Optical fiber network intelligent inspection and fault closed-loop processing system, method and device

Abstract

The application discloses an optical fiber network intelligent inspection and fault closed-loop processing system, method and equipment, and belongs to the technical field of optical fiber communication network intelligent operation and maintenance. The method comprises the following steps: collecting and fusing multi-source heterogeneous data such as resource management, fault network management, distributed sensing and geographic information; analyzing the fused data by using a multi-modal model based on a convolutional neural network and a graph neural network, realizing millimeter-level accurate positioning of faults, and dynamically optimizing an inspection plan by using a long short-term memory network model; based on the analysis and decision results, automatically creating, circulating and associating fault orders, mode orders and repair orders, and realizing a full-process closed loop from diagnosis order dispatching to repair verification. Through the above scheme, the application solves the problems of scattered data, single analysis dimension and lack of closed loop in the prior art, and significantly improves the accuracy, efficiency and automation level of optical fiber network operation and maintenance.

Description

Technical Field

[0001] This invention belongs to the field of intelligent operation and maintenance technology of optical fiber communication networks. Specifically, it relates to intelligent scheduled inspection and fault closed-loop processing system, method and equipment for optical fiber networks. Background Technology

[0002] With the profound transformation of power systems towards digitalization and intelligence, power communication networks, as the "neural network" supporting the safe and stable operation of the power grid, have become core elements in ensuring energy supply security due to their reliability and operational efficiency. Among these, fiber optic communication networks, with their inherent advantages such as high bandwidth, strong resistance to electromagnetic interference, and long transmission distance, form the physical foundation for carrying critical services such as relay protection, dispatch automation, and intelligent inspection. Therefore, ensuring the health of fiber optic communication networks and achieving rapid response and accurate location of fiber optic line faults is a major challenge facing modern power operation and maintenance systems.

[0003] Currently, for the operation and maintenance of power communication fiber optic networks, the main technical solutions in this field are periodic testing and single-system fault analysis. On the one hand, maintenance personnel use specialized instruments such as optical time domain reflectometers (OTDRs) to conduct regular offline tests on fiber optic cable lines. This method, by analyzing the reflection and attenuation events of the OTDR curves, can, to a certain extent, determine the continuity and loss status of the fiber optic cable, playing a fundamental role in ensuring the physical connectivity of the basic communication links of the fiber optic network. On the other hand, to improve the automation level of fault diagnosis, attempts are being made to analyze OTDR data using algorithms such as neural networks to identify fault types. These technical means together constitute the cornerstone of the existing fiber optic network operation and maintenance model, providing basic tools for daily maintenance and fault diagnosis.

[0004] However, with the increasing demands for communication reliability and real-time performance in power grid operations, and the growing scale and complexity of networks, the inherent limitations of the aforementioned operation and maintenance methods that rely on fixed cycles and single data sources are becoming increasingly apparent. First, existing technologies generally suffer from the problem of "data silos." Resource management, fault management, and scheduled maintenance systems are often independent business systems with inconsistent data standards and storage architectures (such as SQL and NoSQL databases), leading to a failure to effectively integrate critical information. This fragmentation directly results in a single dimension of fault analysis; for example, analyzing only OTDR data while ignoring strongly correlated information such as network management alarms and resource topology leads to poor fault location accuracy, with an average location deviation typically exceeding 200 meters. Second, existing technologies lack sufficient intelligence and automated closed-loop capabilities. Scheduled maintenance plans often rely on fixed time periods (e.g., 6-12 months), making it difficult to dynamically adjust inspection strategies based on external environmental threats such as typhoons and icing. In the fault handling process, there is a lack of effective automatic correlation and closed-loop verification mechanisms from diagnosis and dispatch to the verification of post-repair effectiveness. Data shows that over 78% of fault handling cases did not automatically verify the repair results. Finally, the existing solutions are poorly adapted to power business scenarios, fail to fully consider the stringent requirements of advanced services such as relay protection for communication recovery time (the service switching time in the existing solutions often exceeds 5 minutes), and lack a mechanism for collaborative risk assessment and protection of multiple optical cables laid in the same trench.

[0005] Therefore, how to overcome the bottleneck of scattered data and single analysis dimensions in the current operation and maintenance systems, and build a collaborative processing method that can integrate multi-source heterogeneous data, realize intelligent analysis and decision-making and automatic closed-loop work order process, so as to significantly improve the accuracy of fault location, dynamic adaptability of scheduled maintenance plan and overall efficiency of fault handling in power fiber optic network, and meet the high reliability communication guarantee requirements in complex power scenarios, has become a technical problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0006] The technical problem this invention aims to solve is to overcome the limitations of existing fiber optic network operation and maintenance systems, such as fragmented data, limited analytical dimensions, low intelligence, and lack of closed-loop processes. These limitations result in poor fiber optic fault location accuracy, rigid scheduled maintenance plans, and low fault handling efficiency, failing to meet the high reliability and real-time requirements of modern power systems for communication networks. Therefore, this invention provides a method, system, and device for intelligent scheduled maintenance and fault closed-loop processing of fiber optic networks. It aims to achieve accurate perception of the power fiber optic network status, rapid fault location and handling, and dynamic optimization and closed-loop management of operation and maintenance tasks by deeply integrating multi-source heterogeneous data, constructing intelligent analysis models, and automating business processes.

[0007] To achieve the above-mentioned objectives, the present invention provides a technical solution: an intelligent scheduled inspection and fault closed-loop processing system for optical fiber networks, comprising: a multi-source heterogeneous data acquisition and preprocessing unit, an intelligent analysis and decision-making unit, a collaborative work order and closed-loop processing unit, and a comprehensive management and presentation unit.

[0008] The multi-source heterogeneous data acquisition and preprocessing unit is designed to collect and standardize operation and maintenance data from multiple heterogeneous business systems related to the power communication network in real time or near real time, providing a unified, clean, and spatiotemporally aligned data foundation for upper-level intelligent analysis. The unit's inputs are connected to the Resource Management System (RMS), the TMS, the Distributed Fiber Optic Sensing System (DAS), and the Geographic Information System (GIS); its output generates a structured data stream with precise timestamps and geographic location labels, and transmits this data stream to the intelligent analysis and decision-making unit.

