation and maintenance method for an optical fiber network as described in any one of claims 1 to 7.

Images