Optical cable intelligent label full life cycle management method and system

Abstract

The invention discloses an optical cable intelligent label full life cycle management method and system, and the method comprises the steps: extracting structured data through an XML analysis engine based on a configuration file, carrying out the semantic analysis of an unstructured text, carrying out the regression quantification of text features through combining symbol quantiles, and constructing a knowledge graph; according to the knowledge graph, dynamically adjusting the character size, the two-dimensional code position and the error-tolerant rate by adopting a generative adversarial network technology according to the optical cable type and the pasting scene, and generating optical cable intelligent labels adaptive to different scenes; collecting a label image, and converting the label image into structured label identification data including an optical cable type, a connection relation and a transmission signal type by adopting a computer vision technology; the identification data and the knowledge graph are compared, consistency is checked, potential abnormity is analyzed, influences are evaluated, and graded early warning is generated; a micro-service architecture system is constructed, data analysis, AI generation, an intelligent recognition engine and knowledge graph service are integrated, and full-life-cycle management is achieved. The intelligent level of full-life-cycle management of the optical cable label is improved.

Description

Technical Field

[0001] This invention relates to the field of intelligent management technology for power systems, and in particular to a method and system for full lifecycle management of optical cable smart tags, applicable to the generation, identification, and operation and maintenance management of optical cable tags in intelligent substations. Background Technology

[0002] In the construction and operation of smart grids and substations, fiber optic cable tag management is a crucial link in ensuring the reliable operation of communication networks. Currently, the mainstream fiber optic cable tag management methods mainly include two modes: manual data entry to generate tags and barcode / QR code scanning and recognition. For example, traditional fiber optic cable tag management systems typically generate tags by manually entering design data, and then use handheld scanning devices for identification and querying. This method has low initial deployment costs and is simple to operate. Another common approach is to use a database-based management system, storing fiber optic cable information in a central database, and enabling information retrieval and updates through simple barcode scanning.

[0003] The closest current technology is a fiber optic cable tag management system that integrates basic image recognition capabilities. This system can use computer vision technology to identify simple tag content and match it with a backend database. The system uses a static data structure to store fiber optic cable information, generates fiber optic cable tags using preset tag templates, and achieves tag recognition and parsing through basic image processing algorithms. In the operation and maintenance phase, the system also provides a simple maintenance record function, supporting the recording of fiber optic cable maintenance history.

[0004] However, this existing technology has obvious technical shortcomings: First, the tag generation process relies on manual input of design data, which is prone to errors and inefficient; second, the tag recognition technology has poor adaptability to complex scenarios such as dirt and tilt on optical cable tags, and the recognition success rate is less than 90% in actual harsh environments; third, it lacks intelligent data association and analysis capabilities, and cannot effectively cope with the dynamic changes in physical and logical circuit scenarios of smart substations, making it difficult to support real-time monitoring and anomaly early warning of optical cable connections; finally, the system architecture is closed, making it difficult to interact and integrate with other systems, which limits the management efficiency of the entire life cycle of optical cables. Summary of the Invention

[0005] The purpose of this invention is to provide a method and system for full lifecycle management of optical cable smart tags, aiming to solve the technical problems in the prior art, such as reliance on manual input for optical cable tag generation, poor adaptability of identification technology, lack of intelligent data correlation analysis capabilities, and closed system architecture.

[0006] To achieve the above objectives, this invention provides a method for full lifecycle management of optical fiber smart tags, comprising the following steps: Based on the intelligent substation configuration file, which includes intelligent substation SCD and SPCD files, the structured data including descriptions of logical nodes, logical devices, communication connections and physical devices is extracted through an XML parsing engine. Natural language processing technology is used to perform semantic analysis on unstructured text, and the text features are quantified by the symbolic quantile regression method to construct a knowledge graph containing the relationship between devices, ports, optical cables and virtual terminals. Based on the knowledge graph, generative adversarial network technology is used to dynamically adjust the text size, QR code position and fault tolerance according to the type of optical cable and the pasting scenario, so as to generate smart optical cable labels that are adapted to different scenarios. Based on the aforementioned smart optical cable tags, tag images of the optical cables on site are collected, and multimodal fusion computer vision technology is used to convert them into structured tag recognition data containing optical cable type, connection relationship and transmission signal type. Based on the structured tag recognition data and the knowledge graph, the consistency between tag information and expected connection relationships is checked through real-time data comparison algorithms, potential anomalies are analyzed, and the security impact, reliability impact, and functional impact of anomalies are evaluated using multi-objective evolutionary algorithms to generate graded early warning information and targeted handling suggestions. A modular system based on a microservice architecture is constructed, including a data parsing engine, an AI generation engine, an intelligent recognition engine, and a knowledge graph service. Through standardized API interfaces, it supports data exchange and functional integration with other systems in smart substations, realizing intelligent management of the entire lifecycle of optical cable tags from generation and recognition to operation and maintenance.

[0007] Preferably, the intelligent substation configuration file, based on intelligent substation SCD and SPCD files, extracts structured data including descriptions of logical nodes, logical devices, communication connections, and physical devices using an XML parsing engine. Natural language processing techniques are used to perform semantic analysis on the unstructured text, and the text features are quantified using a symbolic quantile regression method. This constructs a knowledge graph containing relationships between devices, ports, optical cables, and virtual terminals, including: Based on the intelligent substation configuration file, graph neural network technology is used to mine the implicit connection relationships between devices, identify the indirect associations across optical cables, and establish a multi-level graph structure association network with an entity layer containing physical devices, ports, optical cables and virtual terminals, a relationship layer containing connection relationships, transmission relationships and attribution relationships, and an attribute layer containing device models, port protocols and the number of optical fiber cores. By using reservoir computing technology to predict potential optical cable connection patterns based on historical connection patterns, and employing Lyapunov index analysis to dynamically adjust the reliability of the prediction strategy, missing optical cable usage labels and spare fiber core information are automatically completed to obtain the completed label information. Based on the completed tag information and historical operation and maintenance data including historical operation and maintenance records, alarm logs, optical cable test data and tag usage feedback, a multi-objective evolutionary algorithm with diverse decision space is used to perform multi-dimensional data fusion while considering data integrity, consistency and timeliness. Based on the uniformity metric of Hamming distance, connection relationship conflicts, outdated information and abnormal patterns are detected and corrected to obtain the verified and optimized knowledge graph.

[0008] Preferably, the quantization of text features using the combined signed quantile regression method includes: For the unstructured text containing cabinet name, port description, and virtual terminal signal, use a power industry professional thesaurus and rule base to perform word segmentation and part-of-speech tagging, and perform symbolic processing to map professional terms and descriptions into symbolic representations to generate symbolic text; Based on the symbolized text, a quantile-based regression model is established to map the semantic features of the text to different quantiles, analyze the correlation strength between symbols, calculate the conditional quantiles of different features to determine the importance of symbols in optical cable tags, and determine the importance weights in optical cable tags. Based on the aforementioned importance weights, the following are extracted: optical cable types including trunk optical cable, pigtail cable, and single-mode / multimode optical fiber; connection relationships of signal flow and logical relationships between devices; transmission signal types including GOOSE signal, SV sample value, and MMS message; priority information including critical protection signal and non-critical monitoring signal; and quantized text features are obtained.

[0009] Preferably, the generation of fiber optic smart tags adapted to different scenarios includes: The knowledge graph is used to analyze the physical space constraints, environmental factors and viewing conditions of different application scenarios, including narrow cabinet spaces, outdoor environments and high-frequency viewing areas. The reservoir computing technology is used to learn from historical tag usage data to predict tag usage environment parameters and generate an initial tag template candidate set including compact tags, waterproof and weather-resistant tags and wear-resistant tags. Based on the initial candidate set of label templates, label content data containing fiber optic cable ID and connection information, and target scene parameters, generative adversarial network technology is used to generate label design schemes according to fiber optic cable type and pasting scene. A discriminator evaluates the information integrity, readability and recognition rate of the label design based on historical label samples. Content constraints are added to ensure that key information is clearly visible, QR code constraints are added to adjust the size, position and error tolerance according to the usage environment, spatial constraints are added to optimize the overall layout according to the pasting position, and visual constraints are added to ensure the contrast between text and background, thus generating an optimized label layout scheme. Based on the optimized label layout scheme and historical label usage data including label durability and scanning success rate, a reinforcement learning model is constructed. The Lyapunov index is used to analyze the impact of parameter adjustments on label performance, adaptively optimizing QR code encoding density, error correction level, and UV resistance encoding capability, adaptively optimizing text font size, thickness, and contrast, and adaptively optimizing material parameters including label thickness, adhesive type, and protective layer processing, to obtain the optical cable smart label.

[0010] Preferably, generative adversarial network technology is used to optimize the label design, including: Using a large number of historical tag samples containing successful cases, failed cases, and expert annotations, along with corresponding usage effect evaluations, as training data, a generative adversarial network model was trained, which included a generator responsible for generating tag design schemes and a discriminator for evaluating whether the generated tag designs met the expected standards. Content constraints were added to ensure that key information was clearly visible, QR code constraints to automatically adjust the size and position of the QR code and the fault tolerance rate according to the usage environment, spatial constraints to optimize the overall layout according to the available space of the pasting position, and visual constraints to ensure the contrast between text and background to improve readability, thus forming a GAN model under constraints. Based on the GAN model under the aforementioned constraints, special adjustments were made to different types of ODF tags that simplify information display and highlight port numbers and optical cable routes, device connection tags that emphasize information at both ends of the equipment and clearly indicate signal types, and tail cable tags that optimize information layout under small size to ensure recognizability. The resolution limitations of printing equipment, the characteristics of commonly used label materials with different ink absorption properties, possible deformation during the pasting process, and visual changes during the aging process were all taken into account to generate a variety of label design schemes. Based on the various label design schemes, through multiple rounds of optimization and iteration, the optimized label layout scheme is obtained by simultaneously considering functional requirements to meet label recognition and information transmission requirements, aesthetic enhancement to improve visual effects, and practicality to ensure ease of use.

[0011] Preferably, the conversion into structured tag identification data containing optical cable type, connection relationship, and transmission signal type includes: Based on the aforementioned optical cable smart tag, the tag image is collected by a mobile terminal, combined with supplementary environmental data obtained in areas with tag damage, obstructed view, and high-density optical cable, and the light intensity, collection angle, and collection distance are recorded simultaneously. The image enhancement algorithms of adaptive histogram equalization, nonlocal mean filtering, and super-resolution reconstruction are applied, and the geometric correction algorithm and illumination compensation algorithm are applied to generate a standardized tag image with uniform brightness, contrast, viewing angle, and resolution. Based on the standardized label image, the QR code region is located and extracted through gradient analysis, morphological verification, and localization point detection. A super-resolution reconstruction technique combining residual learning network and perceptual loss function is used to process low-resolution or blurry QR codes. Edge repair techniques that preserve structural integrity, context-aware filling, and topology are used to process partially worn or broken QR codes. An attention mechanism that combines spatial attention to focus on the effective region, channel attention to dynamically adjust feature weights, and multi-scale fusion to capture local and global features is used. QR code recognition is achieved through adaptive binarization, perspective correction, and error correction enhancement decoding to obtain the QR code recognition result. Based on the QR code recognition results, a multi-objective evolutionary algorithm with decision space diversity is introduced, which encodes key parameters and strategies of the recognition process into decision vectors. An initial solution set with diversity is generated based on the uniformity metric of Hamming distance. Multiple complementary recognition strategies are obtained through a two-layer selection process using Pareto advantage and decision space diversity. Simultaneously, optical character recognition is performed on the text information on the label image. A multi-modal fusion technique, employing information consistency verification, complementary information integration, and redundant information utilization, is used to fuse the QR code information with the text recognition results. Natural language processing technology is then used to convert this data into a standardized data structure, resulting in structured label recognition data containing information about the optical cable type, connection relationship, and transmission signal type.

[0012] Preferably, the step of acquiring tag images of the optical cables on-site based on the smart optical cable tags, and converting them into structured tag recognition data containing optical cable type, connection relationship, and transmission signal type using multimodal fusion computer vision technology, further includes: The key parameters and strategies in the identification process are encoded into decision vectors, the Hamming distance between each decision vector in the solution set is calculated, and a density-based diversity evaluation function is designed to use the diversity index as an additional optimization objective. Based on the diversity evaluation function, an initial solution set with diversity is generated. New candidate solutions are generated through mutation and crossover to explore the decision space. A two-layer selection is performed based on Pareto advantage and then based on the diversity of the decision space. The diversity weight is dynamically adjusted according to the convergence of the optimization process, the learning rate is adjusted according to the sensitivity of different parameters, and an adaptive adjustment mechanism is used to dynamically balance exploration and utilization to obtain a variety of complementary recognition strategies. Based on the aforementioned complementary recognition strategies, a strategy that focuses more on illumination compensation is selected in strong light interference environments, a more aggressive image restoration strategy is selected in severe damage situations, and a simplified strategy with higher computational efficiency is selected when computational resources are limited. The recognition method most suitable for the current situation is dynamically selected or combined according to the environmental conditions detected in real time, thereby enhancing the diversity of the decision space.

[0013] Preferably, the step of relying on the structured tag recognition data and the knowledge graph, checking the consistency between tag information and expected connection relationships through a real-time data comparison algorithm, analyzing potential anomalies, and using a multi-objective evolutionary algorithm to evaluate the security, reliability, and functional impacts of the anomalies and generate graded early warning information and targeted handling suggestions includes: Based on the structured label recognition data and the knowledge graph, multi-dimensional feature matching is used to compare topological relationships and logical functions. A fuzzy matching strategy of character similarity calculation, semantic similarity evaluation and structural similarity analysis is adopted. Design document data, historical scanning records and related optical cable data are integrated for multi-source data collaborative verification. The effective prediction time method of reservoir computer is used to learn from historical anomaly cases and configure anomaly patterns including mismatch, label error, unauthorized change and potential risk. The Lyapunov index is calculated to evaluate the time scale and range of anomaly impact. The case of physical connection not conforming to design drawings is detected to obtain anomaly detection results. Based on the anomaly detection results and tagged operation and maintenance history data including fault history, maintenance records, and performance trends, a multi-objective evolutionary algorithm is used to assess the severity and potential impact of the anomaly from multiple perspectives, including the degree of security threat, reliability impact, functional impact (assessing the scope of functional loss), and economic impact (assessing repair costs and resources). A balance is sought across multiple assessment dimensions, taking into account load conditions, seasonal factors, and the contextual sensitivity of concurrent events. Based on the severity of the assessment, tiered warning information is generated, including emergency warnings, important warnings, general warnings, and attention prompts. Case reasoning based on similar anomalies is retrieved from a historical case library, and rule reasoning based on application domain expert rules is used to provide targeted handling suggestions, including emergency response plans, temporary mitigation plans, fundamental solutions, and prevention strategies. Based on the execution results of the targeted handling suggestions and the actual connection status confirmed on-site, a change source identification and multi-level change confirmation process is established, including planned changes, on-site discovery and fault handling. Graph neural network technology is used to add, modify or delete entity nodes and connection relationships affected by changes using an incremental update strategy. The direct impact, indirect impact and redundancy assessment of network topology changes are re-evaluated. The connection relationship data in the knowledge graph is dynamically updated and the change version history is maintained to achieve dynamic association and anomaly early warning.

[0014] Preferably, the modular system built on a microservice architecture includes a data parsing engine, an AI generation engine, an intelligent recognition engine, and a knowledge graph service. It supports data exchange and functional integration with other systems in the smart substation through standardized API interfaces, enabling intelligent management of the entire lifecycle of optical cable tags from generation and recognition to operation and maintenance. The design includes a mobile terminal equipped with a GPU to support AI inference, a high-definition camera with image stabilization and fill light, an RFID identification module for auxiliary identification, and a hardware system that supports label printing equipment for various materials. By using edge computing technology to optimize the terminal's processing capabilities to support AI inference and image processing, the design ensures the availability of data processing and label identification in weak network or offline environments, resulting in a usable hardware system. Based on the available hardware system, a core module is developed, including a data parsing engine for processing the configuration files of the smart substation using NLP and GNN technologies, an AI generation engine for label optimization and generation using GAN and reinforcement learning technologies, an intelligent recognition engine for multimodal parsing using CV and multimodal technologies, and a knowledge graph service for storing relationships and historical data. Containerization technology is used to achieve loose coupling and flexible deployment between modules, forming a microservice architecture that supports independent development, deployment, and expansion. Based on the aforementioned microservice architecture, a layered data storage strategy is designed for different types of data to achieve separation of hot and cold data. A multi-level security mechanism is constructed, including data encryption to protect sensitive information, access control to restrict operation permissions, and operation audit records for system use. Standardized API interfaces are developed to support data exchange and functional integration with other substation systems such as SCADA systems and asset management systems. A flexible plug-in mechanism that allows third parties to develop extended functions is designed to achieve full lifecycle management.

[0015] This invention also provides a full lifecycle management system for optical fiber smart tags, comprising: The data parsing module is used to extract structured data, including descriptions of logical nodes, logical devices, communication connections, and physical devices, based on the intelligent substation configuration files, including SCD and SPCD files. It uses an XML parsing engine to extract structured data, including descriptions of logical nodes, logical devices, communication connections, and physical devices. It uses natural language processing technology to perform semantic analysis on unstructured text and combines the symbolic quantile regression method to quantify text features, and constructs a knowledge graph containing the relationships between devices, ports, optical cables, and virtual terminals. The tag generation module is used to dynamically adjust the text size, QR code position and fault tolerance based on the knowledge graph and generative adversarial network technology according to the optical cable type and the pasting scenario, so as to generate smart optical cable tags that are adapted to different scenarios. The tag recognition module is used to collect tag images of the optical cable on site based on the optical cable smart tag, and use multimodal fusion computer vision technology to convert them into structured tag recognition data containing optical cable type, connection relationship and transmission signal type; The anomaly warning module is used to check the consistency between the tag information and the expected connection relationship by relying on the structured tag recognition data and the knowledge graph through real-time data comparison algorithm, analyze potential anomalies, evaluate the safety impact, reliability impact and functional impact of the anomalies using multi-objective evolution algorithm, and generate graded warning information and targeted handling suggestions. The system integration module is used to build a modular system based on a microservice architecture, including a data parsing engine, an AI generation engine, an intelligent recognition engine, and a knowledge graph service. It supports data exchange and functional integration with other systems in smart substations through standardized API interfaces, enabling intelligent management of the entire lifecycle of optical cable tags from generation and recognition to operation and maintenance.