[0009] The intelligent analysis and decision-making unit is the core of this invention. Its function is to receive fused data from the multi-source heterogeneous data acquisition and preprocessing unit, and utilize a series of lightweight artificial intelligence models to perform in-depth analysis, prediction, and decision-making regarding the state of the fiber optic network. The unit's output includes a series of high-precision diagnostic results and optimization strategy suggestions, such as accurate fault location coordinates, dynamically adjusted scheduled maintenance cycle lists, and optimal backup optical path schemes. These results will serve as the basis for the execution of collaborative work orders and closed-loop processing units.

[0010] The collaborative work order and closed-loop processing unit functions by automatically creating, associating, and circulating three core business work orders—"method orders," "fault orders," and "repair orders"—based on the output of the intelligent analysis and decision-making unit. It also constructs a fully automated closed-loop process from fault diagnosis, dispatch, repair to effect verification. This unit interacts with the enterprise's existing work order system or execution team through standardized interfaces, ensuring effective implementation of decisions and full traceability of the processing.

[0011] The integrated control and presentation unit integrates and visualizes the analysis results from the intelligent analysis and decision-making unit, the process status of collaborative work orders and closed-loop processing units, and the underlying raw data. This provides operations and maintenance managers with a comprehensive, multi-dimensional monitoring dashboard and report generation tool. The report not only includes detailed fault handling processes but also network health trends and scheduled maintenance plan execution status, providing data support for management decisions.

[0012] Furthermore, the multi-source heterogeneous data acquisition and preprocessing unit specifically includes: a communication network management data acquisition module, a distributed sensor data interface module, a spatial geographic information acquisition module, and a data fusion and alignment engine.

[0013] The communication network management data acquisition module is used to acquire data from existing operation and maintenance systems via standard protocols or adapters. For example, it polls network devices via the SNMP protocol to obtain real-time alarms (Traps), extracts static resource data such as optical cables, fiber cores, routes, and service bearers from a resource management system (RMS) using an SQL database via a JDBC or ODBC connector, and retrieves historical fault records from a fault management system (TMS) using a NoSQL database via an API interface.

[0014] The distributed sensing data interface module is specifically designed for accessing distributed optical fiber sensing (DAS) systems. This module receives the raw vibration or temperature data stream output by the DAS demodulator and converts it into structured event records using preset event recognition algorithms (such as threshold detection and pattern matching). For example, "At K12+345 meters, an abnormal vibration lasting 3 seconds with frequency characteristics of excavator operation was detected at time T."

[0015] The spatial geographic information acquisition module is responsible for connecting to the enterprise's internal GIS platform. This module obtains the precise GIS coordinates and elevation of facilities such as poles, manholes, and base stations along the optical cable route through a spatial query interface, as well as environmental information layers along the route, such as geological disaster risk areas and areas with highly corrosive soil.

[0016] The data fusion and alignment engine is the foundation for collaborative data analysis. This engine preferentially employs high-performance in-memory computing frameworks such as Apache Arrow. It loads data from different sources into a unified columnar in-memory format, avoiding the costly overhead of cross-system data serialization. Using high-precision timestamps as the core alignment key, the engine utilizes a sliding time window algorithm to correlate and align OTDR test events, network management alarm events, DAS vibration events, and related resource and geographic information occurring near a specific moment in milliseconds, forming a multimodal, spatiotemporally synchronized data snapshot for use by the intelligent analysis unit.

[0017] Specifically, the intelligent analysis and decision-making unit is the core technology of this invention. It integrates a series of analysis models to collaboratively complete the intelligent operation and maintenance of the optical fiber network. This unit includes: a multimodal fusion fault location model, a dynamic scheduled maintenance decision-making model based on Long Short-Term Memory (LSTM) networks, and a service-driven automatic optical path design engine.

[0018] A multimodal fusion fault location model aims to improve fault location accuracy to the meter level. This model innovatively combines a one-dimensional convolutional neural network (1D-CNN) and a graph neural network (GNN). The 1D-CNN processes one-dimensional OTDR test curve data, automatically extracting waveform features of events such as reflection, attenuation, and breakage, and outputting a feature vector for the fault type. The GNN models the topology of the fiber optic network, treating towers, junction boxes, etc., as nodes in the graph, and cable segments as edges; network management alarms and DAS vibration events are attached as attributes to the corresponding nodes or edges. The GNN learns the impact range of the fault on the topology by propagating and aggregating information on the graph. Finally, a fusion layer (such as a fully connected network) concatenates and jointly infers the waveform features extracted by the 1D-CNN and the topology-event features extracted by the GNN, outputting the GPS coordinates of the fault point with the highest confidence. The model's input is a spatiotemporally aligned data snapshot generated by a data fusion engine, and the output is the fault location accurate to within ±30 meters.

[0019] A dynamic scheduled maintenance decision model based on Long Short-Term Memory (LSTM) networks is used to achieve intelligent scheduled maintenance from "periodic" to "optimal". This model treats the historical status data of each optical cable (such as historical failure rate, previous attenuation test values, and repair records) as a time series, using the LSTM network to learn the inherent patterns of aging and degradation, thereby predicting its future health trend. The model's decision-making process incorporates a weighted scoring formula: T_new = T_base × (1 - sigmoid(w1*α+ w2*β + w3*γ)), where T_base is the base period (e.g., 12 months), α is the future failure probability predicted by the LSTM, β is the service weight carried by the optical cable (e.g., the weight for carrying relay protection services is 0.9), γ is the environmental risk level obtained from GIS (e.g., the high-corrosion area level is 0.8), and w1, w2, and w3 are configurable weight coefficients. The model dynamically outputs a global list of scheduled inspection priorities, allowing resources to be prioritized for the highest-risk optical cables. For example, the scheduled inspection cycle for optical cables in highly corrosive areas that carry important services can be automatically shortened by more than 50%.

[0020] Optionally, the dynamic scheduled maintenance decision model can also employ reinforcement learning. The maintenance system is viewed as an agent, the network state as the state space, the execution or postponement of scheduled maintenance as the action space, and the minimization of total maintenance costs (including inspection costs and failure losses) as the reward function. Through continuous interaction with the environment, the agent can learn an optimal scheduled maintenance strategy to maximize long-term maintenance benefits.