[0016] The present invention has the following beneficial effects: 1. It enables automatic parsing of optical cable connection relationships from intelligent substation configuration files, constructs a comprehensive knowledge graph, significantly reduces manual data entry errors, and improves data accuracy; 2. By using generative adversarial networks and reinforcement learning techniques, adaptive optimization of label design was achieved, making the labels more suitable for the usage needs of different scenarios and improving the practicality and durability of the labels; 3. Innovatively, a multi-objective evolutionary algorithm with diverse decision space is introduced, combined with a uniformity metric based on Hamming distance, which significantly improves the label recognition accuracy under non-ideal conditions such as dirt and tilt, increasing the recognition success rate from 90% of the traditional method to over 99.5%; 4. An effective prediction time method based on reservoir computing, combined with a multi-objective evolutionary algorithm, enables accurate detection and assessment of optical cable connection anomalies, supporting the generation of tiered early warnings and targeted handling suggestions; 5. By adopting a microservice architecture and standardized API interfaces, an open and scalable system framework has been built, which supports seamless integration with other systems and enables full lifecycle management from tag generation and recognition to operation and maintenance. Attached Figure Description

[0017] To more clearly illustrate the technical solutions in the embodiments of this application, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1 This is a flowchart of the optical cable smart tag full lifecycle management method of the present invention; Figure 2 This is a structural diagram of the optical cable smart tag full lifecycle management system of the present invention. Detailed Implementation

[0019] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0020] like Figure 1 As shown, the optical cable smart tag full lifecycle management method provided by the present invention includes the following steps: S1: Based on the intelligent substation configuration file including the intelligent substation SCD file and SPCD file, the structured data including the description of logical nodes, logical devices, communication connections and physical devices is extracted through the XML parsing engine. Natural language processing technology is used to perform semantic analysis on the unstructured text, and the text features are quantified by the symbolic quantile regression method to construct a knowledge graph containing the relationship between devices, ports, optical cables and virtual terminals. In the fiber optic cable smart tag full lifecycle management method of this invention, step S1 is executed first, which involves constructing the fiber optic cable tag smart data model layer. This step is the foundation of the entire process. By deeply parsing and processing the configuration files of the smart substation, a knowledge graph required for fiber optic cable tag management is established. The SCD (System Configuration Description) file and SPCD (Substation Physical Connection Diagram) file in the smart substation are two core configuration files, describing the logical and physical connection relationships of the substation, respectively. First, these files are parsed using a specially designed XML parsing engine to extract key structured data, including logical node (LN) information (such as protection functions, measurement functions, etc.), logical device (LD) information (clearly defining functional groups), communication connection information (including message types such as GOOSE and SMV and communication paths), and physical device descriptions (such as cabinet location, terminal block number, etc.). During the parsing process, a distributed processing architecture is adopted to process the complex configuration files of large substations in segments in parallel, and a strict error detection and fault tolerance mechanism is established to ensure the accuracy and integrity of the data.

[0021] For unstructured text in substation configuration files, Natural Language Processing (NLP) technology was introduced for deep analysis. This unstructured text includes equipment descriptions, port names, signal specifications, etc., typically written in natural language and containing rich implicit semantics. A custom-designed power industry-specific thesaurus and rule base were used for word segmentation and part-of-speech tagging, paying particular attention to power-specific terms and expressions, such as "110kV busbar PT signal" and "circuit breaker trip control circuit." To transform this textual information into quantifiable features, a novel Signed Quantile Regression (SQR) method was employed. The SQR method maps textual semantic features to different quantiles, analyzes the correlation strength between symbols, calculates the conditional quantiles of different features, and determines the importance and weight of each element in the optical cable tag. This method can accurately extract fiber optic cable types (such as trunk cables, pigtails, single-mode / multimode fibers, etc.), connection relationships (signal flow and logical relationships between devices), transmission signal types (such as GOOSE signals, SV sample values, MMS messages, etc.), and priority information (such as critical protection signals, non-critical monitoring signals, etc.) from unstructured text. Using these extracted features, a knowledge graph containing complete relationships between devices, ports, fiber optic cables, and virtual terminals is constructed, providing a solid data foundation for subsequent tag generation and identification.

[0022] S2: Based on the knowledge graph, generative adversarial network technology is used to dynamically adjust the text size, QR code position and fault tolerance according to the type of optical cable and the pasting scenario to generate smart optical cable tags adapted to different scenarios. Step S2 implements the intelligent generation layer for optical cable tags. This step, based on the previously constructed knowledge graph, uses advanced artificial intelligence technology to generate intelligent optical cable tags adapted to different scenarios. First, Reservoir Computing (RC) technology is used to analyze the characteristics of different application scenarios, including physical space constraints (such as confined spaces in cabinets and densely packed cable trays), environmental factors (such as outdoor exposure environments and high-temperature areas), viewing conditions (such as low-light environments and long-distance identification requirements), and usage frequency. RC technology processes historical tag usage data to learn the correlation between tag parameters and usage effects under different scenarios, thereby predicting the optimal tag design parameters. A training dataset containing scenario features and tag effect evaluations is constructed, the key hyperparameters of the RC model are optimized, and an initial set of candidate tag templates optimized for different scenarios is generated, such as compact tags (suitable for confined spaces in cabinets), waterproof and weather-resistant tags (suitable for outdoor environments), and wear-resistant tags (suitable for high-frequency viewing areas).

[0023] After obtaining the initial candidate set of label templates, Generative Adversarial Network (GAN) technology is introduced to deeply optimize the label design. The GAN model consists of a generator and a discriminator. The generator is responsible for generating label design schemes based on the fiber optic cable type and pasting scenario, while the discriminator evaluates the information completeness, readability, and recognition rate of the design based on historical label samples. Multiple constraints are incorporated into the GAN training: content constraints ensure key information is clearly visible; QR code constraints adjust size, position, and fault tolerance according to the usage environment; spatial constraints optimize the overall layout based on the pasting location; and visual constraints ensure contrast between text and background. Special adjustments are also made for different types of fiber optic cable labels, such as ODF labels (simplifying information display, highlighting port numbers and fiber optic cable routing), inter-device connection labels (emphasizing information from both ends of the device and clearly indicating the signal type), and tail cable labels (optimizing information layout in small sizes). The label designs generated by GAN consider various practical factors such as the resolution limitations of the printing equipment, label material characteristics, possible deformation during pasting, and visual changes during aging.

[0024] Among them, the application of Generative Adversarial Network (GAN) technology to the design optimization of optical cable labels is an innovative artificial intelligence solution. It automatically generates label design schemes that meet specific requirements through deep learning algorithms. This technology combines artistic design with engineering needs, solving problems such as high reliance on human experience, low standardization, and poor adaptability in traditional label design, and realizing intelligent and personalized customization of label design.

[0025] From a technical perspective, the GAN model of this invention comprises specially designed generator and discriminator networks. The generator adopts a U-Net architecture, including an encoder-decoder structure and skip connections, which effectively preserves information details. The encoder extracts feature representations of input conditions (such as fiber optic cable type and scene parameters) through multi-layer convolutions, while the decoder reconstructs the complete label design through transposed convolutions. The discriminator employs a multi-task learning structure, evaluating not only the realism of the design but also key metrics such as information completeness, readability, and expected recognition rate. To adapt to the specific needs of label design, the GAN model introduces a conditional control mechanism, injecting design requirements (such as label type and usage environment) as conditional vectors into the generation process to ensure that the generated results meet the specific scenario requirements. Model training uses an improved Wasserstein GAN loss function, adding multiple task-specific loss terms, such as content preservation loss, clarity loss, and structural consistency loss, which are weighted and combined to form the final optimization objective.

[0026] The detailed implementation steps include six key stages: First, requirements analysis and parameter extraction are conducted, analyzing the label usage requirements and extracting key design parameters such as functional requirements, environmental conditions, space constraints, and recognition methods, converting these requirements into structured condition vectors. The second stage involves initial design generation, where a generator network generates 10-20 candidate design schemes based on the condition vectors and random noise. Each scheme includes a complete layout, color scheme, font settings, and QR code configuration. The third stage performs multi-dimensional quality assessment, with a discriminator scoring each scheme for information completeness, readability, expected recognition rate, and aesthetics, and calculating the overall quality score through weighted aggregation. The fourth stage verifies constraints, checking whether the design meets hard constraints such as information priority, QR code quality, minimum font size, and space utilization, and eliminating unqualified schemes. The fifth stage conducts environmental adaptability simulation, simulating the visual effects and performance of the labels under different conditions in a virtual environment, including different lighting, viewing angles, material aging, and soiling, and evaluating the QR code recognition rate and calculating the environmental adaptability index. The final stage involves iterative optimization and solution selection. The system selects 2-3 optimal solutions for 3-5 rounds of in-depth optimization, addressing weaknesses specifically, such as increasing the contrast of key information, adjusting QR code parameters, or optimizing the layout. Finally, by integrating all evaluation metrics, the optimal solution is selected, and high-resolution design documents and detailed implementation guidelines are generated.

[0027] Through this deeply optimized process, GAN technology has transformed from traditional manual design to intelligent, data-driven design. It not only reduces design time from hours to minutes but also increases the tag recognition success rate by 15-20%. At the same time, it ensures the applicability of the design in various complex environments, providing strong technical support for the optical cable management of smart substations.

[0028] To further optimize label parameters, a reinforcement learning model was constructed, utilizing the Lyapunov exponent to analyze the impact of parameter adjustments on label performance. The reinforcement learning model includes a state space (describing the current parameter configuration of the label), an action space (possible parameter adjustment operations), a reward function (evaluating the merits of parameter adjustments based on label usage performance), and a policy network (learning the optimal parameter adjustment strategy). The Lyapunov exponent is used to assess parameter stability and predict long-term effects, adaptively optimizing QR code parameters (encoding density, error correction level, UV resistance), text parameters (font size, thickness, contrast), and material parameters (label thickness, adhesive type, protective layer treatment). The reinforcement learning model continuously adjusts its parameter selection strategy based on successful and failed cases in historical data, such as analyzing labels that failed in high-temperature environments to optimize material selection and adjusting text and QR code parameters based on the output effects of different printing devices. Through this data-driven parameter optimization, high-quality fiber optic smart labels adapted to different scenarios can be generated, significantly improving the practicality and durability of the labels.

[0029] The construction and training process of the reinforcement learning model can be illustrated through a field application case in a substation. First, we conducted a comprehensive data collection on 100 installed fiber optic tags within the substation, including their complete parameter configurations (such as QR code size, error correction level, material type, etc.) and actual performance data under different environmental conditions (such as recognition success rate, aging rate, etc.). Based on this data, an initial state space was constructed, using a 16-dimensional vector to represent the tag parameter configuration, with each dimension corresponding to a key parameter. The action space was designed as discrete parameter adjustment operations, such as "increase the QR code size by 10%" and "increase the error correction level by one level," defining a total of 42 basic adjustment operations.

[0030] The reward function design is the core of the model, and a comprehensive score is constructed by combining multiple performance indicators: R = 0.4 × recognition success rate + 0.3 × durability index + 0.2 × information integrity + 0.1 × cost-effectiveness. Here, the recognition success rate is obtained through actual testing under different conditions; the durability index is estimated based on accelerated aging tests and field usage data; information integrity assesses the ability of the label to maintain readability during use; and cost-effectiveness considers the cost changes resulting from parameter adjustments.

[0031] The policy network employs a Deep Q-Network (DQN) architecture, comprising four fully connected layers, each with 128, 256, 128, and 42 neurons respectively (the last layer corresponds to the number of actions). Training utilizes an experience replay technique, maintaining a replay buffer containing 10,000 experience records. The model is first pre-trained in a simulated environment using an environment simulator built with historical data, allowing the model to learn the relationship between parameters and performance by experimenting with different parameter adjustments. After pre-training, the model is deployed to a real-world environment for online learning. Whenever new labels are generated and real-world usage feedback is obtained, the experience base is updated and the model is fine-tuned.

[0032] Lyapunov exponent analysis played a crucial role in the training process. For example, it was found that in outdoor high-humidity environments, when the QR code module size was less than 0.8 mm and the label thickness was less than 0.2 mm, the Lyapunov exponent rose sharply, indicating extreme instability in this parameter range, and the recognition rate would rapidly decline over time. Based on this analysis, the model learned a strategy to automatically increase the QR code module size and label thickness for outdoor high-humidity areas.

[0033] After about three months of training, the model gradually converged to an efficient strategy. A typical application case is that when it is detected that the tag will be deployed near a cooling device (where temperature fluctuations are large and humidity is high), the model automatically recommends upgrading the QR code error correction level from the standard L level to H level, while changing the tag material from ordinary PVC to composite PET material and adding a UV-resistant coating.

[0034] S3: Based on the aforementioned optical cable smart tag, collect the tag image of the optical cable on site, and use multimodal fusion computer vision technology to convert it into structured tag recognition data containing optical cable type, connection relationship and transmission signal type; Step S3 develops an intelligent identification layer for fiber optic cable tags. This step utilizes multimodal fusion computer vision technology to achieve high-precision identification of fiber optic cable tags in the field. In the complex environment of substations, tags may be in various non-ideal states, such as uneven lighting, tilted angles, partial obstruction, or dirt, which poses challenges to tag identification. First, high-definition cameras mounted on mobile terminals (such as industrial tablets or smartphones) are used to acquire tag images. These cameras have features such as high resolution, autofocus, image stabilization, and supplemental lighting. Simultaneously, supplemental environmental data is also collected, including RFID-assisted identification (especially in areas with damaged tags, obstructed vision, or high-density fiber optic cables) and environmental parameters (such as light intensity, acquisition angle, and distance). After acquiring the raw data, a series of preprocessing steps are performed, including image enhancement (adaptive histogram equalization, nonlocal mean filtering, super-resolution reconstruction), geometric correction (perspective transformation, distortion correction, size normalization), illumination compensation (illumination model estimation, reflection component separation, shadow removal), and region localization (tag detection, region of interest extraction, multi-tag separation), etc. These processes convert raw images acquired under various conditions into standardized, high-quality label images, providing ideal input data for subsequent QR code recognition.

[0035] For QR codes on labels, the QR code region is located and extracted based on a standardized image, and then processed using integrated CV enhancement algorithms. These algorithms include super-resolution reconstruction technology (based on a residual learning network combined with a perceptual loss function, targeting low-resolution or blurry QR codes), edge restoration technology (handling partially worn or broken QR codes through structural integrity analysis, context-aware filling, and topology preservation), and attention mechanisms (spatial attention focusing on the effective region, channel attention dynamically adjusting feature weights, and multi-scale fusion capturing local and global features). After these processes, high-precision QR code recognition is achieved through adaptive binarization, perspective correction, and error correction-enhanced decoding. To further improve recognition capabilities under extreme conditions, a multi-objective evolutionary algorithm with diverse decision spaces is innovatively introduced. This algorithm encodes key parameters and strategies in the recognition process into decision vectors, evaluates the diversity of the decision space using a Hamming distance-based uniformity metric, and generates multiple complementary recognition strategies. This enables the dynamic selection or combination of the most suitable recognition method based on real-time detected environmental conditions, significantly improving label recognition accuracy under non-ideal conditions such as soiling and tilting. Finally, optical character recognition is performed on the text information on the label, and the QR code information is integrated with the text recognition results through multimodal fusion technology to transform it into structured label recognition data containing fiber optic cable type, connection relationship and transmission signal type.

[0036] Among them, the multi-objective evolutionary algorithm with decision space diversity is an innovative computational optimization method specifically designed for the extreme condition recognition problem of QR codes in optical cable labels. Based on the principle of multi-objective evolutionary computation, this algorithm achieves robust optimization of QR code recognition under various complex environments by maintaining a highly diverse solution set in the decision space. Unlike traditional single recognition strategies, this method constructs an integrated system containing multiple complementary recognition strategies, capable of handling various unforeseen recognition challenges.

[0037] From a technical perspective, this algorithm encodes key decision parameters in the QR code recognition process into multi-dimensional decision vectors, including preprocessing parameters (such as filter type, kernel size, enhancement intensity, etc.), binarization parameters (such as thresholding methods, local window size, etc.), perspective correction parameters (such as transformation matrix estimation methods, reference point selection strategies, etc.), and decoding parameters (such as sampling rate, fault tolerance strategies, etc.). Based on this, the algorithm defines two main optimization objectives: maximizing recognition accuracy and maximizing decision space diversity. Decision space diversity is evaluated using a Hamming distance-based uniformity metric to ensure that the generated strategy set is uniformly distributed in the decision space, covering different types of recognition scenarios.

[0038] The algorithm's implementation comprises four key stages. First, population initialization involves randomly generating a set of decision vectors, each representing a recognition strategy. The population size is typically set to 20-30 individuals to ensure initial diversity. The second stage is fitness evaluation, assessing the performance of each decision vector across multiple objectives. Recognition accuracy is calculated by measuring the average success rate on a standard test set (containing QR code samples with varying degrees of dirt, tilt, and lighting variations). Decision space diversity is evaluated by calculating the average Hamming distance between the current individual and other individuals in the population; a larger distance indicates higher diversity. The third stage involves evolutionary operations, including elite retention, crossover, and mutation. The algorithm selects elite individuals using non-dominated sorting and crowding distance calculation; it generates new offspring strategies in the decision space using simulated binary crossover (SBX); and it introduces random variations using polynomial mutation to prevent premature convergence. The evolutionary process typically runs for 100-200 generations until the Pareto front stabilizes. The final stage is strategy selection and deployment, selecting a representative subset of recognition strategies (typically 5-8 strategies) from the final non-dominated solution set and deploying them into the actual recognition system. During real-time recognition, the environmental conditions of the current QR code are first quickly assessed (such as the degree of dirt, tilt angle, and uniformity of illumination). Then, the most suitable strategy or combination of strategies is selected for recognition through simple decision rules or machine learning classifiers.

[0039] S4: Based on the structured tag recognition data and the knowledge graph, the consistency between the tag information and the expected connection relationship is checked through a real-time data comparison algorithm, potential anomalies are analyzed, and the security impact, reliability impact and functional impact of the anomalies are evaluated using a multi-objective evolutionary algorithm, and graded early warning information and targeted handling suggestions are generated. Step S4 constructs an intelligent association layer for optical cable tags, enabling dynamic association and anomaly warning between optical cable tags and actual physical connections. First, a real-time data comparison mechanism is established, matching the tag information identified in step S3 with the knowledge graph constructed in step S1. This comparison involves not only primary key information such as the optical cable ID but also comprehensive comparison of multiple dimensions of features, including basic attributes (optical cable type, specifications, length, etc.), topological relationships (starting device, ending device, port number, etc.), and logical functions (transmission signal type, communication protocol, etc.). Considering potential naming variations or subtle differences in the actual environment, a fuzzy matching algorithm is adopted, including character similarity calculation, semantic similarity evaluation, and structural similarity analysis. Simultaneously, information from multiple sources is integrated for cross-validation, including design document data, historical scan records, and relevant optical cable data. To detect potential anomalies, an innovative analysis using the effective prediction time method of the reservoir computer is employed. By learning historical anomaly cases through the RC model, various anomaly patterns can be identified, such as configuration mismatch, tag errors, and unauthorized changes. The severity and impact range of the anomalies are assessed using the Lyapunov index.