[0021] To achieve the aforementioned objectives, another technical solution provided by this invention is: a method for intelligent scheduled inspection and fault closed-loop processing of optical fiber networks. This method closely corresponds to the aforementioned system and includes the following steps:

[0022] Step S1: Establish a collaborative data view. The system continuously collects data from RMS, TMS, DAS, GIS, and other systems through a multi-source heterogeneous data acquisition and preprocessing unit. The data fusion and alignment engine cleans and standardizes this heterogeneous data, and aligns it based on timestamps and spatial locations, forming a unified, multi-dimensional real-time status view of the fiber optic network in memory.

[0023] Step S2: Trigger Fault Diagnosis and Precise Location. When the system detects a fault-triggered event, such as an OTDR test reporting a breakpoint event or a TMS receiving a link interruption alarm, the collaborative processing flow is immediately initiated. The intelligent analysis and decision-making unit acquires a snapshot of the collaborative data view before and after the event (e.g., ±5 minutes) and inputs it into the multimodal fusion fault location model. This model comprehensively analyzes the OTDR waveform, associated alarm content, topology connections, and surrounding DAS vibration events to calculate and output high-precision fault geographic coordinates and possible fault causes.

[0024] Step S3: Automated generation and association of work orders. Based on the location results, the collaborative work order and closed-loop processing unit automatically perform the following operations: First, create a "fault order," whose key fields include the fault ID, GPS coordinates output by the location model, fault type, and estimated impact range. Then, the system automatically queries and associates all "mode orders" (i.e., business circuit information orders) affected by this fault using the optical path ID, and marks the fault order with risk according to the business level recorded in the mode order; for example, those affecting relay protection services are marked as "Level 1 Emergency."

[0025] Step S4: Intelligent Dispatch of Maintenance Tasks. Based on the "Fault Order" and its associated "Mode Order," the system automatically generates a "Maintenance Order." During this generation process, the optical path automatic design engine is activated. It queries the resource library to find available spare fiber cores or routes. The engine's optimization goal is not only to meet business SLA requirements but also to follow risk avoidance principles, such as prioritizing spare routes that are not in the same trench or on the same pole as the faulty optical cable. The maintenance order is automatically populated with recommended spare resource IDs, estimated repair time based on historical data of similar faults, and acceptance standards that comply with power safety regulations (e.g., splice loss ≤ 0.1dB).

[0026] Step S5: Perform Repair and Closed-Loop Verification. After completing the on-site repair operation according to the "Repair Order," the maintenance personnel update the work order status to "Repair Completed" on their mobile terminal. This operation triggers the system's closed-loop verification mechanism. The system automatically sends an instruction to the scheduled maintenance system to immediately perform an OTDR retest on the repaired optical path. The collaborative work order and closed-loop processing unit compares the retested OTDR curve with the historical baseline curve and the fault curve before repair. If the test results meet the acceptance criteria, the system automatically updates the fiber core status in the relevant "Mode Order," closes the "Fault Order" and "Repair Order," and generates a closed-loop archive report containing a correlation graph of the three orders. If acceptance fails, the escalation mechanism is automatically triggered, setting the repair order status to "Review Failed" and notifying the next-level expert team to intervene.

[0027] Optionally, in step S4, when multiple emergency faults occur concurrently, causing contention for backup resources, the system initiates a resource conflict detection mechanism. This mechanism preferably employs optimistic locking based on Compare-and-Swap (CAS) to ensure the atomicity of resource allocation. Simultaneously, resource requests are queued according to preset service priority rules (e.g., relay protection services > real-time production control services > management information services) to ensure that the recovery time of critical services is prioritized.

[0028] This invention also provides an intelligent scheduled inspection and fault closed-loop processing device for fiber optic networks. Physically, this device can be a server or server cluster deployed in a data center. The device includes a high-performance processor (such as a multi-core CPU or GPU), a large-capacity memory (RAM and SSD), and multiple high-speed network interfaces communicating with the processor. The memory stores a computer program, which, when executed by the processor, can completely implement all the steps of the aforementioned detection method, providing an integrated technical support platform for the intelligent operation and maintenance of power communication networks.

[0029] Compared with the prior art, the present invention has the following beneficial effects:

[0030] Revolutionary improvement in positioning accuracy and diagnostic dimensions: This invention integrates multi-source data such as OTDR, network management, DAS, and GIS, and uses a model combining CNN and GNN to improve fault analysis from single waveform interpretation to multimodal joint reasoning, thereby improving fault location accuracy from the hundred-meter level to the meter level, significantly shortening the fault finding time for repair personnel.

[0031] The shift from passive response to proactive prediction in operation and maintenance: By introducing a dynamic scheduled inspection model based on LSTM, this invention transforms the traditional fixed and periodic inspection mode into a data-driven and risk-oriented predictive maintenance mode, which can allocate operation and maintenance resources more scientifically and prevent and resolve major network risks in advance.

[0032] High degree of automation and closed-loop system in fault handling process: This invention constructs a collaborative work order model with "three orders" linkage, realizing full-process automation from fault discovery, diagnosis, work order dispatch, resource allocation to repair verification. In particular, the automated review and verification mechanism ensures the quality of repair work, solves the problem of verification deficiency that is common in existing processes, and improves operation and maintenance efficiency and reliability.

[0033] Deeply adapted to power business scenarios: When performing resource scheduling and path planning, the solution of this invention fully considers the SLA requirements of key services such as relay protection and physical risk factors such as co-channel laying. Its design closely fits the safe production needs of the power system, providing a more solid communication guarantee for the safe and stable operation of the power grid. Attached Figure Description

[0034] Figure 1 This is a schematic diagram of the application environment of the intelligent scheduled inspection and fault closed-loop processing system for optical fiber networks in one embodiment of the present invention.

[0035] Figure 2 yes Figure 1 The diagram shows the functional modules of the system.

[0036] Figure 3 This is a flowchart of a fiber optic network intelligent scheduled inspection and fault closed-loop processing method in one embodiment of the present invention.

[0037] Figure 4 This is a schematic diagram illustrating the working principle of a multimodal fusion fault location model in one embodiment of the present invention.

[0038] Figure 5 This is a schematic diagram of a closed-loop processing flow for the coordinated use of three forms: "fault order", "method order" and "maintenance order" in one embodiment of the present invention. Detailed Implementation

[0039] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of the embodiments of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention.

[0040] Please see Figure 1 and Figure 2This invention provides an intelligent scheduled inspection and fault closed-loop processing system for optical fiber networks. It aims to overcome the limitations of existing optical fiber network operation and maintenance systems, such as fragmented data, limited analytical dimensions, low intelligence levels, and lack of closed-loop processes. By deeply integrating multi-source heterogeneous data, constructing intelligent analysis models, and automating business processes, it achieves accurate perception of the power optical fiber network status, rapid fault location and handling, and dynamic optimization and closed-loop management of operation and maintenance tasks. Logically, the system includes a multi-source heterogeneous data acquisition and preprocessing unit, an intelligent analysis and decision-making unit, a collaborative work order and closed-loop processing unit, and a comprehensive control and presentation unit. These four units work collaboratively to form a complete intelligent operation and maintenance closed loop.

[0041] Specifically, the core function of the multi-source heterogeneous data acquisition and preprocessing unit is to build the data foundation for upper-level intelligent analysis. It does not simply aggregate data, but rather collects and standardizes operational data from multiple heterogeneous business systems related to the power communication network in real-time or near real-time, providing a unified, clean, and spatiotemporally aligned data foundation for subsequent analysis. The unit's inputs are connected to the Resource Management System (RMS), the TMS, the Distributed Fiber Optic Sensing System (DAS), and the Geographic Information System (GIS); its output is a structured data stream with precise timestamps and geographic location labels, which is then delivered to the intelligent analysis and decision-making unit.

[0042] The intelligent analysis and decision-making unit is the core of this invention. Its function is to receive fused data from the multi-source heterogeneous data acquisition and preprocessing unit, and utilize a series of lightweight artificial intelligence models to perform in-depth analysis, prediction, and decision-making on the state of the fiber optic network. It is responsible for transforming raw data into high-value insights and executable instructions. The output of this unit is no longer a single alarm message, but a series of high-precision diagnostic results and optimization strategy suggestions, such as accurate fault location coordinates, dynamically adjusted scheduled maintenance cycle lists, and optimal backup optical path schemes. This data comprehensively and meticulously characterizes the state of the fiber optic network in the current and future period, and serves as the basis for the execution of collaborative work orders and closed-loop processing units.

[0043] The collaborative work order and closed-loop processing unit functions to create an automated business execution and verification process. Based on the output of the intelligent analysis and decision-making unit, it automatically creates, associates, and circulates three core business work orders: "method orders," "fault orders," and "maintenance orders," constructing a fully automated closed loop from fault diagnosis, dispatch, repair to effect verification. Its internal integrated logic ensures the effective implementation of decisions and full traceability of the processing. This unit interacts with the enterprise's existing work order system or execution team through standardized interfaces, serving as a key hub for transforming intelligent analysis results into actual operational actions.

[0044] Finally, the integrated control and presentation unit systematically integrates and user-friendly visualizes the analysis results from the intelligent analysis and decision-making unit, the process status of collaborative work orders and closed-loop processing units, and the underlying raw data. This ultimately provides operations and maintenance managers with a comprehensive, multi-dimensional monitoring dashboard and report generation tool. The report not only includes detailed fault handling processes but also network health trends and scheduled maintenance plan execution status, clearly identifying potential risks and bottlenecks in the network, thus providing direct and powerful data support for higher-level management decisions.

[0045] Furthermore, to achieve the above functions, the internal structure and working mechanism of each unit were designed to be more specific and refined. Please refer to [link / reference]. Figure 2 This demonstrates the specific composition of a multi-source heterogeneous data acquisition and preprocessing unit. This unit integrates a communication network management data acquisition module, a distributed sensor data interface module, a spatial geographic information acquisition module, and a data fusion and alignment engine.

[0046] The communication network management data acquisition module is the entry point for the system to obtain basic network status data. Its core function is to acquire data from a wide variety of existing operation and maintenance systems through standard protocols or dedicated adapters. For example, this module periodically polls network devices via the SNMP protocol to obtain real-time performance metrics and alarm information (Traps); it extracts relatively static but crucial resource data such as optical cables, fiber cores, routing, and service bearers from Resource Management Systems (RMS) using SQL databases via JDBC or ODBC connectors; and it obtains massive amounts of historical fault records from Fault Management Systems (TMS) that may use NoSQL databases through API interfaces, providing data accumulation for subsequent fault mode learning and trend analysis.

[0047] The distributed sensing data interface module is specifically designed for accessing distributed optical fiber sensing (DAS) systems, a novel sensing method. This module does not simply receive data; instead, it performs preliminary intelligent processing on the raw sensing data. It receives high-frequency raw vibration or temperature data streams from the DAS demodulator in real time and, through internally pre-set event recognition algorithms (such as threshold detection, short-time Fourier transform, or simple pattern matching), extracts structured and meaningful event records from the massive, undifferentiated data stream. For example, a record formatted as "At K12+345 meters, at time T, an abnormal vibration lasting 3 seconds with frequency characteristics of excavator operation was detected," significantly reducing the computational burden on upper-level analysis.

[0048] The spatial geographic information acquisition module is responsible for imbuing network data with a geographic context. This module proactively connects to the enterprise's internal GIS platform through standardized spatial query interfaces (such as WFS / WMS). It not only acquires the precise GIS coordinates and elevation of network facilities along the fiber optic cable route, such as towers, manholes, and base stations, but also obtains environmental information layers along the route, such as historical geological disaster risk areas, areas with highly corrosive soil, and areas of large-scale construction activities. This provides crucial geographic and environmental background information for subsequent risk assessment and fault cause inference.

[0049] The data fusion and alignment engine is the technological cornerstone for achieving collaborative analysis of multi-source data. To address the performance challenges posed by heterogeneous multi-source data, this engine preferentially employs high-performance in-memory computing frameworks such as Apache Arrow. It loads data from different sources into a unified columnar in-memory format, thereby avoiding the expensive cross-system data serialization and deserialization overhead of traditional solutions. Using high-precision timestamps as the core alignment key, the engine leverages an efficient sliding time window algorithm to instantaneously correlate and align OTDR test events, network management alarm events, DAS vibration events, and static resource and geographic environment information related to the event occurrence points within a tiny time window (e.g., milliseconds). This forms a multimodal, spatiotemporally synchronized data snapshot, providing a "three-dimensional on-site view" for in-depth analysis by the intelligent analysis unit.