[0040] Among them, the Reservoir Computing (RC) model learning of historical anomaly cases is an innovative method combining nonlinear dynamic systems theory and machine learning, specifically designed for the detection of time-series anomalies in optical cable tag management. This technology leverages RC's computational efficiency and superior processing capabilities for time-series data to analyze historical anomaly patterns, construct predictive models, and achieve early identification and warning of potential optical cable connection anomalies.

[0041] The core principle of the RC model is based on dynamic systems theory, projecting the input signal into a high-dimensional nonlinear space for processing. Its basic structure comprises three key components: an input layer, a reservoir, and an output layer. The input layer receives the data stream from the fiber optic tag, including identification results, scan timestamps, and operator information. The reservoir is a recurrent network containing a large number of randomly connected neurons, exhibiting "echo state" characteristics, capable of retaining historical information from the input signal. The output layer extracts useful features from the reservoir state through simple linear regression to predict anomaly probabilities. Unlike traditional deep learning methods, the RC model only trains the output layer weights, keeping the reservoir weights fixed, significantly reducing computational complexity and training difficulty, making it highly suitable for edge computing environments.

[0042] The implementation of this model in optical cable tag management comprises four main stages. The first stage is data preprocessing and feature engineering. The system collects historical optical cable tag scanning records, extracts temporal features (such as scanning frequency and time interval changes), spatial features (such as scanning location distribution and movement patterns), and content features (such as tag information consistency and change frequency), and performs normalization processing. The second stage is the construction of an anomaly case library. Through historical record analysis and expert annotation, a case library containing various typical anomalies is established, covering configuration mismatches (such as optical cable type not matching system records), tag errors (such as incorrect port information), unauthorized changes (such as connection changes during unplanned maintenance), and other situations. A severity level is assigned to each anomaly. The third stage is RC model design and training. Based on anomaly detection requirements, a suitable reservoir topology is designed, typically using an echo state network (ESN) with 500-1000 neurons. Appropriate sparsity (approximately 5%) and spectral radius (typically 0.8-0.95) are set to ensure the network has sufficient memory capacity and dynamic complexity. Input data is injected into the RC model in time series order to activate reservoir neurons. Then, only the output layer weights are trained to match the model output with known anomaly case labels. To improve model robustness, cross-validation and regularization techniques are used to prevent overfitting. The fourth stage is anomaly prediction and Lyapunov index analysis. Real-time fiber optic cable label data is processed by the trained RC model to generate anomaly probability scores. A preliminary warning is triggered when the score exceeds a preset threshold. The Lyapunov index is further calculated to quantify the instability of the system state. This index is calculated based on the divergence rate of the RC model's state vector; a higher index value indicates that the system may undergo drastic changes in a short period, indicating the severity and potential impact range of the anomaly. Based on the Lyapunov index and anomaly type, a graded warning is automatically generated, along with possible root cause analysis and suggested remedial measures.

[0043] When anomalies are detected, a multi-objective evolutionary algorithm is used to assess the severity and potential impact of the anomaly from multiple perspectives, including security impact (the degree of threat to security), reliability impact (the impact on reliability), functional impact (the range of potential functional losses), and economic impact (the cost and resources required for repair). The multi-objective evolutionary algorithm seeks a balance across multiple assessment dimensions, while considering contextual factors of the current operating environment, such as load, seasonal factors, and concurrent events. Based on the assessment results, tiered early warning information is generated, including emergency warnings (serious anomalies requiring immediate attention), important warnings (anomalies requiring prompt attention but not immediate response), general warnings (minor anomalies that can be handled during routine maintenance), and attention alerts (potential risks or situations to be observed). Simultaneously, based on historical handling experience and best practices, targeted handling recommendations are generated, including emergency response plans, temporary mitigation plans, fundamental solutions, and prevention strategies. Finally, dynamic updates to connectivity relationships are supported. Based on anomaly handling results and the actual connection status confirmed on-site, graph neural network technology is used to incrementally update the knowledge graph, reassess the impact of network topology changes, keep the data model synchronized with the actual physical connections, and form a closed-loop management system.

[0044] S5: Build a modular system based on a microservice architecture, including a data parsing engine, an AI generation engine, an intelligent recognition engine, and a knowledge graph service. Support data exchange and functional integration with other systems in the smart substation through standardized API interfaces, and realize intelligent management of the entire lifecycle of optical cable tags from generation and recognition to operation and maintenance.

[0045] Step S5 establishes the architecture for a full lifecycle management system for optical fiber tags. This step integrates the aforementioned technical layers to build a complete system framework. A comprehensive solution, including both hardware and software, was designed first. At the hardware level, the system includes a mobile terminal equipped with a GPU (supporting AI inference), a high-definition camera (with image stabilization and supplemental lighting), an RFID-assisted identification module, and a tag printing device. Edge computing technology optimizes the terminal's processing capabilities, ensuring data processing and tag identification availability in weak network or offline environments. At the software level, the system is based on a microservice architecture and has developed several core modules: a data parsing engine (combining NLP and GNN technologies to process SCD / SPCD files), an AI generation engine (combining GAN and reinforcement learning to optimize tags), an intelligent recognition engine (combining CV and multimodal technologies for tag recognition), and a knowledge graph service (storing relationships and historical data). Containerization technology is used to achieve loose coupling and flexible deployment between modules, forming a microservice architecture that supports independent development, deployment, and expansion.

[0046] In terms of data management, a hierarchical data storage strategy was designed, employing appropriate storage schemes for different types of data such as knowledge graphs, historical data, and user information, achieving separation of hot and cold data. Simultaneously, a multi-level security mechanism was constructed, including data encryption (protecting sensitive information), access control (restricting operational permissions), and operation auditing (recording system usage), ensuring the security and integrity of optical cable data. To support integration with other systems, a standardized API interface was developed, supporting data exchange and functional integration with other substation systems (such as SCADA and asset management systems). Furthermore, a flexible plug-in mechanism was designed, allowing third parties to develop extended functions, forming an open optical cable tag management ecosystem. Through this modular and scalable system architecture, this invention achieves intelligent management of the entire lifecycle of optical cable tags, from generation and identification to operation and maintenance, improving the efficiency and reliability of optical cable management in smart substations.

[0047] Step S1 includes the following sub-steps: S1.1: Based on the intelligent substation configuration file, graph neural network technology is used to mine the implicit connection relationships between devices, identify the indirect associations across optical cables, and establish a multi-level graph structure association network with an entity layer containing physical devices, ports, optical cables and virtual terminals, a relationship layer containing connection relationships, transmission relationships and attribution relationships, and an attribute layer containing device models, port protocols and the number of optical cable cores. S1.2: Based on the multi-level graph structure association network, reservoir computer technology is used to predict potential optical cable connection patterns based on historical connection patterns. The reliability of the prediction strategy is dynamically adjusted by Lyapunov index analysis. Missing optical cable usage labels and spare fiber core information are automatically filled in to obtain the completed label information. S1.3: Based on the completed tag information and historical operation and maintenance data including historical operation and maintenance records, alarm logs, optical cable test data and tag usage feedback, a multi-objective evolutionary algorithm with diverse decision space is used to perform multi-dimensional data fusion while considering data integrity, consistency and timeliness. Based on the uniformity metric of Hamming distance, connection relationship conflicts, outdated information and abnormal patterns are detected and corrected to obtain the verified and optimized knowledge graph.

[0048] Specifically, step S1.1 details how to construct a multi-level graph structure network based on the smart substation configuration file. In a smart substation, complex connections exist between devices, including directly visible physical connections and implicit logical relationships. Traditional methods often only capture explicit direct connections, neglecting potential indirect connections, which limits the depth and breadth of fiber optic cable management. This step innovatively introduces Graph Neural Network (GNN) technology to deeply explore the implicit connections between devices. GNN can simultaneously consider node characteristics and topology, learning deep interactions between nodes through message passing mechanisms, making it particularly suitable for handling complex connections within a substation. Through GNN, indirect connections across fiber optic bays can be identified. For example, when two protection devices in different bays are connected by multiple fiber optic segments, GNN can infer their logical connections, even if such connections are not directly represented in the original configuration file. Meanwhile, a multi-layered graph structure was constructed based on the extracted data: the entity layer includes physical devices (such as transformers, circuit breakers, protection devices, etc.), ports (such as fiber optic interfaces, Ethernet ports, etc.), optical cables (backbone optical cables, pigtail cables, etc.), and virtual terminals (such as GOOSE publishing points, subscription points, etc.); the relationship layer defines the connection relationships between entities, such as "connected to," "transmitting," "belonging to," etc.; the attribute layer describes the characteristics of each entity, such as the device model, the communication protocol of the port, the number of fiber cores in the optical cable, etc. This multi-layered graph structure design enables the comprehensive capture of the complex relationships between devices within the substation, including physical connection relationships and logical functional relationships, providing a comprehensive and in-depth data view for subsequent tag generation and anomaly detection.

[0049] Among them, Graph Neural Networks (GNN) technology, applied to the mining of implicit connections between devices in smart substations, is a cutting-edge deep learning method specifically designed to handle complex association patterns in non-Euclidean data structures. In the substation optical cable management scenario, GNN can overcome the limitations of traditional methods, discovering functional associations that do not exist in the configuration files but are not directly labeled, providing a deeper semantic understanding for the optical cable labeling system.

[0050] The core principle of GNNs is based on the message passing mechanism of graph-structured data. Unlike traditional neural networks that process images or sequences, GNNs are specifically designed to process graph structures composed of nodes (substation equipment) and edges (connections). Through an iterative message passing process, each node continuously aggregates information from its neighboring nodes and updates its own representation. As the number of layers increases, nodes can acquire information from increasingly distant neighbors, thereby capturing indirect relationships in multi-hop connections. This characteristic makes GNNs particularly suitable for discovering implicit functional relationships in substations formed by multiple fiber optic cable connections, such as the logical relationships between cross-bay protection devices.

[0051] The implementation of GNN in fiber optic cable management in smart substations includes four main stages. The first stage is graph construction and initialization. The system extracts device information, communication configurations, and connection relationships from the IEC 61850 configuration file (SCD file) to construct the initial graph structure. Each physical device, communication port, and fiber optic cable is represented as a node in the graph, and direct physical connections are represented as edges. The initial feature vector of a node contains attribute information such as device type, function code, and physical location; the features of an edge include connection type and media type. For communication services such as GOOSE and MMS, virtual nodes are created to represent publishing and subscribing points, and connections are established with the corresponding physical devices. The second stage is GNN model design and training. Considering the characteristics of substations, a hybrid architecture including a multi-layer graph convolutional network (GCN) and a graph attention network (GAT) is adopted. The GCN layer captures the overall structural features, while the GAT layer highlights important connections through an attention mechanism, paying particular attention to fiber optic cable paths carrying critical protection signals. The model training employs a semi-supervised learning method, utilizing known device functional associations as training labels, while introducing a topology-preserving loss function to ensure the retention of key topological characteristics of the substation network during the learning process. The third stage is the discovery and verification of implicit connections. The trained GNN model identifies potential implicit functional associations by analyzing the similarity and connection patterns in the node embedding space. The system uses a path importance-based algorithm to evaluate the criticality of different optical cable paths in information transmission and identify device pairs that are physically separated but functionally closely related. To verify the discovered implicit connections, the system combines electrical protection logic analysis and historical operating data to confirm the validity and importance of these connections. The final stage is multi-level graph structure integration. The system integrates the discovered implicit connections into the original graph structure, constructing a more complete multi-level association network. At the entity layer, virtual association nodes are added to represent functional associations; at the relation layer, new edge types such as "functional dependency" and "alternate path" are introduced to describe the properties of implicit connections; at the attribute layer, attributes such as importance scores and signal types are added to these new associations. The integrated graph structure not only represents physical connections, but also includes functional dependencies and information flow paths, providing a comprehensive semantic context for the optical fiber tagging system.

[0052] Step S1.2 utilizes reservoir computing technology to conduct in-depth analysis and completion of the multi-level graph structure association network. In practical applications, substation configuration files may have missing or incomplete data, such as unclear labeling of optical cable usage or missing spare fiber core information, which can affect the accuracy of subsequent label generation and management. To address this issue, this step introduces reservoir computing (RC) technology to predict potential optical cable connection patterns based on historical connection patterns. RC is a special type of recurrent neural network whose core feature is a fixed, randomly initialized reservoir, training only the output layer weights. This gives RC significant advantages in training efficiency and processing time-series data. First, the RC model is trained using historical connection data, optimizing its hyperparameters such as reservoir size and spectral radius. For the prediction of optical cable connection patterns, historical connection data is serialized and input into the RC model, utilizing its "memory" capability to capture the time dependence of connection patterns. The reliability of the prediction results is evaluated using the effective prediction time (VPT) characteristic of RC. VPT (Virtual Predictive Time) is a key metric for RC (Reliable Response Time), representing the longest period during which effective predictions can be maintained, and is crucial for assessing the long-term validity of prediction results. Furthermore, Lyapunov index analysis is employed to dynamically adjust the prediction strategy. The Lyapunov index quantifies sensitivity to changes in initial conditions, helping to identify areas of uncertainty in the prediction and adjusting the prediction strategy accordingly to improve the accuracy of long-term predictions. In this way, missing tag information can be automatically completed, such as inferring the purpose of unlabeled optical cables based on historical connection patterns, or inferring the possible use of spare fiber cores based on equipment function and topological location. This automatic completion capability significantly reduces the need for manual intervention, improves the completeness and accuracy of the knowledge graph, and provides a more reliable data foundation for generating high-quality optical cable tags.

[0053] Step S1.3 achieves multi-dimensional data fusion and consistency verification through a multi-objective evolutionary algorithm with diverse decision spaces. In practical applications, data from different sources may be inconsistent, conflicting, or outdated. Directly using this data may lead to incorrect labeling information and management decisions. This step integrates the previously constructed knowledge graph with historical operation and maintenance data in multiple dimensions and ensures data quality through rigorous consistency verification. The integrated data sources include historical operation and maintenance records (actual maintenance status), alarm logs (indicating potential problems), fiber optic cable test data (such as OTDR test results), and label usage feedback (such as label scan success rate, durability, etc.). Data fusion adopts a hierarchical strategy, including data cleaning (standardizing data from each source), entity alignment (identifying different descriptions of the same equipment or fiber optic cable in different data sources), temporal correlation (establishing an event timeline to correlate the state changes of the same fiber optic cable or equipment at different points in time), and spatial correlation (establishing a spatial distribution model of the data by combining substation physical layout information). To ensure data consistency, a multi-objective evolutionary algorithm with diverse decision spaces is innovatively introduced. Unlike traditional multi-objective optimization algorithms that primarily focus on the diversity of the objective space, this algorithm specifically focuses on the diversity of the decision space, evaluating the distribution of candidate solutions in the decision space using the uniformity metric of Hamming distance. In its implementation, multiple evaluation metrics (such as structural integrity, data timeliness, and conflict level) are set to generate multiple candidate solutions, each representing a possible data correction strategy. A Pareto optimal solution is selected through non-dominated sorting, balancing multiple optimization objectives such as data integrity, consistency, and timeliness. This step detects and corrects connection conflicts (such as mismatches between logical connections described in the SCD file and physical connections recorded in the SPCD file), outdated information (connections that have changed but are not reflected in the configuration file), abnormal patterns (relationships that significantly deviate from normal connection patterns), and missing data (necessary but missing connection descriptions inferred from contextual information). Finally, this step outputs a validated and optimized high-quality intelligent data model for optical cable tags. This model not only contains accurate connection relationships but also integrates historical operational experience and best practices, providing a comprehensive and reliable data foundation for subsequent tag generation and management.

[0054] The quantization of text features using the combined signed quantile regression method includes: For the unstructured text containing cabinet name, port description, and virtual terminal signal, use a power industry professional thesaurus and rule base to perform word segmentation and part-of-speech tagging, and perform symbolic processing to map professional terms and descriptions into symbolic representations to generate symbolic text; Based on the symbolized text, a quantile-based regression model is established to map the semantic features of the text to different quantiles, analyze the correlation strength between symbols, calculate the conditional quantiles of different features to determine the importance of symbols in optical cable tags, and determine the importance weights in optical cable tags. Based on the aforementioned importance weights, the following are extracted: optical cable types including trunk optical cable, pigtail cable, and single-mode / multimode optical fiber; connection relationships of signal flow and logical relationships between devices; transmission signal types including GOOSE signal, SV sample value, and MMS message; priority information including critical protection signal and non-critical monitoring signal; and quantized text features are obtained.

[0055] Specifically, processing unstructured text is a key challenge in the configuration file processing of smart substations. This text includes cabinet names (e.g., "110kV surge arrester cabinet", "primary equipment monitoring and control panel"), port descriptions (e.g., "fiber optic interface TX1", "optical splitter port P3"), and virtual terminal signals (e.g., "circuit breaker trip command", "measurement output"). To effectively process this unstructured text and extract valuable features, this invention innovatively combines the symbolic quantile regression method to achieve precise quantification of text features. The specific implementation process of this method consists of three main stages: symbolic processing, quantile regression modeling, and feature extraction.

[0056] In the symbolization stage, a specialized thesaurus and rule base customized for the power industry are first used to accurately segment and tag the unstructured text. This thesaurus contains tens of thousands of power-specific terms, covering equipment names (such as "circuit breaker," "disconnector," and "transformer"), technical parameters (such as "rated voltage," "rated current," and "protection setting"), and professional abbreviations (such as "PT" for voltage transformer and "CT" for current transformer). Simultaneously, the rule base includes naming patterns and expression rules specific to the power industry, such as the naming convention of "equipment number + function + parameter" (e.g., "circuit breaker trip control circuit No. 1"). Using a specialized word segmenter based on these resources, complex power-related text is accurately decomposed into the smallest semantic units. Subsequently, these segmentation results are symbolized, mapping the specialized terms and descriptions to symbolic representations. For example, "220kV line protection device" is mapped to the symbol "HVL_PROT," and "fiber optic splitter output port" is mapped to the symbol "FOSP_OUT." This symbolization process not only standardizes text representation but also reduces the dimensionality and complexity of the text, facilitating subsequent mathematical modeling and analysis. In this way, the original unstructured text is transformed into a sequence of symbols containing specific semantic information, forming symbolized text.

[0057] In the quantile regression modeling stage, a quantile-based regression model is established based on the symbolic text generated in the previous step. Unlike traditional mean regression, quantile regression can analyze different quantiles in the data distribution, providing more comprehensive distribution information, and is particularly suitable for handling data with outliers and asymmetric distributions. First, the symbolic text is converted into feature vectors, with each symbol corresponding to one dimension. Then, the quantile regression model is trained using an expert-annotated training dataset (containing symbols and their importance scores in different fiber optic cable labels). This model can map the semantic features of the text to different quantiles; for example, it can analyze the weight performance of each symbol at the 25th, 50th, 75th, and 90th quantiles. Through this analysis, the performance characteristics of symbols in different contexts can be obtained, avoiding the information loss that may be caused by simple mean. At the same time, the correlation strength between symbols is analyzed, and the mutual influence between symbols is considered through the conditional quantile model. For example, when the symbols "circuit breaker" and "trip" appear simultaneously, their combined importance may be much higher than the sum of their individual importance. By calculating the conditional quantiles of different features, the importance of each symbol in the optical cable tag is determined, and finally, the importance weight of each symbol in the optical cable tag is determined. These weights reflect the relative importance of each symbol in the tag information, providing a scientific basis for subsequent feature extraction.

[0058] In the feature extraction stage, based on the importance weights determined in the previous step, deep feature extraction is performed on the original text. First, fiber optic cable type information is extracted, including trunk cables (backbone cables connecting major devices), pigtail cables (short cables extending from the trunk cables to connect to terminal devices), and single-mode / multimode fibers (distinguished by the fiber's transmission mode). This is achieved by identifying specific identifiers (such as "TRUNK"). CABLEBy combining the signal flow indicators (such as "->", "=>") and analysis symbol patterns, the type information of the optical cable can be accurately extracted. Secondly, connection relationship information is extracted, including the signal flow direction (transmitter and receiver) and logical relationships (such as control relationships, dependency relationships) between devices. This process relies on the combination of signal flow indicators (such as "->", "=>") and relationship descriptors (such as "control", "PIGTAIL") to accurately extract the type information of the optical cable. The process involves identifying and analyzing signals belonging to specific protocol identifiers (such as "GOOSE", "SMV", and "MMS") and analyzing contextual information. Next, the transmission signal type is extracted, including GOOSE signals (used for rapid event transmission between devices), SV sample values ​​(used for measurement data transmission), and MMS messages (used for routine data exchange). The transmission signal type is determined by identifying specific protocol identifiers (such as "GOOSE", "SMV", and "MMS") and analyzing contextual information. Finally, priority information is extracted to distinguish between critical protection signals (such as trip signals and interlocking signals) and non-critical monitoring signals (such as status monitoring and parameter display). Signal priority is determined by identifying keywords (such as "protection", "trip", and "monitoring") and combining their importance weights. Through these steps, the raw unstructured text is transformed into structured, quantified text features, providing an accurate data foundation for the generation and management of optical cable tags.

[0059] This text quantization method based on symbolic quantile regression offers significant advantages. First, it effectively handles specialized terminology and expressions unique to the power industry, overcoming the limitations of general natural language processing methods in specialized applications. Second, the quantile regression model captures the full picture of data distribution, focusing not only on the mean but also analyzing feature performance at different quantiles, providing more comprehensive information. Third, by analyzing the correlation strength between symbols through conditional quantile analysis, important symbol combinations can be identified and reinforced, improving the accuracy of feature extraction. Fourth, the feature extraction process based on importance weights ensures the relevance and importance of the extracted results, providing a scientific basis for the design of fiber optic cable tags. Finally, the quantized text features possess good interpretability and traceability, contributing to continuous optimization and knowledge accumulation. In practical applications, this method has been validated in multiple smart substation projects, achieving a text feature extraction accuracy rate exceeding 95%, significantly improving the quality and efficiency of fiber optic cable tag generation.

[0060] Step S2 includes the following sub-steps: S2.1: Utilize the knowledge graph to analyze the physical space constraints, environmental factors, and viewing conditions of different application scenarios, including narrow cabinet spaces, outdoor environments, and high-frequency viewing areas. Learn from historical tag usage data using reservoir computing technology to predict tag usage environment parameters and generate an initial tag template candidate set, including compact tags, waterproof and weather-resistant tags, and wear-resistant tags. S2.2: Based on the initial candidate set of label templates, label content data containing optical cable ID and connection information, and target scene parameters, generative adversarial network technology is adopted. The generator generates a label design scheme according to the optical cable type and pasting scene. The discriminator evaluates the information integrity, readability and recognition rate of the label design based on historical label samples. Content constraints are added to ensure that key information is clearly visible, QR code constraints are added to adjust the size, position and fault tolerance according to the usage environment, spatial constraints are added to optimize the overall layout according to the pasting position, and visual constraints are added to ensure the contrast between text and background, thereby generating an optimized label layout scheme. S2.3: Based on the optimized label layout scheme and historical label usage data including label durability and scanning success rate, a reinforcement learning model is constructed. The Lyapunov exponent is used to analyze the impact of parameter adjustments on label performance. The QR code encoding density, error correction level, and UV resistance encoding capability are adaptively optimized. The text font size, thickness, and contrast are adaptively optimized. The material parameters including label thickness, adhesive type, and protective layer processing are adaptively optimized to obtain the optical cable smart label.

[0061] Specifically, step S2.1 details the scene-adaptive label template design process. In the generation of fiber optic labels, different application scenarios have different requirements for the labels. For example, outdoor environments require weather-resistant labels, while confined spaces require compact labels. To achieve precise label design adaptation, the characteristics and constraints of different application scenarios are comprehensively analyzed based on the intelligent data model of fiber optic labels output in step S1. These characteristics and constraints include physical space constraints (such as confined spaces in cabinets, dense cable trays, and under-ceiling / floor cabling spaces), environmental factors (such as outdoor exposure environments, high-temperature areas, humid environments, and vibration areas), viewing conditions (such as low-light environments, long-distance identification requirements, and high-frequency viewing scenarios), and usage frequency (such as high-frequency areas for daily inspections, emergency handling areas, and areas for regular maintenance). For these different scenarios, labels with different parameters need to be designed to meet actual usage needs.

[0062] To scientifically predict parameter requirements for various scenarios, Reservoir Computing (RC) technology was innovatively applied. RC is a special type of recurrent neural network with a "reservoir" structure, capable of efficiently processing time-series data. It has low computational resource requirements and a simple training process, making it particularly suitable for embedded and real-time applications. A training dataset containing various scene features and label performance evaluations was constructed. Scene features included temperature range (-40℃ to +85℃), humidity (relative humidity 0% to 100%), lighting conditions (0 to 100,000 lux), and spatial dimensions (from a few millimeters to tens of centimeters). Label performance evaluations included recognition success rate (percentage of success rate under different conditions), durability (from several months to over ten years), and readability (ease of recognition by the human eye). Key hyperparameters for optimizing the RC model include reservoir size (which determines model complexity, typically ranging from hundreds to thousands of nodes), spectral radius (which affects kinetic properties and is usually set between 0.8 and 1.2), and regularization coefficient (which controls overfitting and is typically adjusted between 0.01 and 0.1). By adjusting these parameters, the predictive performance of the RC model can be maximized, generating optimal label parameter configurations for different scenarios.

[0063] Based on the optimized RC model, the optimal label parameters for new scenarios are predicted, generating an initial candidate set of label templates. These candidate templates are specifically optimized for different scenarios. For example, for confined spaces like display cabinets, compact labels are generated, using the minimum effective size (typically 20×40mm), optimizing information layout to ensure core information (such as fiber optic cable ID and connection endpoints) is prominently displayed, while secondary information (such as detailed descriptions) can be appropriately simplified or reduced. For outdoor environments, waterproof and weather-resistant labels are designed, increasing material thickness (typically above 0.5mm), using UV-resistant ink, adding a waterproof layer and an anti-oxidation coating, and improving text contrast to adapt to strong light environments. For frequently accessed areas, abrasion-resistant labels are designed, using bold fonts and high-contrast color schemes, using abrasion-resistant materials (such as polyester, polycarbonate, etc.), and adding a protective layer to extend service life. For low-light environments, labels with enhanced visibility are designed, using fluorescent materials or reflective elements, increasing font size, improving contrast, and optimizing the error correction level of the QR code to improve scanning success rate. This data-driven tag template design method can generate the most suitable initial tag design for different application scenarios, providing a good starting point for further optimization.

[0064] Step S2.2 introduces Generative Adversarial Network (GAN) technology to optimize the label design. After the initial label template is generated, it needs to be deeply optimized based on the specific optical cable information to ensure the practicality and readability of the labels. Therefore, based on the initial label template candidate set generated in step S2.1, and combined with label content data (such as optical cable ID, connection information, signal type, etc.) and target scene parameters, GAN technology is used for deep optimization. A GAN consists of a generator and a discriminator; through adversarial learning between these two networks, high-quality label design schemes can be generated.

[0065] The generator network is responsible for generating label designs based on input conditions (such as scene parameters, fiber optic cable information, etc.). Its architecture employs a deep convolutional neural network (DCNN), containing multiple convolutional layers, batch normalization layers, and activation functions. To improve generation performance, a conditional control mechanism is introduced, allowing the generation process to be controlled by additional inputs. For example, the generation strategy can be adjusted using scene conditional vectors (encoding physical space, environmental factors, etc.), or the information layout can be controlled using content conditional vectors (encoding fiber optic cable type, importance, etc.). The discriminator network evaluates whether the generated label designs meet expected standards, including information completeness, readability, and recognition rate. The discriminator also uses a DCNN architecture but incorporates an attention mechanism to focus on key areas in the labels (such as QR codes, core text information, etc.). During training, a large number of historical label samples and their performance evaluations are used as training data, including successful cases (label designs with high recognition rates and high durability), failed cases (label designs that are difficult to recognize and easily damaged), and expert-annotated label design samples (high-quality labels designed and evaluated by professionals).

[0066] The innovation of GAN training lies in the inclusion of multiple constraints to ensure that the generated label designs are both aesthetically pleasing and practical. Content constraints ensure the clear visibility of key information by grading different content elements through information importance scoring (based on the weight analysis in step S1.3), ensuring that key information (such as fiber optic cable ID and connection endpoints) is highlighted, while secondary information (such as detailed descriptions) can be appropriately simplified. QR code constraints automatically adjust the QR code size, position, and fault tolerance based on the usage environment. In high-interference environments (such as outdoors or dusty areas), the QR code size is increased and the fault tolerance level is improved (typically Q or H level). In controlled environments, a more compact QR code can be used to save space. Spatial constraints optimize the overall layout based on the available space at the pasting location, considering the space limitations of different installation locations (such as fiber optic cable surfaces, near ports, patch panels, etc.), and dynamically adjusting the label shape and size. Visual constraints ensure the contrast between text and background, improving readability by automatically selecting the optimal text-background color combination to ensure good readability under various lighting conditions, and considering colorblind-friendly design. During the optimization process, GAN makes specific adjustments to different types of optical cable tags, such as ODF tags (simplifying information display and highlighting port numbers and cable routing), inter-device connection tags (emphasizing information about both ends of the device and clearly indicating signal type), and tail cable tags (optimizing information layout in small sizes). Through multiple rounds of iteration and fine-tuning, an optimized tag layout scheme is finally generated, which meets functional requirements while also taking into account aesthetics and practicality.

[0067] Step S2.3 employs a reinforcement learning model to adaptively adjust the label parameters. In the final stage of label design, various parameters need fine-tuning to adapt to the specific needs of different usage environments. To this end, this sub-step innovatively introduces reinforcement learning (RL) technology to adaptively optimize key label parameters based on historical usage data. The core idea of ​​reinforcement learning is to learn the optimal strategy through interaction with the environment, making it highly suitable for solving parameter optimization problems that require balancing multiple objectives and constraints.

[0068] The core components of a reinforcement learning model include a state space, an action space, a reward function, and a policy network. The state space describes the current parameter configuration of the label, including QR code parameters (size, position, fault tolerance level, etc.), text parameters (font, size, color, etc.), material parameters (material type, thickness, protective layer, etc.), and scene features (temperature, humidity, lighting, etc.). The action space defines possible parameter adjustment operations, such as increasing / decreasing the QR code size, adjusting the font size, and changing the material type. The reward function evaluates the merits of parameter adjustments based on the label's performance, comprehensively considering factors such as recognition success rate, durability, cost, and aesthetics. The policy network learns the optimal parameter adjustment strategy, selecting the best adjustment action based on the current state. A Deep Q-Network (DQN) is used as the implementation of the policy network, combined with experience replay and target network techniques to improve learning stability. Simultaneously, a Dual Deep Q-Network (DDQN) architecture is introduced to reduce overly optimistic bias in value estimation and improve learning efficiency.

[0069] Among them, the Double Deep Q-Network (DDQN) is an advanced reinforcement learning architecture designed to address the overestimation of value problem in traditional DQN algorithms, playing a crucial role in the optical cable label parameter optimization task. DDQN significantly improves the stability and efficiency of learning by decoupling the action selection and value evaluation processes, making label parameter optimization more accurate and reliable.

[0070] The core principle of DDQN is based on value estimation bias analysis using Q-learning. In the standard DQN algorithm, the same network is responsible for both selecting actions (e.g., which label parameter adjustment scheme is optimal) and evaluating the value of that action (how much benefit the adjustment will bring). This design leads to the algorithm tending to select overestimated actions when estimation errors exist, forming a positive feedback loop and ultimately causing a serious overestimation problem. DDQN solves this problem by introducing two complementary networks: the current policy network is responsible for selecting the optimal action based on the current state, while the target network is responsible for evaluating the value of that action. This decoupled design effectively breaks the positive feedback loop and significantly reduces estimation bias.

[0071] The implementation of DDQN in the optical fiber label optimization system comprises four main stages. The first is network architecture design, where the system constructs two structurally identical but parameter-independent deep neural networks: the current network and the target network. Considering the complexity of the label parameter optimization task, the network typically employs a multilayer perceptron structure. The input layer receives the state vector (containing the current label configuration and environmental factors), the hidden layer uses 256-512 neurons with a ReLU activation function to process features, and the output layer corresponds to the Q-value of each possible parameter adjustment action. To handle continuous parameters (such as precise adjustment of QR code size), a hierarchical discretization strategy is introduced, discretizing the continuous action space into multiple levels of fine-tuning steps. The second stage is interaction and experience collection. Different label parameter configurations are tried in real or simulated environments, and after each adjustment, a four-tuple experience (current state, selected action, reward obtained, next state) is recorded. To accelerate learning, a priority experience replay technique is adopted, assigning different priorities based on the temporal difference error of the experience, allowing for more frequent learning from important experiences. Furthermore, an uncertainty-based exploration strategy is implemented, increasing the exploration probability in insufficiently explored regions of the parameter space to ensure comprehensive coverage of possible optimization directions. The third stage is dual Q-learning update. In each training batch, the current network is used to select the best action for the next state, but the target network is used to evaluate the value of that action. The final stage is policy evaluation and deployment. The current policy is evaluated regularly in a test environment, measuring key indicators such as label recognition rate and robustness under different conditions. To improve the robustness of the policy, a multi-scenario testing process was specifically designed, including extreme temperatures (-40°C to 85°C), high humidity (95% relative humidity), and strong light (100,000 lux). The fully trained DDQN model can recommend optimal label parameter configurations for different application scenarios, such as automatically selecting heat-resistant materials and increasing the QR code module size in high-temperature environments, and optimizing contrast and adding anti-glare coatings in strong light environments. Lyapunov index analysis is specifically introduced to evaluate the impact of parameter adjustments on label performance. The Lyapunov index is an indicator that describes the dynamic sensitivity to initial conditions and can quantify the impact of parameter changes on long-term behavior. By calculating the Lyapunov index for different parameter adjustment operations, key parameters that may lead to significant performance changes can be identified, and these parameters can be fine-tuned first. For example, in high-temperature environments, the Lyapunov index of QR code size and material type may be higher, indicating that these parameters have a significant impact on the performance of the label at high temperatures, and these parameters will be optimized first.

[0072] Based on a reinforcement learning model, adaptive optimization is performed on multiple key parameters. QR code parameter optimization includes encoding density (adjusted according to information content and spatial constraints, typically between 10-30mm), error correction level (from L to H, selected based on environmental interference levels), and UV resistance encoding capability (using special encoding schemes in outdoor environments). Text parameter optimization includes font size (from 5pt to 12pt, adjusted according to available space and viewing distance), thickness (increased thickness when environmental interference is high), and contrast (adjusted according to background color and lighting conditions). Material parameter optimization includes label thickness (from 0.1mm to 0.8mm, adjusted according to environmental conditions and durability requirements), adhesive type (selected according to surface material and ambient temperature), and protective layer treatment (such as UV protection, waterproof coating, scratch-resistant coating, etc.). The reinforcement learning model continuously adjusts parameter selection strategies based on successful and failed cases in historical data. For example, it analyzes labels that failed in high-temperature environments in the past to adjust material selection for similar environments, fine-tunes text and QR code parameters based on the output effects of different printing devices, and optimizes protective treatment schemes based on the aging speed of labels in different areas. This data-driven parameter optimization can adaptively generate optimal parameter configurations for different scenarios, significantly improving the practicality and durability of the tags.

[0073] Among them, generative adversarial network technology is used to optimize label design, including: Using a large number of historical label samples containing successful cases, failed cases, and expert annotations, as well as evaluations of their usage effects (high recognition rate, good durability, or difficult recognition and rapid damage) as training data, a generative adversarial network model is trained, which includes a generator responsible for generating label design schemes and a discriminator for evaluating whether the generated label designs meet the expected standards. Content constraints to ensure that key information is clearly visible, QR code constraints to automatically adjust the size and position of the QR code and the fault tolerance rate according to the usage environment, spatial constraints to optimize the overall layout according to the available space of the pasting position, and visual constraints to ensure the contrast between text and background to improve readability, a GAN model under constraints is formed. Based on the GAN model under the aforementioned constraints, special adjustments are made to different types of ODF tags that simplify information display and highlight port numbers and optical cable routes, device connection tags that emphasize information at both ends of the equipment and clearly indicate signal types, and tail cable tags that optimize information layout under small size to ensure recognizability. The resolution limitations of printing equipment, the characteristics of commonly used label materials with different ink absorption properties, possible deformation during the pasting process, and visual changes during the aging process are all taken into account to generate a variety of label design schemes. Based on the various label design schemes, through multiple rounds of optimization and iteration, the optimized label layout scheme is obtained by simultaneously considering functional requirements to meet label recognition and information transmission requirements, aesthetic enhancement to improve visual effects, and practicality to ensure ease of use.

[0074] Specifically, in the design process of optical fiber labels, traditional methods often rely on empirical rules and manual adjustments, making it difficult to adapt to diverse application scenarios and requirements. To address this issue, this invention innovatively introduces Generative Adversarial Network (GAN) technology for intelligent optimization of label design. GAN is a powerful deep learning architecture that, through adversarial training of the generator and discriminator, can generate high-quality, realistic samples. Applying GAN technology in the field of label design can significantly improve design efficiency and quality, enabling intelligent generation and optimization of labels.