[0050] Specifically, the intelligent analysis and decision-making unit is the core technology of this invention. It integrates a series of analytical models to collaboratively complete the intelligent operation and maintenance of the fiber optic network. Please refer to [link / reference]. Figure 4 This unit includes a multimodal fusion fault location model, a dynamic scheduled inspection decision model based on long short-term memory networks, and a business-driven automatic optical path design engine.

[0051] The multimodal fusion fault location model aims to elevate the accuracy and precision of fault location to a new level. This model innovatively combines the advantages of one-dimensional convolutional neural networks (1D-CNN) and graph neural networks (GNN). The 1D-CNN component is designed to process one-dimensional time-series signals, i.e., OTDR test curve data. Like an experienced technician, it automatically extracts subtle waveform features of key events such as reflection, attenuation, and breakage from complex waveforms and outputs a feature vector representing the physical morphology of the fault. Simultaneously, the GNN component is responsible for modeling the physical topology of the fiber optic network. In this model, physical facilities such as towers and junction boxes are abstracted as nodes in the graph, and fiber optic cable segments are abstracted as edges connecting nodes. Information such as network management alarms and DAS vibration events obtained from other systems are dynamically attached as attributes to the nodes or edges corresponding to their occurrence locations. The GNN learns the impact range and propagation patterns of fault events on the network topology by propagating and aggregating neighborhood information on the graph structure. Finally, a fusion layer (such as a standard fully connected neural network) concatenates and jointly infers the waveform features extracted by the 1D-CNN and the topology-event features extracted by the GNN, thereby integrating multi-dimensional information and outputting a fault point GPS coordinate with the highest confidence. Its theoretical positioning accuracy can reach ±30 meters, which is far superior to the traditional method of judging solely by OTDR curves.

[0052] The core objective of the dynamic scheduled maintenance decision model based on Long Short-Term Memory (LSTM) networks is to shift from a "periodic" to a "selective" intelligent scheduled maintenance mode. This model treats the historical status data of each optical cable, including its historical failure frequency, attenuation values ​​from previous tests, repair records, and the severity of its environment, as a complex time series. Leveraging the excellent ability of LSTM networks to capture long-sequence dependencies, the model can learn and understand the inherent patterns of optical cable performance degradation over time, thereby predicting its health evolution trend and failure probability in a future cycle. The model's decision-making process further incorporates a configurable weighted scoring mechanism. Specifically, the next scheduled maintenance time for an optical cable is dynamically adjusted based on a baseline period (e.g., 12 months). The adjustment range depends on a comprehensive risk score, calculated by integrating factors from multiple dimensions. These factors primarily include: the future failure probability predicted by the LSTM model, the importance level of the services carried by the optical cable (e.g., optical cables carrying critical services such as relay protection have higher weights), and the environmental risk level of the optical cable's location obtained from the GIS system (e.g., optical cables located in highly corrosive areas have higher risks). By weighting and summing these factors, the model can dynamically output a global list of scheduled maintenance priorities, allowing limited maintenance resources to be prioritized for the optical cables with the highest current risk and greatest need for attention. For example, the scheduled maintenance cycle of an optical cable located in a highly corrosive area and carrying important services may be automatically shortened by more than 50% by the system.

[0053] Optionally, the dynamic scheduled inspection decision model can also employ reinforcement learning. Within this framework, the entire operation and maintenance system is treated as an agent, and the set of health states of all optical cables in the network constitutes the state space of the environment. The agent's actions determine which optical cables to immediately perform scheduled inspections on and which to postpone. The system's reward function is set to maximize operational efficiency in the long run, i.e., minimize the total operational cost (including inspection costs and losses due to faults). Through extensive exploration and trial and error in a simulated environment, the agent can autonomously learn an optimal scheduled inspection strategy to maximize long-term operational efficiency.

[0054] To achieve the aforementioned objectives, another technical solution provided by this invention is: a method for intelligent scheduled inspection and fault closed-loop processing of optical fiber networks. Please refer to [link / reference]. Figure 3 This method closely corresponds to the aforementioned system and includes the following steps:

[0055] Step S1: Establish a collaborative data view. After system startup, its multi-source heterogeneous data acquisition and preprocessing unit begins continuous operation, collecting operational data from multiple source systems such as Resource Management System (RMS), Troubleshooting Management System (TMS), Distributed Fiber Optic Sensing System (DAS), and Geographic Information System (GIS) through its internal data acquisition modules. The collected raw data is then sent to the data fusion and alignment engine, which performs standardization processing such as cleaning, formatting, and unit unification on this heterogeneous data. Finally, it aligns the data based on high-precision timestamps and spatial location information, forming and dynamically maintaining a unified, multi-dimensional real-time status view of the fiber optic network in the system's high-speed memory, serving as the foundation for all subsequent intelligent analyses.

[0056] Step S2: Trigger Fault Diagnosis and Precise Location. When the system detects a preset fault trigger event, such as a new breakpoint event reported by the OTDR test system or an alarm indicating a critical link interruption received by the fault management system (TMS), the collaborative processing flow is immediately and automatically triggered. The intelligent analysis and decision-making unit immediately acquires a snapshot of the collaborative data view within a short time window (e.g., ±5 minutes) before and after the fault event occurs, and inputs this snapshot data, which contains rich contextual information, into the multimodal fusion fault location model. This model comprehensively analyzes the OTDR waveform data in the snapshot, other alarm content associated with the optical path, the topological connection relationship of the fiber optic network, and DAS vibration event records (if any) around the fault point. Through its internal joint reasoning mechanism, it calculates and outputs a high-precision fault geographic coordinate and a preliminary judgment on possible fault causes (such as external damage, natural aging, etc.).

[0057] Step S3: Automated generation and association of work orders. Please refer to [link / reference]. Figure 5 Based on the precise positioning results output from the previous step, the collaborative work order and closed-loop processing unit automatically execute a series of work order operations. First, it creates a new "fault order" and automatically fills in key information, including a unique fault ID generated by the system, GPS coordinates output by the positioning model, the inferred fault type, and the estimated scope of affected services. Next, the system uses the optical path IDs in the resource library to automatically query and associate all "mode orders" (i.e., service circuit information configuration orders) directly affected by this fault. Then, the system parses the service level information recorded in these "mode orders" and uses this information to risk-label the "fault order." For example, if relay protection services are found to be included in the affected services, the fault order is automatically marked as "Level 1 Emergency."