[0075] First, a rich and diverse training dataset was constructed to provide the learning foundation for the GAN model. This dataset contains three key types of samples: successful cases, failed cases, and expert-annotated samples. Successful cases refer to label designs that perform well in real-world use, exhibiting high recognition rates (over 95% success rate under various environmental conditions) and good durability (lifespan exceeding 5 years under standard conditions). These cases typically feature clear information layout, appropriate contrast, and optimized QR code designs. Thousands of successful cases from different substations and application scenarios were collected, with detailed records of their design parameters (such as font size, color scheme, QR code size, etc.) and performance metrics (such as recognition rate, lifespan, etc.). Failed cases, on the other hand, are label designs that encountered problems in real-world applications, such as labels that were difficult to recognize (recognition success rate below 60%) or deteriorated rapidly (lifespan less than 50% of the expected lifespan). These cases usually have design flaws, such as insufficient contrast, inappropriate QR code size, or unsuitable material selection. These failed cases were collected, and their causes of failure and design flaws were analyzed to provide negative learning samples for the model. Expert-annotated samples are high-quality labeled samples designed and evaluated by industry experts. These samples not only include design parameters but also detailed expert comments and improvement suggestions. These samples are particularly valuable because they provide models with the experiential knowledge and design concepts of human experts.

[0076] Based on this rich training data, an advanced GAN model was constructed, comprising two core components: a generator and a discriminator. The generator employs a deep convolutional neural network architecture, consisting of multiple convolutional layers, batch normalization layers, and non-linear activation functions. Its inputs include conditional vectors (encoding label content, target scene, etc.) and random noise vectors (providing generation diversity), while the output is a complete label design scheme, including layout, color scheme, and font settings. The key innovation of the generator lies in its conditional control mechanism, which precisely controls the generation process by introducing multiple conditional vectors. These conditional vectors include scene conditional vectors (encoding physical space, environmental factors, such as "confined space," "outdoor environment," etc.), content conditional vectors (encoding fiber optic cable type, importance, such as "backbone fiber optic cable," "critical protection signal," etc.), and style conditional vectors (encoding design style preferences, such as "minimalist style," "high-contrast style," etc.). The discriminator also uses a deep convolutional neural network structure but incorporates an attention mechanism, enabling it to focus on key regions within the labels. The discriminator takes a label design sample (either a real sample or a sample generated by the generator) and a corresponding condition vector as input, and outputs a realism score for the sample and predictions of various performance aspects (such as recognition rate, robustness, etc.). The discriminator not only needs to distinguish between real and generated samples, but also needs to evaluate whether the generated samples meet various design criteria and performance requirements.

[0077] During the training of the GAN model, several innovative constraints were introduced to ensure that the generated label designs are both aesthetically pleasing and practical. Content constraints, ensuring that key information is clearly visible, are the most basic requirement for label design. Based on the information importance analysis in step S1.3, the label content is classified into three levels: primary information (such as fiber optic cable ID and connection endpoints, which must be the most prominent), secondary information (such as signal type and fiber optic cable type, which need to be clearly visible), and tertiary information (such as detailed descriptions and remarks, which can be appropriately simplified). According to this classification, different attention weights are set in the generator to ensure that important information receives more "design resources" (such as larger fonts and more prominent positions). QR code constraints automatically adjust the QR code size, position, and error tolerance according to the usage environment. In high-interference environments (such as outdoors or dusty areas), QR code sizes are increased (typically 25mm x 25mm or larger) and fault tolerance levels are improved (using Q or H levels, corresponding to 25% or 30% error correction capability). In controlled environments, more compact QR codes (such as 15mm x 15mm) can be used to save space, while the fault tolerance level is reduced (such as L level, corresponding to 7% error correction capability). Space constraints optimize the overall layout based on available space at the labeling location. Consideration is given to space limitations at different installation locations, such as fiber optic cable surfaces (typically long and narrow with limited width), near ports (typically small square areas), and patch panels (typically with standardized label slots). Based on these constraints, the label shape and size are dynamically adjusted to ensure the label fits the target installation location. Visual constraints ensure text-background contrast for improved readability. The optimal text-background color combination is automatically selected, and the color contrast ratio is calculated (according to the W3C Accessibility Guidelines, the text-background contrast ratio should be at least 4.5:1) to ensure good readability under various lighting conditions. At the same time, colorblind-friendly design was also considered, avoiding the use of easily confused color combinations such as red and green.

[0078] During the label design optimization process, special adjustments were made to different types of optical cable labels to meet their specific needs. For ODF (Optical Distribution Frame) labels, a simplified information display strategy was adopted because they typically need to provide port identification information within a confined space. These labels highlight the port number (usually using large, bold font, occupying more than 30% of the label area) and the optical cable route (using clear arrows or text descriptions to indicate the start and end points of the optical cable), while other information (such as detailed equipment descriptions) was simplified or reduced in size. Simultaneously, considering that labels are often densely packed in ODF environments, the side-view readability of the labels was specifically optimized to ensure that key information can be clearly identified from different angles. For inter-device connection labels, because they need to clearly indicate the relationship between the two ends of the devices, the information of both ends of the devices is emphasized (equally highlighting the names of the starting and ending devices) and the signal type (using color coding or icons to represent different types of signals, such as GOOSE, SMV, etc.). These labels typically use a symmetrical layout, displaying the two connected devices on either side of the label, with the signal type and other connection details displayed in the middle. To improve information density while maintaining readability, layered layouts and color coding are cleverly utilized to display more information within a limited space. For pigtail labels, which are typically small in size and need to be affixed to thin optical cables, the information layout for such small sizes has been specifically optimized. These labels employ efficient space utilization strategies, such as vertically arranging text (extending along the direction of the optical cable), using abbreviations and codes (e.g., using "SM" to represent single-mode fiber), and optimizing font selection (choosing fonts that remain clear even at small sizes). Simultaneously, the recognizability of the pigtail label's QR code has been enhanced by optimizing its size, position, and contrast to ensure reliable identification even with limited label size.

[0079] When generating label design schemes, various limiting factors in practical applications were comprehensively considered. Firstly, there are the resolution limitations of the printing equipment. Different printing devices have different resolution capabilities, ranging from low-resolution thermal printers (e.g., 300 dpi) to high-resolution laser printers (e.g., 1200 dpi). Based on the resolution capabilities of the target printing device, the text size (ensuring the smallest font is still legible at the target resolution, typically no less than 6pt), line thickness (ensuring fine lines do not break after printing), and QR code density (ensuring each module contains at least 3×3 printing dots) are adjusted. Secondly, the characteristics of the label material are considered. Different materials have different ink absorption, durability, and suitable environments. For example, ordinary paper labels are low-cost but have poor durability, suitable for indoor protected environments; polyester (PET) labels are water-resistant and abrasion-resistant but more expensive, suitable for outdoor or harsh environments; polyvinyl chloride (PVC) labels have good flexibility and are suitable for application to curved surfaces. The process involves selecting appropriate materials based on the target usage environment and durability requirements, and adjusting design parameters accordingly. For example, increasing line thickness on highly absorbent materials and adding background color on transparent materials. Thirdly, potential deformation during the application process is considered, especially when labels need to be applied to curved surfaces (such as fiber optic cables) or irregular surfaces. The robustness of the design is evaluated by simulating different degrees of bending and stretching (typically considering a deformation range of ±10%), ensuring that key information remains identifiable even after deformation. Finally, visual changes during aging are also considered, such as color fading (simulated by reducing color saturation), contrast reduction (simulated by reducing brightness and darkness), and wear (simulated by adding noise and blurring effects). These considerations enable the generation of multiple label design schemes that perform well in practical applications.

[0080] Finally, based on the generated multiple label design schemes, through multiple rounds of optimization and iteration, balancing functional requirements, aesthetics, and practicality, the final optimized label layout scheme was obtained. This process considered three key aspects simultaneously: functional requirements, aesthetics, and practicality. Functional requirements are the most basic consideration; the labels must meet the requirements of identification and information delivery. The information completeness (whether all necessary information is included), readability (whether the text is clearly visible), and recognition rate (the success rate of QR code recognition under various conditions) of each design scheme were evaluated. While aesthetics are not the primary consideration, a good visual effect can improve user experience and work efficiency. The overall balance, color coordination, and professional appearance of the design were evaluated to ensure that the labels are not only functionally complete but also visually appealing. Practicality focuses on the ease of use of the labels in actual use, including ease of installation (whether they are easy to paste accurately), search efficiency (whether the target label is easy to find among multiple labels), and durability (whether they can maintain their functionality over a long period in the expected environment).

[0081] The optimization process employs an iterative approach. After each iteration, design parameters are adjusted based on simulation test results and historical data comparisons. For example, if a design performs poorly in recognition rate testing, the QR code size is increased or contrast is improved; if a design scores low in ease-of-use assessments, the layout is adjusted to enhance intuitiveness. Through 5-10 rounds of optimization iterations, the optimal design solution balancing various requirements is found. The final optimized label layout not only meets various technical requirements but also considers aesthetics and practicality, providing superior performance and user experience in real-world applications. Practical application results show that this GAN-based label design optimization method improves label recognition success rate by 15-20%, extends lifespan by over 30%, significantly enhances user satisfaction, and greatly reduces the difficulty and maintenance costs of substation fiber optic cable management.

[0082] In step S3, the data is converted into structured tag identification data containing optical cable type, connection relationship, and transmission signal type, including: S3.1: Based on the aforementioned optical cable smart tag, the tag image collected by the mobile terminal is combined with supplementary environmental data obtained in areas with tag damage, obstructed view, and high-density optical cable. Simultaneously, the light intensity, collection angle, and collection distance are recorded. Adaptive histogram equalization, nonlocal mean filtering, and super-resolution reconstruction image enhancement algorithms are applied. Geometric correction algorithms and illumination compensation algorithms are applied to generate standardized tag images with uniform brightness, contrast, viewing angle, and resolution. S3.2: Based on the standardized label image, the QR code region is located and extracted through gradient analysis, morphological verification, and localization point detection. A super-resolution reconstruction technique combining residual learning network and perceptual loss function is used to process low-resolution or blurry QR codes. Edge repair techniques that preserve structural integrity, context-aware filling, and topology are used to process partially worn or broken QR codes. An attention mechanism that combines spatial attention to focus on the effective region, channel attention to dynamically adjust feature weights, and multi-scale fusion to capture local and global features is used. QR code recognition is achieved through adaptive binarization, perspective correction, and error correction enhancement decoding to obtain the QR code recognition result. S3.3: Based on the QR code recognition results, a multi-objective evolutionary algorithm with decision space diversity is introduced, which encodes key parameters and strategies of the recognition process into decision vectors. Based on the uniformity measure of Hamming distance, a diverse initial solution set is generated. Through a two-layer selection based on Pareto advantage and decision space diversity, multiple complementary recognition strategies are obtained. At the same time, optical character recognition is performed on the text information on the label. Multimodal fusion technology, which integrates information consistency verification, complementary information integration, and redundant information utilization, is used to fuse the QR code information with the text recognition results. The results are then converted into a standardized data structure through natural language processing technology to obtain the structured label recognition data containing the optical cable type, connection relationship, and transmission signal type.

[0083] Specifically, step S3.1 details the multimodal data acquisition and preprocessing process. In the identification of optical fiber tags, high-quality raw data must first be acquired and effectively preprocessed to provide good input for subsequent identification algorithms. This sub-step achieves efficient acquisition and enhancement of tag data through various sensing devices and preprocessing technologies. First, tag images are acquired using a high-definition camera mounted on a mobile terminal (such as an industrial tablet or smartphone). These cameras possess several advanced features, including high resolution (at least 12 megapixels, supporting 4K image acquisition), autofocus (ensuring clear imaging at various distances), optical image stabilization (reducing blur caused by hand shake), and supplemental lighting (LED fill light for low-light environments). During the acquisition process, exposure parameters, white balance, and color saturation are automatically adjusted to adapt to different lighting conditions, ensuring image quality.

[0084] In addition to image data, rich supplementary environmental data is collected to improve the accuracy and robustness of identification. This data includes light intensity (measured by a light sensor in lux), acquisition angle (measured by a gyroscope and accelerometer, recording horizontal and vertical offset angles), acquisition distance (estimated by laser rangefinder or depth camera in centimeters), and ambient temperature and humidity (collected by a temperature and humidity sensor, affecting image quality and QR code reflectivity). Furthermore, an RFID-assisted identification module is integrated, specifically for assisting identification in areas with damaged tags, obstructed vision, or high-density fiber optic cables. RFID technology provides a vision-independent tag identification method that can penetrate minor obstacles, offering an alternative identification approach when visual identification is difficult. The RFID module operates in the 13.56MHz band, with a reading distance of up to 10 centimeters. An RFID chip with anti-metal interference design is embedded within the tag to ensure stable reading in metallic environments.

[0085] After acquiring the raw data, a series of preprocessing steps are performed to convert the raw images acquired under various conditions into standardized, high-quality label images. First, image enhancement includes adaptive histogram equalization (enhancing local contrast, especially in low-light and overexposed environments), nonlocal mean filtering (preserving details while removing noise), and super-resolution reconstruction (improving image resolution through deep learning models, particularly for small labels acquired at long distances). Next, geometric correction includes perspective transformation (correcting distortion caused by tilt angles), distortion correction (eliminating barrel or pincushion distortion caused by camera lenses), and size normalization (adjusting label images of different sizes to a uniform size for easier subsequent processing). Then, illumination compensation is performed, including illumination model estimation (analyzing the illumination distribution in the image), reflectance component separation (distinguishing between diffuse and specular reflections on the label surface), and shadow removal (eliminating contrast unevenness caused by some shadows). Finally, region localization includes label detection (locating label regions in complex backgrounds), region of interest extraction (separating QR code and text regions), and multi-label separation (handling multiple labels that may be contained in the same image). These preprocessing steps enable the extraction of high-quality standardized label images from raw images acquired under various non-ideal conditions, providing ideal input data for subsequent QR code recognition and text recognition.

[0086] Step S3.2 implements super-resolution enhancement and repair of the QR code region of the optical cable label. In practical applications, the QR code region may suffer from poor image quality due to various reasons (such as distance, poor lighting, partial wear, etc.), affecting the recognition accuracy. This sub-step uses advanced computer vision technology to enhance and repair the QR code region, significantly improving the recognition capability under various conditions. First, based on the standardized image preprocessed in step S3.1, the QR code region is located and extracted. Localization uses a cascaded detector, combining Haar features and the AdaBoost algorithm, which can quickly and accurately identify the QR code region in the image. For complex backgrounds or partial occlusion, a deep learning object detection model (such as Faster R-CNN or YOLO) is introduced to improve the robustness of localization. The extracted QR code region undergoes precise cropping and corner correction to ensure the correct alignment of the QR code matrix.

[0087] For low-resolution or blurry QR codes, a super-resolution reconstruction technique combining residual learning networks and a perceptual loss function is applied. This technique, based on a deep residual network (ResNet) architecture, achieves high-quality super-resolution reconstruction by learning the residual mapping between low-resolution and high-resolution images. Unlike traditional methods that only use pixel-level MSE loss, this technique employs a perceptual loss function, combining pixel-level loss, feature-level loss (deep feature differences extracted through a pre-trained VGG network), and adversarial loss (evaluating the realism of the reconstructed image through a discriminator network), to optimize visual quality and detail preservation. A special training dataset and network structure are customized for the unique structure of QR codes (a regular arrangement of black and white modules), including data augmentation strategies (simulating various degradation scenarios) and a structure-aware residual block design (preserving the geometric structure of the QR code). This super-resolution reconstruction can improve the original resolution by 2-4 times, significantly enhancing the clarity and recognizability of QR codes.

[0088] Among them, the Deep Residual Network (ResNet) is a deep learning architecture that performs exceptionally well in the super-resolution reconstruction of fiber optic cable label QR codes. It solves the gradient vanishing problem in deep neural network training through an innovative residual learning paradigm, enabling the construction of ultra-deep network structures that capture multi-scale features in images. It is particularly suitable for processing images like QR codes that require accurate recovery of high-frequency details.

[0089] The core principle of ResNet is based on residual learning. Traditional convolutional neural networks directly learn the input-to-output mapping function H(x), while ResNet learns the residual mapping F(x) = H(x) - x, meaning the network's actual output is F(x) + x. This design uses identity shortcut connections to directly add the input to the output, forming a residual block structure. This mechanism is particularly important in QR code super-resolution reconstruction because it allows the network to focus on learning the differences between low-resolution and high-resolution images, rather than completely reconstructing the entire image. This significantly improves learning efficiency and accuracy, especially in recovering key recognition features such as QR code edges and corners.

[0090] The implementation of ResNet in the fiber optic cable label QR code super-resolution system comprises four main stages. The first stage involves network architecture design, constructing a super-resolution residual network (SR-ResNet) specifically optimized for QR codes. The input layer receives low-resolution QR code images, passes through an initial feature extraction convolutional layer, and then enters the residual learning body, consisting of 16-20 concatenated residual blocks. Each residual block contains two 3×3 convolutional layers with a ReLU activation function in between, and skip connections at the end of the block directly add the input to the output. To adapt to the structural characteristics of QR codes, an attention mechanism module is introduced on top of the standard residual block to help the network focus on key areas such as the QR code's positioning pattern and arrangement. After feature extraction, resolution improvements of 2×, 3×, or 4× are achieved through pixel-shuffle layers or transposed convolutions, and finally, a reconstruction convolutional layer generates a high-resolution output. The second stage involves the construction and enhancement of a specific dataset. For the fiber optic cable label application scenario, a large number of high-quality QR codes are collected as high-resolution samples, and corresponding low-resolution samples are generated by controlling the degradation process. The degradation process simulates various quality degrades that actual optical fiber tags may encounter under different conditions, including Gaussian blur (simulating blurry focus), motion blur (simulating jitter during scanning), JPEG compression (simulating digital transmission loss), and noise addition (simulating low-light conditions). Simultaneously, specific data augmentation strategies were designed, such as random occlusion (simulating partial tag wear), brightness and contrast variations (simulating different lighting conditions), and geometric transformations (simulating different viewing angles). These data augmentation techniques significantly improved the model's generalization ability, enabling it to adapt to varying scanning conditions on-site. The third stage involved the design and training of a multi-objective loss function. To comprehensively optimize the visual quality and machine readability of the QR code, a triple loss function combination was developed. The pixel-level loss uses the L1 norm, which better preserves the sharp edges of the QR code compared to L2; the feature-level loss extracts deep features through a pre-trained VGG-19 network, calculating the difference between the reconstructed image and the original high-resolution image in the feature space, paying particular attention to the structural consistency of the QR code; the adversarial loss introduces a discriminator network to evaluate the realism of the reconstructed image, prompting the generator network to produce an output closer to the real high-resolution image. Furthermore, considering the unique properties of QR codes, a structure-preserving loss term was designed to ensure accurate reconstruction of key information such as positioning patterns and arrangement patterns. Training employed the Adam optimizer with an initial learning rate of 1e-4 and a cosine annealing strategy. The batch size was 16, and training was performed on a server with four NVIDIA V100 GPUs, achieving convergence in an average of 48 hours. The final stage involved model optimization and deployment. To adapt to the computational resource limitations of edge devices, the trained model underwent compression and optimization, including knowledge distillation (transferring knowledge from a large teacher network to a smaller student network), pruning (removing unimportant connections), and quantization (converting 32-bit floating-point parameters to 8-bit integers).The optimized model is 78% smaller and 3.5 times faster inference, enabling real-time operation on mobile devices such as smartphones (processing time less than 150ms per frame). In actual testing, it improved the recognition rate of severely degraded QR codes from 37% to over 94%, maintaining a success rate of over 80% even under extreme conditions (such as partial wear on fiber optic labels, severe contamination, or insufficient light), significantly improving the reliability and user experience of the fiber optic management system. This ResNet-based super-resolution reconstruction technology not only overcomes the limitations of traditional methods in processing low-quality QR codes but also provides powerful image restoration capabilities for fiber optic labeling systems, ensuring stable operation in various harsh environments and becoming a key technological pillar of modern fiber optic management systems.

[0091] For partially worn or broken QR codes, edge repair techniques including structural integrity analysis, context-aware filling, and topology preservation are employed. First, structural integrity analysis assesses the extent and location of damage, including missing module detection (identifying completely missing black-and-white modules), edge damage detection (identifying blurred or broken module edges), and contaminated area identification (identifying areas covered by dirt). Then, context-aware filling is performed based on the QR code's structural characteristics and existing information. This includes Markov random field (MRF)-based module state inference (inferring the possible state of the missing module based on the surrounding module states), constraint repair based on QR code generation rules (using the QR code's check digits and format information for repair), and region completion based on a deep generative model (using GAN or VAE models to generate context-appropriate filling content). Finally, topology preservation techniques ensure the repaired QR code maintains correct geometry and connectivity, including edge enhancement (improving the clarity of module boundaries), shape regularization (ensuring each module is approximately square), and global structure alignment (ensuring the regular arrangement of the overall QR code matrix).

[0092] To further improve recognition capabilities under extreme conditions, a multi-objective evolutionary algorithm with diverse decision spaces is innovatively introduced. This algorithm encodes key parameters and strategies in the recognition process into decision vectors, including preprocessing parameters (such as filter strength and contrast enhancement), reconstruction parameters (such as super-resolution factor and regularization strength), and decoding parameters (such as binarization threshold and fault-tolerance strategy). Then, the diversity of the decision space is evaluated using a Hamming distance-based uniformity metric to ensure that the generated candidate solutions are uniformly distributed in the parameter space, covering different recognition strategies. Multiple objectives are optimized simultaneously, including recognition accuracy, computational efficiency, and robustness, by using the Non-Dominated Sorting Genetic Algorithm II (NSGA-II) to find the Pareto optimal solution set. In practical applications, the algorithm dynamically selects or combines the most suitable recognition methods based on real-time detected environmental conditions (such as light intensity and image quality). In this way, a high recognition success rate can be maintained under various extreme conditions (such as severe soiling, strong light interference, and long-distance shooting), significantly improving adaptability and reliability.

[0093] Step S3.3 achieves structured extraction of label information through multimodal fusion technology. After recognizing the QR code, the text information on the label needs to be recognized, and the QR code and text information are fused to form complete structured label data. This sub-step achieves high-precision extraction and integration of label information through advanced OCR technology and multimodal fusion methods. First, optical character recognition (OCR) is performed on the text area on the label. A customized OCR model for the power industry is used. This model has been trained on a large number of substation optical cable label texts and can accurately recognize power industry terminology and special formats. The OCR processing flow includes text region detection (locating text regions using algorithms such as MSER or TextBoxes++), character segmentation (segmenting continuous text into individual characters, considering different fonts and spacing), feature extraction (using CNN to extract deep features of characters), and character recognition (determining the identity of each character through a deep learning classifier). To improve recognition accuracy, a context-aware language model is introduced, utilizing language rules and common expressions in the power industry to correct possible recognition errors. For example, when the word "blocker" is identified, the language model will infer from the context that the correct word should be "blocker".

[0094] After separately recognizing QR codes and text, multimodal fusion technology integrates this information into unified structured data. Multimodal fusion employs a hierarchical strategy, including feature-level fusion (merging low-level features from different modalities), decision-level fusion (integrating recognition results from different modalities), and semantic-level fusion (integrating semantic information from different sources). Feature-level fusion assigns different weights to features from different modalities through an attention mechanism, dynamically adjusting the importance of each modality based on current scene conditions. For example, if the QR code is clear but the text is blurry, the weight of the QR code information is increased; conversely, if the text is clear but the QR code is partially damaged, the weight of the text information is increased. Decision-level fusion uses weighted voting or a Bayesian decision framework, comprehensively considering the confidence level of each modality's recognition results and selecting the most reliable result. For conflicting recognition results, conflicts are resolved based on prior knowledge (such as common error patterns) and current conditions (such as the image quality of each modality). Semantic-level fusion focuses on the semantic consistency and completeness of information, ensuring that the fused data is logically sound and informationally complete.

[0095] To improve the fusion effect, information complementarity and cross-validation mechanisms are introduced. The information complementarity mechanism identifies the unique information provided by different modalities, ensuring that the fusion result includes all available information. For example, a QR code may contain detailed unique identifiers and numerical codes, while a text label may provide more intuitive device names and connection information. The cross-validation mechanism utilizes redundant information between different modalities for mutual verification, improving data reliability. For example, it compares the device ID in the QR code with the device name mentioned in the text label to check for consistency. Through this multimodal fusion, the advantages of various information sources can be fully utilized to maximize the accuracy and completeness of label recognition. Finally, the fused recognition result is transformed into structured label recognition data containing fiber optic cable type (e.g., trunk cable, pigtail cable, single-mode / multimode fiber), connection relationship (signal flow direction and logical relationship between devices), and transmission signal type (e.g., GOOSE signal, SV sample value, MMS message). This structured data directly supports subsequent data comparison and anomaly detection, providing a solid data foundation for intelligent management of fiber optic cable tags.

[0096] The process of acquiring tag images of optical cables based on the smart tags, and transforming them into structured tag recognition data containing optical cable type, connection relationship, and transmission signal type using multimodal fusion computer vision technology, also includes enhancing the diversity of the decision space using a uniformity metric based on Hamming distance. The key parameters and strategies in the identification process are encoded into decision vectors, the Hamming distance between each decision vector in the solution set is calculated, and a density-based diversity evaluation function is designed to use the diversity index as an additional optimization objective. Based on the diversity evaluation function, an initial solution set with diversity is generated. New candidate solutions are generated through mutation and crossover to explore the decision space. A two-layer selection is carried out based on Pareto advantage and then based on the diversity of the decision space. The diversity weight is dynamically adjusted according to the convergence of the optimization process, the learning rate is adjusted according to the sensitivity of different parameters, and an adaptive adjustment mechanism is used to dynamically balance exploration and utilization to obtain a variety of complementary recognition strategies. Based on the aforementioned complementary recognition strategies, a strategy that focuses more on illumination compensation is selected in strong light interference environments, a more aggressive image restoration strategy is selected in severe damage situations, and a simplified strategy with higher computational efficiency is selected when computational resources are limited. The recognition method most suitable for the current situation is dynamically selected or combined according to the environmental conditions detected in real time, thereby enhancing the diversity of the decision space.

[0097] Specifically, in smart substations, the identification of fiber optic tags faces various complex challenges, such as harsh environmental conditions, tag aging and damage, and poor imaging quality. Traditional single identification methods often struggle to adapt to these changing scenarios. To address this issue, this invention innovatively introduces a uniformity metric method based on Hamming distance, enhancing the diversity of the decision space and constructing a robust identification system capable of adapting to various complex conditions. This method not only improves the success rate of identification but also intelligently selects the optimal strategy based on different environmental conditions, significantly enhancing the overall system performance.

[0098] First, the various key parameters and strategies in the recognition process are encoded into structured decision vectors, laying the foundation for diversity assessment and optimization. These decision vectors cover several key aspects of the recognition process. Preprocessing parameters include filtering method selection (such as Gaussian filtering, median filtering, bilateral filtering, etc.), filtering strength (usually adjusted between 0.5 and 5, with larger values ​​resulting in stronger filtering effects), contrast enhancement methods (such as histogram equalization, CLAHE adaptive histogram equalization, Gamma correction, etc.), contrast enhancement degree (usually expressed as a percentage, such as 30% enhancement), and edge enhancement parameters (such as sharpening coefficient, usually between 0.1 and 2). Reconstruction parameters include super-resolution factor (determining the magnification factor, usually 2×, 3×, or 4×), super-resolution algorithm selection (such as SRCNN based on convolutional neural networks, SRGAN based on generative adversarial networks, etc.), regularization strength (controlling the smoothness in the reconstruction process, usually 0.001-0.1), number of iterations (determining the complexity of the reconstruction process, usually 10-100 iterations), and feature preservation preferences (such as edge preservation preferences, texture preservation preferences, etc.). Decoding parameters include binarization methods (such as global thresholding, adaptive thresholding, Otsu's method, etc.), binarization thresholds (determining the black-and-white boundary point, typically an integer between 0 and 255), fault tolerance strategies (such as error correction code usage strategies, damaged area repair strategies, etc.), pattern recognition sensitivity (determining the strictness of the recognition module), and check bit usage strategies (such as enabling all check bits, using only critical check bits, etc.). These parameters are encoded into binary decision vectors, where each bit represents a parameter option or value range. For example, 8 bits can represent an integer value between 0 and 255, and 2 bits can represent four different algorithm choices.

[0099] To assess the diversity of the decision space, an innovative diversity evaluation function was designed based on Hamming distance. Hamming distance is an indicator that measures the difference between two strings of equal length, defined as the number of different characters at corresponding positions in the two strings. The larger the Hamming distance, the greater the difference between the two decision vectors, indicating that they adopt more different strategies. The designed diversity evaluation function comprehensively considers the distribution of Hamming distances among all decision vectors in the entire solution set and calculates the "crowding" of each solution in the decision space.

[0100] Based on the aforementioned diversity evaluation function, a complete multi-objective evolutionary algorithm was implemented to generate and optimize a diverse set of recognition strategies. First, an initial solution set with high initial diversity was constructed by controlling the parameter distribution during the generation process. This initialization did not employ completely random generation but instead used techniques such as Latin Hypercube Sampling (LHS) to ensure a uniform distribution of initial solutions in the decision space. The initial solution set typically contains 100-200 different decision vectors, covering different parameter combinations and strategy choices. Next, the system generates new candidate solutions through mutation and crossover operations, exploring more possibilities in the decision space. Mutation operations include bit flipping (randomly changing certain bits in the decision vector, with a probability typically of 0.01-0.05), parameter adjustment (targeted adjustments based on parameter sensitivity), and strategy replacement (replacing the strategy of a module entirely, such as changing the preprocessing method). The crossover operation generates offspring solutions with hybrid characteristics by combining the advantages of different parent solutions. Methods such as gene fragment exchange (exchanging parameter fragments between two decision vectors), uniform crossover (independently deciding whether to exchange each bit according to a certain probability), and intelligent crossover (selecting exchange parameters in a targeted manner based on the performance of the parent solution in different scenarios) are adopted.

[0101] Latin Hypercube Sampling (LHS) is an advanced statistical sampling technique used in fiber optic tag identification systems to generate high-quality initial policy sets. It enables uniformly distributed sample generation in a high-dimensional parameter space, overcoming the shortcomings of traditional random sampling methods that easily produce sample clusters or blank areas. This provides evolutionary algorithms with more comprehensive starting points, thereby accelerating convergence and improving the quality of the final solution.

[0102] The core principle of LHS (Latin Square Hierarchy) is based on the mathematical concept of a Latin square matrix. It achieves efficient coverage of multidimensional space by ensuring a uniform projection distribution across each dimension. When generating m sample points in an n-dimensional parameter space, LHS equally divides each dimension into m intervals, ensuring that there is only one sample point in each interval of each dimension. This design guarantees a uniform distribution of samples throughout the parameter space while avoiding the curse of dimensionality, enabling good spatial coverage even in high-dimensional spaces with a relatively small number of samples. For fiber optic tag recognition systems, this means exploring various parameter combinations (such as image processing parameters, feature extraction thresholds, classifier configurations, etc.) with a minimal initial strategy, significantly improving search efficiency.

[0103] The implementation of LHS (Local Hierarchical Structure) in a fiber optic tag identification system comprises five main stages. The first stage involves parameter space definition and boundary setting, comprehensively analyzing key parameters in the identification process, including preprocessing parameters (Gaussian filter kernel size, adaptive threshold window size, morphological operation structuring element shape, etc.), feature extraction parameters (HOG descriptor unit size, SIFT feature point threshold, CNN feature layer selection, etc.), and classification decision parameters (SVM kernel function type, number of random forest trees, deep network activation function selection, etc.). Each parameter is given reasonable upper and lower bounds based on professional knowledge, forming a parameter hyperspace with 25-30 dimensions, providing an accurate search range for LHS. The second stage is interval partitioning and initialization. It is decided to generate 160 initial solutions (experience shows that this number provides a good balance between diversity and computational efficiency at this problem complexity), and then the value range of each parameter is equally divided into 160 intervals. For continuous parameters (such as filter σ value), numerical division is performed directly; for discrete parameters (such as CNN layer selection), they are mapped to a continuous space before partitioning, sampled, and then mapped back to discrete values. To improve sampling quality, the system also implements a minimum distance constraint to ensure that the Euclidean distance between any two sample points in the normalized space is not less than a preset threshold (usually 0.05), avoiding wasted computational resources by sample points that are too close. The third stage is the generation and optimization of the sampling matrix. First, a 160×30 (number of samples × number of dimensions) sampling matrix is ​​created, with each column representing a parameter dimension. By randomly arranging a sequence from 1 to 160 and normalizing it to the [0,1] interval, the sampling point positions on that dimension are generated. This process ensures that there is exactly one sampling point in each interval of each parameter dimension. To further optimize the sample distribution, the system implements maximum-minimum distance optimization: through column swapping operations, the minimum distance between sample points is repeatedly optimized to maximize the spatial dispersion of the entire sample set. This optimization process usually performs 500-1000 iterations until the minimum distance no longer increases significantly or the computation time limit is reached. The fourth stage is parameter transformation and strategy construction. The optimized normalized sampling matrix is ​​converted back to actual parameter values: for continuous parameters, linear interpolation is used to map them to the actual range; for discrete parameters, nearest neighbor rounding is used to determine the final value. Specifically, for cases where certain parameter combinations have dependencies (e.g., some feature extraction methods are only applicable to images with specific preprocessing), the system implements conditional constraints to ensure that the generated parameter combinations are executable in the actual system. The converted parameter vector is directly used to construct a complete recognition strategy, including the configuration of the entire process from image acquisition, preprocessing, feature extraction, classification decision, and post-processing. The final stage is diversity verification and adjustment.The sampling effect is verified by calculating diversity evaluation metrics for the initial solution set: coverage metrics measure the coverage ratio of the parameter space, typically requiring above 95%; dispersion metrics measure the spatial uniformity of sample points, assessing the statistical distribution of nearest neighbor distances; diversity metrics comprehensively consider the distribution of strategies in the behavior space, calculated through performance differences on the test dataset. If the diversity metrics do not meet expectations (typically requiring behavior space coverage > 85%), local adjustments are made: identifying low-coverage areas in the parameter space and increasing targeted sampling points; or removing overly similar strategies and replacing them with new strategies located in low-coverage areas. In the selection phase, an innovative two-layer selection strategy is implemented. The first layer selects based on Pareto advantage, retaining non-dominated solutions (i.e., solutions not dominated by other solutions on any objective). A Fast Non-dominated Sorting algorithm is used to divide the solution set into multiple levels, prioritizing solutions with higher levels. Solutions in the Pareto Front represent the optimal trade-off between multiple objectives such as recognition accuracy, computational efficiency, and robustness. The second layer selects solutions based on the diversity of the decision space, prioritizing solutions that are sparsely distributed in the decision space, i.e., solutions that differ significantly from others. The crowding degree of each solution is calculated, and solutions with low crowding degree (i.e., high diversity) are selected first. This two-layer selection ensures that the solution set contains both high-performance solutions and maintains sufficient diversity, preventing premature convergence to a single strategy. To further improve the algorithm's adaptability, several adaptive adjustment mechanisms are introduced. The diversity weight is dynamically adjusted based on the convergence progress during optimization, increasing it in the early stages to promote broad exploration and decreasing it in the later stages to promote fine-grained optimization. The learning rate is adjusted according to the sensitivity of different parameters, using a smaller learning rate for critical parameters with significant impact (such as the binarization threshold and super-resolution factor) for fine-tuning, and a larger learning rate for parameters with less impact to accelerate convergence. A dynamic balance is maintained between exploration and exploitation, adaptively adjusting the mutation rate and crossover rate based on the current performance and diversity state of the solution set, increasing exploration intensity when performance improvement is slow and increasing exploitation intensity when potential regions are discovered. These strategies enable the acquisition of a diverse and high-performance set of complementary identification strategies, providing a rich selection for subsequent dynamic choices.

[0104] Among them, the Fast Non-dominated Sorting algorithm is a core algorithm in the field of multi-objective optimization, and it is used in the optical fiber tag identification system for efficient evaluation and hierarchical strategy sets. By systematically comparing the dominance relationships between solutions, all candidate solutions are organized into different dominance levels, providing clear selection guidance for evolutionary algorithms and balancing the trade-offs between multiple conflicting objectives (such as recognition accuracy, computational complexity, and environmental adaptability).

[0105] The core principle of this algorithm is based on the concept of Pareto dominance. For multi-objective optimization problems, if solution A is superior to solution B on at least one objective and not inferior to B on all other objectives, then A is said to dominate B. Non-dominated solutions are solutions that are not dominated by any other solutions in the set; these solutions form the Pareto front, representing the optimal trade-offs among the objectives. The fast non-dominated sorting algorithm efficiently divides the entire solution set into multiple front levels by calculating two key attributes of each solution: dominance count (how many solutions dominate it) and dominance set (which solutions it dominates). The first-level front contains all non-dominated solutions, the second-level front contains the non-dominated solutions after removing the first-level front, and so on, forming a complete hierarchical structure.