[0058] Step S4: Intelligent Dispatch of Maintenance Tasks. Based on the generated and risk-marked "Fault Orders" and their associated "Method Orders," the system automatically generates a detailed "Maintenance Order" and dispatches it to the corresponding maintenance team. During the generation of the maintenance order, the business-driven automatic optical path design engine is activated. It queries the resource library in real time to find backup fiber cores or routes that can be used to temporarily restore services. The engine's optimization goals are multi-dimensional, not only meeting the Service Level Agreement (SLA) requirements of the affected services (such as bandwidth and latency), but also following risk avoidance principles, such as prioritizing backup routes that are not installed in the same trench or on the same pole as the faulty optical cable to prevent secondary disasters. The final generated maintenance order will automatically be filled with recommended backup resource IDs, estimated repair man-hours based on historical data of similar faults, and post-repair acceptance standards that comply with power safety regulations (e.g., splice loss must be less than or equal to 0.1dB).

[0059] Step S5: Perform Repair and Closed-Loop Verification. After frontline maintenance personnel complete on-site repair operations such as fiber splicing and fiber patching according to the instructions on the "Repair Order," they update the work order status to "Repair Completed" on their handheld mobile terminals. This simple operation triggers the system's closed-loop verification mechanism. The system backend automatically sends an instruction to the scheduled maintenance system or OTDR tester, requiring an immediate retest of the newly repaired optical path. The collaborative work order and closed-loop processing unit obtain the retested OTDR curve and intelligently compares it with the historical baseline curve of the optical path and the fault curve before repair. If the test results show that the optical path performance has been restored to the acceptance standard, the system automatically updates the fiber core status in the relevant "Mode Order" to "Normal," and simultaneously closes the associated "Fault Order" and "Repair Order," finally generating a closed-loop archive report containing a correlation graph of the three orders and a complete record of the processing process. Conversely, if the acceptance fails, the escalation mechanism will be automatically triggered, setting the inspection order status to "review failed" with the reason for the failure, and notifying the next-level technical expert team through system messages to intervene and ensure that the problem is completely resolved.

[0060] Optionally, in step S4, when multiple emergency failures occur in the network, causing competition among maintenance teams when allocating resources, the system will activate a resource conflict detection and arbitration mechanism. This mechanism preferably employs a Compare-and-Swap (CAS)-based optimistic locking strategy to ensure the atomicity of resource allocation operations and prevent multiple processes from simultaneously allocating the same backup fiber. Simultaneously, the system will queue conflicting resource requests according to preset service priority rules (e.g., relay protection services have the highest priority, followed by real-time production control services, and then management information services), ensuring that the recovery time of services most critical to the safe operation of the power grid is given the highest priority.

[0061] This invention also provides an intelligent scheduled inspection and fault closed-loop processing device for fiber optic networks. Physically, this device can be a server or server cluster deployed in a data center, or it can be an integrated dedicated hardware device. The device includes a high-performance processor (e.g., a multi-core CPU or a GPU for accelerating AI model calculations), a large-capacity memory (including high-speed RAM and persistent SSDs), and multiple high-speed network interfaces communicating with the processor for communication with various business systems and field devices. The memory stores a computer program that, when executed by the processor, can completely implement all the steps of the aforementioned detection method, providing an integrated and high-efficiency technical support platform for the intelligent operation and maintenance of power communication networks.

[0062] Example 1

[0063] This embodiment aims to illustrate the application of the system described in this invention in a specific scenario of handling external force damage to an optical cable line. Specifically, the system environment of this embodiment is deployed on a cluster consisting of three computing servers. Each server is equipped with a 32-core multi-core central processing unit (CPU) with a base clock frequency of 2.8GHz, 512GB of DDR4 main memory, and a graphics processing unit (GPU) with 48GB of high-bandwidth video memory (HBM2 type) for accelerating model calculations. The operating system is a Linux distribution containing kernel version 5.10, and it adopts a deep learning framework version 2.5.0 that supports the unified computing device architecture version 11.2. When the system detects an interruption alarm on an optical cable carrying power production control services, it immediately initiates the processing procedure. The multi-source heterogeneous data acquisition and preprocessing unit constructs a 5-minute data snapshot in memory before and after the fault. This snapshot integrates alarms from the fault management system (TMS), optical path topology from the resource management system (RMS), construction area layers along the line from the geographic information system (GIS), and abnormal vibration events with frequency characteristics consistent with excavator operation detected by the distributed optical fiber sensing system (DAS) deployed on the optical cable. The intelligent analysis and decision-making unit inputs this snapshot into the multimodal fusion fault location model. This model analyzes the breakage characteristics of the OTDR curve using a one-dimensional convolutional neural network (1D-CNN) and combines it with a graph neural network (GNN) to analyze the correlation between topology, alarms, and DAS vibration events, outputting the GPS coordinates of the fault point with a positioning error of 8 meters. The collaborative work order and closed-loop processing unit automatically generates "fault orders" and "repair orders" based on this, and associates them with the affected "mode orders." The entire repair and automatic review process is completed within 2.5 hours. Testing shows that this embodiment exhibits extremely high fault handling efficiency, and its specific performance data is shown in Table 1 below.

[0064] Example 2

[0065] This embodiment aims to illustrate the application of the method described in this invention in dynamically optimizing the scheduled maintenance cycle of optical cables to achieve predictive maintenance. The operating environment of this embodiment is the same as that of Embodiment 1. In this scenario, the dynamic scheduled maintenance decision model based on a Long Short-Term Memory (LSTM) network within the intelligent analysis and decision unit is activated to handle a critical 50-kilometer-long optical cable. A section of this cable (approximately 5 kilometers) traverses a highly corrosive soil area and carries two primary services: relay protection and dispatch automation. The model is input with historical data of the optical cable over the past three years, including time-series data such as attenuation values ​​from each OTDR test, ambient temperature and humidity, and historical minor fault records. The LSTM model predicts that the failure probability (α) of the optical cable section traversing the corrosive area in the next 12 months is 0.75. The system calculates the required maintenance cycle for the optical cable based on preset weights (w1=0.5, w2=0.3, w3=0.2), combined with the service weight carried by the optical cable (β=0.9) and the environmental risk level obtained from GIS (γ=0.8), using the formula T_new = T_base × (1 - sigmoid(w1*α + w2*β + w3*γ)). The baseline period T_base is 12 months. The calculation results dynamically shorten the scheduled maintenance cycle of the optical cable from 12 months to 5.5 months, thereby precisely allocating limited maintenance resources to the highest-risk sections and significantly reducing the potential failure rate. Specific performance data is shown in Table 1 below.

[0066] Example 3

[0067] This embodiment aims to illustrate the intelligent scheduling capability of the system described in this invention in dealing with concurrent faults and resource contention. The system environment is the same as in Embodiment 1. The scenario is set as two independent optical cable interruption faults being triggered simultaneously in a certain area due to extreme weather. Fault A affects the "relay protection" service, and fault B affects the "management information" service. Repairing both faults requires calling up the same spare optical cable resource. After the system completes the accurate location of the two faults through the multimodal fusion fault location model, when the collaborative work order and closed-loop processing unit generate the "repair order", the service-driven automatic optical path design engine detects resource conflicts. At this time, the system starts an optimistic locking mechanism based on Compare-and-Swap (CAS) to ensure the atomicity of resource allocation, and according to the preset service priority rules (relay protection > production real-time control > management information), it prioritizes the allocation of the only spare optical cable resource to the repair process of fault A. At the same time, the automatic optical path design engine plans another suboptimal backup route for fault B, although the path is longer, but physically isolated. This decision ensured that relay protection services, which are critical to grid security, were restored within 15 minutes, avoiding delays caused by resource contention. Specific performance data is shown in Table 1 below.

[0068] Comparative Example 1

[0069] This comparative example is used to compare with Example 1. The only difference between them is that the fault location method lacks the core multimodal data fusion and analysis capabilities, and only uses the traditional method based on single OTDR curve analysis in the existing technology to locate the fault. Specifically, the intelligent analysis and decision-making unit does not enable the graph neural network (GNN) module, nor does it integrate data from the DAS and GIS systems, and only analyzes the waveform data reported by the OTDR tester. Apart from this difference, the system hardware environment, software stack, fault scenario, and subsequent work order flow steps used in this comparative example are completely consistent with Example 1. Due to the lack of cross-validation of multi-source information, the error of its fault location result is as high as 250 meters, causing on-site repair personnel to spend a lot of extra time on manual line inspection and fault finding, significantly prolonging the fault repair time. Under the same conditions, the fault handling efficiency of this comparative example is much lower than that of Example 1, and its specific performance data can also be found in Table 1 below.

[0070] Comparative Example 2

[0071] This comparative example is used to compare with Example 2. The only difference between them is that the optical cable scheduled maintenance strategy adopts a fixed and unchanging periodic plan in the prior art, instead of using the dynamic risk prediction and decision-making method based on the LSTM model described in this invention. Specifically, for the same 50-kilometer critical optical cable described in Example 2, the maintenance system of this comparative example strictly follows the standard 12-month cycle for scheduled maintenance, regardless of its health status, the importance of the services it carries, or the environmental risks. Apart from this difference, the system hardware environment, software stack, and historical optical cable data on which the analysis is based are completely consistent with those of Example 2. Because it fails to identify and prioritize high-risk sections, this scheduled maintenance mode cannot effectively intervene in the potential degradation process in advance, resulting in extremely low effectiveness of its predictive maintenance and failing to significantly reduce the probability of future failures. Its specific performance data are also shown in Table 1 below.

[0072] Effect verification

[0073] To more intuitively demonstrate the beneficial effects of the present invention, the key performance test data of the above embodiments and comparative examples are summarized in Table 1.

[0074] Table 1 Performance Comparison Data Table

[0075]

[0076] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention in any other way. Any person skilled in the art may make changes or modifications to the above-disclosed technical content to create equivalent embodiments that can be applied to other fields. However, any simple modifications, equivalent changes, and modifications made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the protection scope of the present invention.

Claims

1. A fiber optic network intelligent scheduled inspection and fault closed-loop processing system, characterized in that, include: The multi-source heterogeneous data acquisition and preprocessing unit is configured to connect to the resource management system, fault network management system, distributed optical fiber sensing system and geographic information system to collect and standardize various operation and maintenance data, thereby generating a set of structured data streams with accurate timestamps and geographic location tags. The intelligent analysis and decision-making unit is configured to receive the structured data stream and perform in-depth analysis and decision-making on the state of the fiber optic network through at least one preset intelligent analysis model, so as to output decision instructions containing high-precision diagnostic results and dynamic strategy suggestions. The collaborative work order and closed-loop processing unit is configured to automatically create, associate, and transfer three types of work orders—method work orders, fault work orders, and repair work orders—based on the decision instructions, and to build and execute a fully automated closed loop from fault diagnosis, work order dispatch, repair to effect verification. The integrated control and presentation unit is configured to systematically integrate the decision-making instructions, the process status of the automated closed loop, and various operation and maintenance data, and generate a multi-dimensional monitoring dashboard and visualization report for operation and maintenance management personnel. The intelligent analysis and decision-making unit specifically includes: The multimodal fusion fault location model is used to receive data snapshots when a fault triggering event is detected, and to comprehensively analyze the optical time domain reflectometer test curve data, fiber optic network topology data, network management alarm content and distributed sensor event data contained in the data snapshot, so as to output the geographical coordinates of the fault point with the highest confidence. The dynamic scheduled inspection decision model based on long short-term memory network is used to learn the historical status data of optical cables as a time series to predict the evolution trend of their health and failure probability, and dynamically calculate and output the adjusted scheduled inspection cycle of optical cables accordingly. The business-driven automatic optical path design engine is used to query available backup fiber cores or routing resources in the network in real time when a failure occurs, and automatically design a backup optical path that meets the service level agreement requirements and follows the risk avoidance principle. The internal structure of the multimodal fusion fault location model includes: A one-dimensional convolutional neural network is designed to process one-dimensional time-series signals, namely the optical time-domain reflectometer test curve data, to automatically extract waveform feature vectors that characterize the physical form of the fault. A graph neural network is designed to model the physical topology of the fiber optic network, where network facilities are abstracted as nodes of a graph, fiber optic cable segments are abstracted as edges, and network management alarms and distributed sensing events obtained from other systems are dynamically attached as attributes to the nodes or edges corresponding to their occurrence locations in order to learn and extract topology-event features. A fusion layer is configured to concatenate and jointly infer the waveform feature vector and the topology-event features to output the geographic coordinates of the fault point by integrating multi-dimensional information.