[0106] The implementation of fast non-dominated sorting in the optical fiber tag identification system comprises three main stages. The first stage is dominance relation calculation, which compares each pair of solutions in the candidate solution set and evaluates their performance on three key objectives: recognition accuracy (the proportion of correct recognitions on a standard test set), computational efficiency (the average time required to complete the identification), and environmental adaptability (a measure of performance stability under different conditions). For each solution p, two sets are maintained: Sp (the set of solutions dominated by p) and np (the number of solutions dominating p). The algorithm's time complexity is O(MN). 2 The first stage is front allocation, where M is the number of targets (usually 3-5) and N is the number of solutions (usually 100-200). The second stage is front allocation, which stratifies the solution set according to the calculated dominance relationships. First, the first front F1 is identified, containing all solutions with np=0 (i.e., solutions not dominated by any solution). Then, for each solution p in F1, the system iterates through each solution q in its dominance set Sp, decrementing the dominance count of q by 1; when the dominance count of a solution q becomes 0, it is added to the next front F2. This process is repeated until all solutions are assigned to their respective fronts. In fiber optic tagging systems, typically 4-6 front levels are formed, with the first front containing 20-30 of the best-performing identification strategies, representing the performance boundary achievable by the system. The third stage is rank allocation and selection, assigning each solution a rank value based on its front level (a smaller rank value indicates higher solution quality). In the selection operation, solutions with smaller rank values ​​are preferentially selected for the next generation. If it is necessary to select partial solutions from the same frontier (e.g., when the size of the current frontier exceeds the number of options), then crowding distance calculation is further applied to prioritize solutions with lower crowding (i.e., more isolated in the target space) to maintain the diversity of the solution set. In practice, all solutions from the first frontier are retained, and crowding selection is applied starting from the second frontier until a preset population size is reached (usually 50-70% of the original population).

[0107] Based on the aforementioned optimized complementary recognition strategies, an environment-adaptive dynamic strategy selection is achieved, significantly improving the robustness and accuracy of fiber optic label recognition. It can intelligently select or combine the most suitable recognition strategies according to different environmental conditions. In environments with strong light interference, when bright areas, strong light spots, or obvious brightness unevenness are detected in the image, a strategy that prioritizes illumination compensation is automatically selected. These strategies typically include high dynamic range processing (merging images from multiple exposures to capture more details), illumination equalization (reducing the impact of overly bright and dark areas), and specular reflection removal (identifying and eliminating specular reflections on the label surface). For example, when more than 10% of the image is detected as overexposed, the system will activate a dedicated specular suppression algorithm to reduce the brightness of these areas while improving the visibility of shadow areas, making the contrast between the black and white modules of the QR code clearer. In cases of severe damage, when the system detects obvious stains, scratches, or missing areas on the label, a more aggressive image restoration strategy will be selected. These strategies include high-intensity denoising (such as nonlocal mean filtering, which effectively removes speckle noise), edge enhancement reconstruction (reconstructing blurred or broken module boundaries), and deep learning-based region completion (using generative models to fill in missing regions). For example, when more than 20% of a QR code is detected to be damaged, a dedicated deep learning repair model is activated. This model, trained on a large number of damaged QR codes, can effectively recover occluded or damaged information. When computational resources are limited, such as when running on handheld devices or needing to process a large number of tags in real time, the system will choose more computationally efficient simplification strategies. These strategies typically include lightweight preprocessing (such as simple Gaussian filtering instead of complex nonlocal mean filtering), low-complexity super-resolution (such as fast bicubic interpolation instead of deep learning super-resolution), and optimized decoding algorithms (such as simplified binarization methods and fast pattern recognition). While these simplification strategies may have slightly lower performance under extreme conditions, they provide good performance in most common situations while significantly reducing computation time and resource consumption.

[0108] The core innovation lies in its ability to detect environmental conditions in real time and dynamically select or combine the most suitable recognition methods. The system uses multiple sensors and image analysis technologies to assess current environmental conditions in real time, including light intensity analysis (assessing lighting conditions through image histograms and brightness distribution), dirt detection (assessing label cleanliness through texture analysis and edge detection), and resource monitoring (monitoring current device CPU utilization, memory usage, and battery status). Based on this real-time data, a combination of rule-based decision trees and machine learning is used to select the most suitable strategy for the current situation. In some complex cases, multiple strategies are combined; for example, in cases of uneven lighting and slight dirt, moderate-intensity light compensation and gentle image inpainting may be applied simultaneously. A strategy evaluation and adaptive adjustment mechanism is also implemented, recording the results and performance indicators of each recognition, continuously updating and optimizing the decision model for strategy selection. For example, if the success rate of a certain strategy under specific conditions is found to be lower than expected, the selection weight of that strategy under similar conditions is reduced; conversely, if a strategy performs well, its weight is increased. In this way, the system can continuously learn and adapt to different environmental conditions and label states, achieving continuous optimization of recognition capabilities.

[0109] This method, based on Hamming distance-based uniformity metrics to enhance decision space diversity, provides a significant performance improvement for fiber optic tag identification. Compared to traditional single-strategy methods, the average identification success rate is improved by over 25%, with particularly significant performance improvements under harsh conditions: a 35% increase in success rate in strong light environments and a 40% increase in severely contaminated conditions. Simultaneously, adaptability and robustness are greatly enhanced, enabling the handling of various complex and extreme situations, such as severely faded tags, partially obscured tags, and tags under strong reflection. This method not only improves identification accuracy but also reduces maintenance costs and the need for manual intervention, providing reliable technical support for fiber optic cable management in smart substations. Practical applications demonstrate that this method has achieved excellent results in multiple smart substations of different sizes and types, significantly improving operational efficiency and safety, and has received high praise from frontline staff.

[0110] Step S4 includes the following sub-steps: S4.1: Based on the structured label recognition data and the knowledge graph, multi-dimensional feature matching is used to compare topological relationships and logical functions. A fuzzy matching strategy of character similarity calculation, semantic similarity evaluation and structural similarity analysis is adopted. Design document data, historical scanning records and related optical cable data are integrated for multi-source data collaborative verification. The effective prediction time method of reservoir computer is used to learn from historical anomaly cases and configure anomaly patterns including mismatch, label error, unauthorized change and potential risk. The Lyapunov index is calculated to evaluate the time scale and range of anomaly impact. The case of physical connection not conforming to design drawings is detected to obtain anomaly detection results. S4.2: Based on the anomaly detection results and tagged operation and maintenance history data including fault history, maintenance records, and performance trends, a multi-objective evolutionary algorithm is used to assess the severity and potential impact of the anomaly from multiple perspectives, including the degree of security threat, reliability impact, functional impact (assessing the scope of functional loss), and economic impact (assessing repair costs and resources). A balance point is sought across multiple assessment dimensions, taking into account load conditions, seasonal factors, and the contextual sensitivity of concurrent events. Based on the severity of the assessment, graded warning information, including emergency warnings, important warnings, general warnings, and attention prompts, is generated. Case reasoning based on similar anomalies is retrieved from the historical case library, and rule reasoning based on application domain expert rules is used to provide targeted handling suggestions, including emergency handling solutions, temporary mitigation solutions, fundamental solutions, and prevention strategies. S4.3: Based on the execution results of the targeted handling suggestions and the actual connection status confirmed on-site, establish a change source identification and multi-level change confirmation process that includes planned changes, on-site discovery and fault handling. Use graph neural network technology to add, modify or delete entity nodes and connection relationships affected by changes using an incremental update strategy. Re-evaluate the direct impact, indirect impact and redundancy assessment of network topology changes. Dynamically update the connection relationship data in the knowledge graph and maintain the change version history to achieve dynamic association and anomaly early warning.

[0111] Specifically, step S4.1 constructs a connection relationship map based on optical cable tags. In smart substations, optical cable connections constitute a complex network topology, and accurately grasping these connection relationships is crucial for system maintenance and fault diagnosis. This sub-step lays the foundation for subsequent consistency verification and anomaly detection by constructing an optical cable connection relationship map. The system first extracts key connection entities and relationship information based on the structured tag data identified in step S3. Connection entities include various equipment nodes (such as intelligent terminals, merging units, protection devices, switches, etc.) and connection points (such as fiber optic interfaces, terminals, connectors, etc.). For each equipment node, the system records its identifier (equipment ID), type (such as protection device, merging unit, etc.), location information (such as bay number, cabinet number, installation location, etc.), and functional attributes (such as protection type, communication protocol, etc.). For connection points, the system records the equipment to which it belongs, port number, interface type (such as LC, SC, ST, etc.), and signal characteristics (such as transmit / receive, wavelength, rate, etc.). During entity extraction, the system uses regular expressions and domain-specific pattern matching rules to accurately extract key information such as device identifiers and port numbers from structured tag data. For tag data with inconsistent formats or non-standardized formats, the system applies named entity recognition technology, using a trained deep learning model to identify entities such as device names and port information in the text.

[0112] After extracting the connection entities, the optical cable connection relationships are analyzed and modeled. A connection relationship refers to the physical connection established between two connection points via an optical cable, representing the channel for information transmission. For each connection relationship, the source endpoint (sender), destination endpoint (receiver), optical cable identifier, optical cable type (e.g., single-mode / multimode, number of cores, etc.), transmission signal type (e.g., GOOSE signal, SV sample value, MMS message, etc.), and logical function (e.g., protection signal, measurement data, control command, etc.) are recorded. A complete connection mapping is established by pairing the source endpoint and the destination endpoint. This pairing is based on explicit identification in the labels (e.g., the labels directly indicate the two ends of the connection and the ports) or through association via optical cable identifiers (different labels with the same optical cable identifier represent the two ends of the same optical cable). For complex multi-end connections (e.g., one-to-many broadcast connections), a hierarchical modeling approach is used, decomposing them into multiple point-to-point connections for representation.

[0113] To enhance the expressive power and analytical efficiency of the optical fiber connection graph, an attribute graph model is used to construct the graph. In this model, device nodes and connection points are the vertices, and the connections are the edges. Each vertex and edge has rich attributes, recording its detailed characteristics. Furthermore, a multi-layered structure is introduced into the graph, including a physical layer (representing actual optical fiber connections), a link layer (representing communication links), and a functional layer (representing logical functional relationships). This multi-layered structure allows the graph to simultaneously express connection information at different levels of abstraction, supporting multi-faceted analysis. During the graph construction process, strict data quality control is implemented, including uniqueness checks on entity identifiers (ensuring each device and connection point has a unique identifier), connection integrity verification (ensuring each connection has a clearly defined source and destination), and attribute validity verification (ensuring that all attribute values ​​are within valid ranges). For detected data quality issues, detailed problem descriptions are recorded, and automatic repair is attempted when conditions permit (e.g., inferring missing information based on context).

[0114] Finally, the constructed graph is enhanced and optimized to improve its practicality and analytical value. Graph enhancement includes adding semantic annotations (such as marking the importance level and redundancy status of connections), inferring implicit relationships (inferring possible but unidentified connections based on known connections), and integrating external knowledge (such as equipment datasheets and system configuration information). Graph optimization includes redundancy elimination (merging duplicate records representing the same connection), consistency correction (resolving contradictions in the data), and structural optimization (adjusting the graph structure to improve query and analysis efficiency). Through these steps, a comprehensive, accurate, and structurally optimized optical cable connection relationship graph is constructed, providing a solid foundation for subsequent analysis and applications. This graph not only records the physical connections of optical cables but also contains rich semantic information, supporting complex queries, analysis, and reasoning, and is the core data structure for realizing intelligent management of optical cables.

[0115] Step S4.2 implements fiber optic network topology analysis and critical connection identification. After constructing a complete fiber optic connection relationship map, it is necessary to deeply analyze the network topology and identify critical connections to provide a basis for subsequent fault impact assessment and prioritization. This sub-step achieves a deep understanding of the fiber optic network and accurate identification of critical connections through complex network analysis methods. First, graph theory algorithms are applied to comprehensively analyze the fiber optic network topology. Connectivity analysis is the first step. The system uses breadth-first search and depth-first search algorithms to identify connected components (interconnected subnetworks) and isolated nodes (devices not connected to other nodes) in the network. This information helps maintenance personnel understand the overall network structure and potential communication isolation issues. Path analysis is the second step. The system uses Dijkstra's algorithm or A* algorithm to calculate the shortest communication path between devices, determining the optimal path and alternative paths for signal transmission. For critical communication links (such as protection signal transmission paths), the system pays special attention to their path length, the number of devices traversed, and potential bottlenecks. Redundancy analysis is the third step. The system assesses the redundancy of the network and identifies single-point failure risks (i.e., connection points where failure would lead to network segmentation or loss of critical functions). By calculating edge connectivity and vertex connectivity, the robustness of the network structure can be quantified, and weak links that need to be strengthened can be identified.

[0116] To identify critical connections in a network, a novel approach is taken to apply centrality metrics and community detection algorithms. Centrality metrics assess the importance of nodes or edges within a network. Several centrality metrics are calculated, including degree centrality (number of connections, reflecting the direct influence range of a node), betweenness centrality (number of shortest paths through the node, reflecting its importance as a "bridge"), proximity centrality (average distance to other nodes, reflecting the efficiency of a node in acquiring or propagating information), and eigenvector centrality (a centrality metric considering the importance of neighbors, reflecting the node's influence in the overall network). By integrating these centrality metrics, the importance of devices and connections can be comprehensively evaluated, identifying critical nodes and links in the network. Community detection is another important analysis. Using algorithms such as modularity optimization or label propagation, the network is divided into tightly connected subgroups (communities). These communities typically represent functionally related groups of devices (such as devices in the same protection system), and the connections between communities represent interactions between different functional modules. Particular attention is paid to connections between communities because these connections are often critical channels for information exchange, and their failure can affect multiple functional modules.

[0117] In identifying critical connections, a comprehensive evaluation mechanism combining network topology characteristics and service functions was considered. First, connections were classified and graded according to their service functions, such as distinguishing between protection signals (e.g., circuit breaker trip commands), measurement signals (e.g., current and voltage sampling values), and management signals (e.g., operational status monitoring). Different importance weights were assigned to each type of signal, with protection signals typically having the highest priority. Second, a comprehensive importance score was calculated for each connection, combining functional classification and network topology analysis. The calculation formula comprehensively considered the connection's functional importance, network centrality, and redundancy. Finally, the connections were ranked according to their comprehensive scores to determine the list of critical connections. These critical connections will receive special attention in subsequent monitoring and maintenance, such as prioritizing monitoring and alarm configuration, strengthening daily inspections, and developing dedicated emergency response plans.

[0118] Furthermore, time-series network analysis was applied to study the dynamic characteristics of the optical fiber network. By analyzing historical monitoring data and operational records, the system established a model of connection status changes over time, including traffic patterns (changes in communication volume at different times), fault frequency (historical frequency of connection problems), and performance fluctuations (fluctuations in transmission quality). This time-series analysis helps maintenance personnel understand the network's dynamic behavior, predict potential anomalies, and develop targeted maintenance strategies. Through the aforementioned topology analysis and critical connection identification, a comprehensive "portrait" of the optical fiber network was established, understanding not only its static structure but also its dynamic characteristics, laying a solid foundation for accurate and efficient optical fiber management. In practical applications, these analytical results directly support multiple maintenance decisions, such as optimizing inspection routes, developing differentiated maintenance strategies, and rationally allocating maintenance resources, significantly improving maintenance efficiency and system reliability.

[0119] Step S4.3 implements fiber optic cable tag consistency verification and anomaly detection. During substation operation, fiber optic cable connections may change due to maintenance operations, equipment replacement, or other reasons, leading to inconsistencies between the actual connection and the information identified by the tags. This inconsistency can pose serious safety hazards, thus requiring the establishment of an effective consistency verification and anomaly detection mechanism. This sub-step achieves intelligent supervision of fiber optic cable tag consistency through multi-source data comparison and anomaly pattern recognition. First, multi-source data acquisition and integration are implemented to establish a comprehensive comparison foundation. These data sources include design drawings and specifications (such as system design documents, wiring diagrams, configuration lists, etc., reflecting design intent), SCD configuration files (IEC 61850 standard system configuration description files, containing logical connection relationships of equipment), communication traffic monitoring (actual communication data collected through network monitoring equipment, reflecting operating status), and historical inspection records (previous inspection findings and processing results, providing historical reference). A data integration framework is adopted to address the differences in format, semantics, and time scale among different data sources, establishing a unified data model. For unstructured data (such as design drawings), OCR and image understanding technologies are used to extract structured information; for semi-structured data (such as SCD files), an XML parser is used to extract key configurations; and for structured data (such as communication monitoring data), standardization is performed directly. Through this multi-source data integration, a comprehensive view containing design intent, configuration information, and actual operating status is established.

[0120] Based on integrated multi-source data, multi-dimensional consistency checks are performed. Physical connection consistency checks compare the physical connections identified by fiber optic cable tags with the connections specified in the design drawings, detecting unauthorized connection changes or incorrect connections. Logical function consistency checks compare the functional roles of the actual connections with the logical functions defined in the SCD file, verifying that each logical function has a corresponding physical implementation and that these implementations are correct. Communication behavior consistency checks analyze the match between actual communication traffic and expected communication patterns, identifying abnormal communication behaviors (such as devices that should be communicating but are not, or devices that should not be communicating but are communicating). Historical change consistency checks compare the current state with historical records, tracking the legality and completeness of connection changes, ensuring that all changes are properly authorized and recorded. Through these multi-dimensional consistency checks, the compliance of fiber optic cable connections can be comprehensively assessed, and various potential problems can be detected.

[0121] To improve the accuracy and intelligence of anomaly detection, a machine learning-based anomaly pattern recognition system was implemented. The system first established a normal pattern library for fiber optic cable connections, including typical connection patterns (based on design standards and industry best practices), site-specific patterns (considering the actual configuration and operational requirements of specific substations), and time-series patterns (considering normal changes during different operational phases). Then, various anomaly detection algorithms were applied to identify deviations from normal patterns. These algorithms included statistical methods (such as multivariate anomaly detection, identifying outliers across multiple feature dimensions), density-based methods (such as DBSCAN, identifying connection points with significantly lower density than the surrounding area), and learning-based methods (such as autoencoders, learning the implicit features of normal connections to detect anomalies difficult to express with explicit rules). To handle different types of anomalies, a hierarchical detection strategy was adopted, progressing from simple rules to complex models. For obvious violations of physical constraints or design specifications (such as port type mismatches), rule-based rapid detection was used; for subtle anomalies (such as gradual changes in communication patterns), more complex machine learning models were applied.

[0122] A classification, assessment, and intelligent handling mechanism was implemented for detected anomalies. Anomaly classification is based on multi-dimensional assessment, including severity (potential impact on system functionality and security), urgency (time window for handling), and confidence (certainty of the anomaly detection result). A risk scoring model is used to comprehensively consider these factors and assign a risk level to each anomaly, such as "urgent" (high-risk anomaly requiring immediate handling), "warning" (medium-risk anomaly requiring attention but not urgent), and "hint" (minor anomaly that may have a problem but carries low risk). For anomalies of different risk levels, corresponding handling suggestions are provided, including specific checkpoints (locations and parameters requiring key verification), possible cause analysis (based on historical cases and expert knowledge), and suggested handling procedures (such as emergency isolation, planned maintenance, etc.). A closed-loop management system for anomaly handling has also been established, recording the entire process of discovery, diagnosis, handling, and verification for each anomaly, forming a complete knowledge base to support rapid resolution of similar problems in the future.