2. The system according to claim 1, characterized in that, The collaborative work order and closed-loop processing unit are configured as follows: After receiving the accurate positioning result output by the intelligent analysis and decision-making unit, a new fault ticket is automatically created, and information such as fault ID, geographic coordinates, inferred fault type and estimated business impact range is automatically filled in. Using the optical path IDs in the resource database, automatically query and associate all mode orders directly affected by this fault; Based on the generated fault tickets and their associated method tickets, a detailed maintenance order is automatically generated and dispatched to the pre-set maintenance team.

3. The system according to claim 2, characterized in that, The collaborative work order and closed-loop processing unit are also configured to execute the closed-loop verification mechanism in the fully automated closed loop, the closed-loop verification mechanism including: After the maintenance personnel update the work order status to "repair completed", the system automatically sends an instruction to the testing system, requiring the repaired optical path to be retested immediately. The optical time domain reflectometer curve of the retest is obtained and intelligently compared with the historical reference curve of the optical path and the fault curve before the repair. If the comparison results show that the optical path performance has been restored to the preset acceptance standard, the fiber core status in the relevant method sheet will be automatically updated to normal, and the associated fault order and maintenance order will be closed at the same time. Finally, a closed-loop archive report containing a complete record of the processing process will be generated.

4. A method for intelligent scheduled inspection and fault closed-loop handling of optical fiber networks, characterized in that, Includes the following steps: Establish a collaborative data view by continuously collecting and integrating operation and maintenance data from multiple source systems such as resource management system, fault network management system, distributed optical fiber sensing system and geographic information system through multi-source heterogeneous data acquisition and preprocessing unit, so as to form and dynamically maintain a unified, multi-dimensional real-time status view of optical fiber network. Triggering fault diagnosis and precise location: When a preset fault triggering event is detected, the intelligent analysis and decision-making unit immediately obtains a data snapshot within a time window before and after the occurrence of the fault event, and inputs the data snapshot into a multimodal fusion fault location model. Through a joint reasoning mechanism, a high-precision fault geographic coordinate is calculated and output. The automated generation and association of work orders involves the collaborative work order and closed-loop processing unit automatically creating a new fault order based on the high-precision fault geographic coordinates output in the previous step, and automatically querying and associating all mode orders affected by the fault. The intelligent dispatch of maintenance tasks involves the collaborative work order and closed-loop processing unit automatically generating a maintenance order based on the generated fault order. This maintenance order is automatically filled with backup resource solutions recommended by the business-driven optical path automatic design engine. The system performs repair and closed-loop verification. After the on-site repair operation is completed, it automatically triggers the retest of the repaired optical path. Based on the comparison between the retest results and the preset acceptance standards, it automatically updates and closes the relevant work orders to complete the fault handling closed loop. The multi-source heterogeneous data acquisition and preprocessing unit specifically includes: The communication network management data acquisition module is used to acquire static resource data, real-time performance indicators and alarm information, and massive historical fault records from the resource management system and the fault network management system through standard protocols or dedicated adapters. The distributed sensing data interface module is used to connect to the distributed optical fiber sensing system and perform preliminary intelligent processing on the received raw high-frequency vibration or temperature data stream, refining it into structured event records. The spatial geographic information acquisition module is used to actively acquire the precise coordinates of the network facilities along the optical cable route and the environmental information layer along the route from the geographic information system through a standardized spatial query interface. The data fusion and alignment engine is used to load and spatiotemporally align the static resource data, real-time performance indicators and alarm information, historical fault records, event records, and precise coordinates and environmental information layers in a unified format to form a multimodal, spatiotemporally synchronized data snapshot. The intelligent analysis and decision-making unit specifically includes: The multimodal fusion fault location model is used to receive the data snapshot when a fault triggering event is detected, and to comprehensively analyze the optical time domain reflectometer test curve data, fiber optic network topology data, network management alarm content and distributed sensor event data contained in the data snapshot, so as to output the geographical coordinates of the fault point with the highest confidence. The dynamic scheduled inspection decision model based on long short-term memory network is used to learn the historical status data of optical cables as a time series to predict the evolution trend of their health and failure probability, and dynamically calculate and output the adjusted scheduled inspection cycle of optical cables accordingly. The business-driven automatic optical path design engine is used to query available backup fiber cores or routing resources in the network in real time when a failure occurs, and automatically design a backup optical path that meets the service level agreement requirements and follows the risk avoidance principle. The internal structure of the multimodal fusion fault location model includes: A one-dimensional convolutional neural network is designed to process one-dimensional time-series signals, namely the optical time-domain reflectometer test curve data, to automatically extract waveform feature vectors that characterize the physical form of the fault. A graph neural network is designed to model the physical topology of the fiber optic network, where network facilities are abstracted as nodes of a graph, fiber optic cable segments are abstracted as edges, and network management alarms and distributed sensing events obtained from other systems are dynamically attached as attributes to the nodes or edges corresponding to their occurrence locations in order to learn and extract topology-event features. A fusion layer is configured to concatenate and jointly infer the waveform feature vector and the topology-event features to output the geographic coordinates of the fault point by integrating multi-dimensional information.

5. The system according to claim 4, characterized in that, The data fusion and alignment engine is configured to: employ an in-memory computing framework, use high-precision timestamps as the core alignment key, and utilize a sliding time window algorithm to instantaneously associate and align test events, network management alarm events, sensor events, and static resource information and geographical environment information related to the event occurrence point that occur within a small time window.

6. The system according to claim 4, characterized in that, The dynamic scheduled maintenance decision model based on long short-term memory network is configured to calculate the adjusted optical cable scheduled maintenance cycle through a configurable weighted scoring mechanism. The inputs of the weighted scoring mechanism include: the future failure probability predicted by the long short-term memory network model, the importance level of the service carried by the optical cable, and the environmental risk level of the location of the optical cable obtained from the geographic information system.

7. A fiber optic network intelligent scheduled inspection and fault closed-loop processing device, which is physically deployed as a server or server cluster in a data center, characterized in that... The device includes a processor, memory, and a network interface that communicates with the processor in terms of hardware. The memory contains a computer program, which, when executed by the processor, implements all the steps of the intelligent scheduled inspection and fault closed-loop processing method for optical fiber networks as described in claim 4.

Images