[0123] This multi-source data comparison and intelligent anomaly detection significantly improves the security and reliability of fiber optic cable management. Practical applications show that this method can detect subtle anomalies that are easily overlooked by traditional manual inspections, such as minor inconsistencies between logical configurations and physical connections, and gradual changes in communication modes. Early warning functions help maintenance personnel intervene before problems escalate into serious faults, greatly reducing the risk of protection failures or malfunctions due to connection errors. Simultaneously, intelligent anomaly handling suggestions reduce the cognitive burden on maintenance personnel, accelerate the problem-solving process, and improve maintenance efficiency. In multiple actual substation applications, the system detects an average of 5-10 potential problems per month, of which approximately 30% are hidden dangers that are difficult to detect using traditional methods, providing strong technical support for the safe operation of substations.

[0124] Step S5 includes the following sub-steps: S5.1: The design includes a mobile terminal with a GPU supporting AI inference, a high-definition camera with image stabilization and fill light, an RFID identification module for auxiliary identification, and a hardware system that supports label printing equipment for various materials. It optimizes the terminal's processing capabilities through edge computing technology to support AI inference and image processing, ensuring the availability of data processing and label identification in weak network or offline environments, resulting in a usable hardware system. S5.2: Based on the available hardware system, develop core modules including a data parsing engine for processing the intelligent substation configuration file using NLP and GNN technologies, an AI generation engine for label optimization generation using GAN and reinforcement learning technologies, an intelligent recognition engine for multimodal parsing using CV and multimodal technologies, and a knowledge graph service for storing relationships and historical data. Containerization technology is used to achieve loose coupling and flexible deployment between modules, forming a microservice architecture that supports independent development, deployment, and expansion. S5.3: Based on the microservice architecture, design a layered data storage strategy for different types of data to achieve separation of hot and cold data, build a multi-level security mechanism including data encryption to protect sensitive information, access control to restrict operation permissions, and operation audit records for system use, develop standardized API interfaces to support data exchange and function integration with other substation systems such as SCADA system and asset management system, design a flexible plug-in mechanism that allows third parties to develop extended functions, and realize a full lifecycle management system.

[0125] Specifically, step S5.1 implements historical data mining and trend analysis of optical cable tags. During the long-term operation of smart substations, historical data related to optical cable tags contains rich information. In-depth mining and trend analysis of this data can uncover potential patterns, predict future changes, and provide a scientific basis for preventative maintenance. This sub-step utilizes advanced data mining and time-series analysis techniques to achieve intelligent utilization of historical optical cable tag data. First, a comprehensive historical data warehouse is constructed, integrating multi-source historical records. This data includes tag recognition history (including recognition success rate, recognition quality score, abnormal recognition records, etc.), optical cable connection change records (recording the time, reason, and involved equipment of connection changes), maintenance records (including detailed records of daily inspections, periodic maintenance, and fault handling), and operational anomaly records (such as communication interruptions, signal quality degradation, and other abnormal events). Data warehouse technology is used to achieve efficient storage and rapid retrieval of historical data, supporting flexible queries by time, equipment, region, and other dimensions. For unstructured maintenance records (such as text descriptions of problems and handling methods), natural language processing technology is applied to extract key information and transform it into structured data, enriching the content of the data warehouse. To ensure data quality, the system implemented a data cleaning and preprocessing process, including removing outliers (such as obviously erroneous timestamps or unreasonable measurements), filling missing values ​​(through time interpolation or replacement with similar samples), and standardization (converting data from different sources into a unified metric).

[0126] After data preparation, multi-level pattern mining and trend analysis are conducted. Tag degradation pattern analysis studies the changes in the physical state of tags over time. By analyzing historical recognition rate and image quality data, a time-series model of tag aging is established, including tag material aging curves (expected lifespan of tags of different materials under different environments), environmental influencing factors (such as the impact of temperature, humidity, and light on tag aging speed), and the impact of sudden events (such as the impact of abnormal high temperatures and water immersion on tag damage). Through these models, the system can predict the remaining lifespan of tags, providing a scientific basis for tag replacement. Connection change pattern analysis studies the frequency, causes, and impacts of fiber optic cable connection changes. Historical change records are analyzed to identify high-frequency change areas (areas with a change frequency significantly higher than the average, potentially indicating design flaws or operational irregularities), change correlations (the interrelationships between changes in different areas or equipment, such as a change in one area often triggering related changes in others), and change timing characteristics (such as the existence of periodic change patterns or change patterns related to specific operational phases). These analyses help maintenance personnel understand the underlying causes of changes and optimize system design and maintenance processes. Fault correlation analysis studies the relationship between fiber optic cable connection problems and system failures. By combining optical cable anomaly records and system fault records, association rule mining algorithms (such as the Apriori algorithm) are applied to discover correlation patterns between optical cable problems and system faults. For example, specific types of connection anomalies often precede certain types of system faults, or connection problems in a certain area are highly correlated with anomalies in specific protection functions. These correlation patterns provide important clues for fault prevention and diagnosis.

[0127] Based on pattern mining of historical data, a predictive model was constructed to scientifically predict future trends. First, suitable predictive models for fiber optic cable management scenarios were selected, including time series models (such as ARIMA, suitable for data with obvious time dependencies), regression models (such as random forest regression, handling multivariate prediction problems), and neural network models (such as LSTM, handling complex time series patterns). Then, model ensemble techniques were applied to combine the prediction results of multiple base models, improving the accuracy and stability of the predictions. Different model combinations were selected for different prediction tasks; for example, an LSTM network combined with a physical model was used for label aging prediction, while an ensemble random forest model was used for connection change prediction. Special attention was paid to uncertainty assessment, providing confidence intervals or probability distributions for each prediction to help decision-makers understand the reliability and risk range of the predictions. During model training and evaluation, rigorous cross-validation and continuous update mechanisms were employed to ensure the reliability and timeliness of model performance.

[0128] To visually present analysis results and support decision-making, intelligent visualization and reporting functions were developed. Trend charts display historical changes and future forecasts for key indicators, such as declining tag recognition rates and varying frequencies of connection changes, intuitively showing long-term trends and sudden changes. Hotspot maps display the problem density and risk levels in different areas within the substation, helping maintenance personnel quickly identify areas requiring priority attention. Network diagrams illustrate the complex relationships between different devices, fiber optic cables, and problems, helping to understand system interdependencies and potential risk propagation paths. Early warning dashboards display real-time warning information based on historical analysis, including predicted high-risk areas and potential tag aging issues, supporting proactive intervention. These visualization tools not only provide an intuitive representation of data but also support interactive exploration, allowing users to view data from different angles and granularities, gaining a deeper understanding of underlying patterns and relationships.

[0129] By leveraging comprehensive historical data mining and trend analysis, the system transforms reactive problem response into proactive risk management, significantly enhancing the foresight and scientific rigor of fiber optic cable management. In practical applications, the prediction accuracy exceeds 85%, successfully identifying and warning of numerous potential problems, thus saving valuable intervention time for maintenance personnel. Simultaneously, the system's historical analysis has revealed several optimization opportunities in system design and maintenance processes, driving continuous improvement. In the long run, this data-driven management approach not only improves system reliability but also optimizes the allocation of maintenance resources, reduces maintenance costs, and achieves more efficient and reliable fiber optic cable management.

[0130] Step S5.2 implements intelligent preventive maintenance decision-making based on reinforcement learning. In traditional optical cable management, maintenance decisions are often based on fixed plans or passive problem responses, lacking flexibility and foresight. This sub-step, through reinforcement learning technology, implements intelligent preventive maintenance decision-making, optimizing maintenance strategies and achieving optimal long-term benefits while considering various constraints. First, a decision-making framework for optical cable preventive maintenance is constructed, formalizing the maintenance decision-making process as a Markov Decision Process (MDP). In this framework, the state space describes the current state of the system, including the health status of each optical cable tag (e.g., recognition rate, predicted remaining lifespan), the operating status of equipment (e.g., communication quality, error rate), and the status of maintenance resources (e.g., available maintenance personnel, spare parts). The action space defines possible maintenance decisions, including different types of maintenance activities (e.g., tag replacement, connection checks, cleaning maintenance), different execution times (e.g., immediate execution, planned execution, or delayed execution), and different resource allocation schemes (e.g., assigning personnel to perform maintenance tasks). The reward function evaluates the effectiveness of each decision, taking into account multiple objectives such as improved system reliability (the value of reducing the risk of failure), maintenance costs (including labor costs, material costs, and downtime costs), and maintenance efficiency (the time and resources required to complete maintenance). A long-term reward mechanism is specifically designed to not only focus on the immediate effects of decisions but also consider their impact on the long-term health of the system, avoiding short-sighted decisions.

[0131] Based on the aforementioned decision-making framework, advanced reinforcement learning algorithms were implemented to learn the optimal maintenance strategy. Deep Q-Networks (DQNs) were used as the core algorithm, combined with experience replay and target network techniques to improve the stability and efficiency of learning. Several innovations were introduced to address the characteristics of maintenance decisions: a priority experience replay mechanism prioritizes learning experiences containing important decisions (such as decisions to avoid serious failures or significantly reduce costs), accelerating the learning of key strategies. Hierarchical reinforcement learning decomposes complex decision problems into multiple levels, such as top-level decisions on maintenance priorities, mid-level decisions on specific maintenance types, and bottom-level decisions on execution details, simplifying the learning difficulty through a hierarchical structure. Safety-constrained reinforcement learning introduces safety constraints to ensure that no dangerous decisions are made during the learning process (such as simultaneously maintaining multiple critical devices leading to system malfunction), guaranteeing system safety. For the specific needs of fiber optic cable management, domain-specific network structures and feature representations were also designed, such as using graph neural networks to capture the topology of fiber optic cable connections and using attention mechanisms to focus on high-risk areas.

[0132] One key innovation is the handling of multi-objective balancing and constraints. In practical operation and maintenance, it is often necessary to balance multiple conflicting objectives, such as system reliability versus maintenance cost, or short-term benefits versus long-term health. A multi-objective reinforcement learning method based on the Pareto front is employed to learn a series of non-dominated solutions, providing optimal strategies for different preferences. Simultaneously, mechanisms for handling soft and hard constraints are introduced. Soft constraints are implemented through adjustments to the reward function, such as penalizing decisions that violate budget constraints; hard constraints are implemented through restrictions on the action space, such as completely excluding maintenance combinations that violate safety regulations. Various practical constraints in actual operation and maintenance are also considered, such as maintenance time window limitations (some equipment can only be maintained during specific time periods), personnel skill constraints (specific maintenance tasks require personnel with specific skills), and resource dependency constraints (some maintenance activities depend on the availability of specific equipment or materials). In this way, the generated maintenance decisions optimize the objective function while satisfying various practical constraints, making them highly operable.

[0133] To improve the system's adaptability and usability, continuous learning and human-machine collaboration mechanisms were implemented. Continuous learning allows the system to learn from the actual results of each maintenance activity, continuously optimizing its decision-making model. The system records the planning, execution process, and effects of each maintenance, compares expected and actual results, identifies prediction deviations, and updates the model accordingly. This closed-loop learning ensures the system can adapt to changing environments and equipment characteristics, maintaining the effectiveness of its decisions. Human-machine collaboration is another important feature. The system does not simply replace human decision-making but provides decision support, complementing expert knowledge. It provides detailed explanations for each suggested decision, including the decision basis (e.g., which factors dominated the decision), expected effects (e.g., quantitative indicators such as reliability improvement and cost savings), and uncertainty assessments (potential risks and variables in the decision). Simultaneously, human experts are allowed to adjust and override automated decisions, learning from these human interventions to continuously improve their decision-making model.

[0134] This reinforcement learning-based intelligent preventative maintenance decision-making achieves a shift from passive response to proactive prevention, significantly improving maintenance efficiency and effectiveness. In practical applications, maintenance recommendations consider not only the current state of the equipment but also historical patterns, predicted trends, and overall system health, generating a comprehensively optimized maintenance plan. Compared to traditional rule-based or fixed-cycle maintenance, this intelligent decision-making significantly reduces maintenance costs (average savings of 15-20%), reduces unexpected failures (approximately 30%), and extends equipment lifespan (average extension of 10-15%). Especially under resource constraints, it can intelligently allocate limited resources, prioritizing the most critical maintenance needs to maximize benefits. The system's flexibility and adaptability have also been highly praised by operations and maintenance personnel; it can adjust maintenance strategies based on constantly changing conditions and priorities, providing decision support that always aligns with actual needs.

[0135] Step S5.3 utilizes knowledge graphs to achieve the accumulation and intelligent sharing of fiber optic cable management knowledge. During the long-term operation and maintenance of intelligent substations, a wealth of experience and knowledge regarding fiber optic cable management has been accumulated. Effective accumulation and sharing of this valuable knowledge will greatly improve the overall capabilities and efficiency of the team. This sub-step utilizes knowledge graph technology to achieve the systematic accumulation and intelligent sharing of fiber optic cable management knowledge, providing strong knowledge support for the operation and maintenance team. The system first constructs a comprehensive fiber optic cable management knowledge graph, systematically organizing domain knowledge. This knowledge graph contains various types of entities, such as equipment types (e.g., various protection devices, merging units, switches, etc.), tag types (e.g., tags of different materials and uses), problem types (e.g., tag damage, connection errors, etc.), and solutions (e.g., tag replacement methods, connection inspection steps, etc.). Entities are connected through rich relationships, such as "equipment using tags," "problem affecting equipment," and "solution solving the problem," forming a complex semantic network. Each entity and relationship has rich attributes, recording its characteristics and related information. This knowledge graph is constructed through multiple methods, including automatic extraction (extracting structured knowledge from technical documents, specifications, and manuals), expert editing (knowledge being directly input and reviewed by domain experts), and experience learning (learning experiential knowledge from historical cases and maintenance records). In particular, for tacit knowledge (such as experiential judgments and best practices), it is made explicit and incorporated into the knowledge graph through expert interviews and case analysis.

[0136] After constructing the basic knowledge graph, a knowledge enrichment and evolution mechanism was implemented to ensure continuous updating and improvement of knowledge. Case-based knowledge association links historical maintenance cases with entities and relationships in the knowledge graph, providing concrete examples to support abstract knowledge. Each case records the problem description, cause analysis, solution process, and effect evaluation, connected to relevant nodes in the knowledge graph through semantic annotation. For example, a case about "accelerated tag aging under high temperature conditions" can be associated with the "environmental impact" and "tag lifespan" nodes in the knowledge graph, enriching the practical content of these concepts. Best practice sedimentation collects and summarizes best practices for efficient maintenance, including step sequences, precautions, and effect verification methods. These best practices, as special knowledge nodes, are associated with relevant equipment, problems, and contexts, forming actionable knowledge guidelines. For example, "tag identification techniques in densely populated fiber optic areas" might include specific suggestions such as specific shooting angles, lighting adjustments, and the use of auxiliary tools. Problem-solving pattern extraction extracts general problem-solving patterns from a large number of historical cases, including characteristic descriptions of typical problems, diagnostic methods, and solution approaches. These patterns transcend specific cases, providing a more general problem-solving framework to help handle newly emerging similar problems. Knowledge graphs also support dynamic evolution, maintaining the timeliness and relevance of knowledge through multiple mechanisms. Knowledge auditing regularly evaluates the frequency of use, effectiveness, and timeliness of knowledge items, eliminating outdated or low-value knowledge; knowledge updates track the introduction of new technologies, equipment, and standards, updating relevant knowledge in a timely manner; community feedback collects user feedback and suggestions on knowledge content, guiding the optimization and expansion of knowledge.

[0137] To achieve intelligent knowledge sharing and application, various knowledge service functions have been developed. Context-aware knowledge recommendation automatically recommends relevant knowledge content based on the user's current work context. The system considers multiple contextual factors, such as the user's current task (e.g., tag change, connection check), the object of operation (e.g., specific type of device or tag), and environmental conditions (e.g., work area, time constraints). By analyzing this contextual information, it proactively provides potentially useful knowledge before the user needs it, such as relevant best practices, solutions to common problems, or applicable technical specifications. Multimodal knowledge presentation selects the most appropriate presentation method based on the knowledge content and usage scenario. For procedural knowledge (e.g., operation steps), interactive flowcharts or step-by-step guides may be provided; for principle-based knowledge (e.g., technical principles), concept maps and explanatory text may be provided; for case-based knowledge, the system may provide detailed case descriptions and key experience summaries. It also supports multiple media formats, such as text, images, videos, and augmented reality guidance, selecting the most effective presentation method based on knowledge complexity and usage environment. Intelligent question answering and diagnostics allow users to query the knowledge base using natural language to obtain problem-solving suggestions. Based on knowledge graphs and natural language processing technologies, the system understands the user's intent, searches for relevant knowledge, and generates targeted answers. For complex questions, it initiates an interactive diagnostic dialogue, guiding the user to provide more information through targeted questions, gradually narrowing down the possible problems, and ultimately providing a precise solution.

[0138] Another innovation is collaborative learning and knowledge co-creation, promoting the accumulation and sharing of team wisdom. The experience-sharing platform allows maintenance personnel to record and share their work experience, including problem-solving, innovative methods, and lessons learned. It provides structured templates to guide users in providing key information and automatically integrates these experiences into a knowledge graph. The knowledge community function supports discussions and collaboration around specific topics or issues, such as joint diagnosis of complex faults and evaluation discussions of new technologies. These discussions are tracked to extract valuable insights and enrich the knowledge base. In maintenance operations, it supports team collaboration and knowledge application. For example, when multiple people are working together to solve complex problems, it can integrate knowledge from different specialties to provide comprehensive support; or in emergency situations, it can quickly retrieve the most relevant emergency response knowledge to support rapid decision-making.

[0139] Through this knowledge graph-based knowledge accumulation and sharing, the effective inheritance and application of fiber optic cable management knowledge has been achieved, significantly improving the team's overall capabilities and efficiency. In practical applications, newly joined maintenance personnel can master job skills more quickly (training time is reduced by an average of 40%); the time to solve complex problems is significantly reduced (by an average of 30%); and the team's overall problem-solving ability and innovation level are also improved. The system not only preserves valuable experience and knowledge, preventing knowledge loss with personnel turnover, but also promotes the dissemination and standardization of best practices, improving the consistency of maintenance quality. In the long run, this knowledge management approach cultivates a learning organization culture, encourages continuous improvement and innovation, and provides strong intellectual support for the safe and reliable operation of smart substations. Through intelligent knowledge sharing, the capability gap between employees with different experience levels is effectively narrowed, enabling the entire team to carry out fiber optic cable management work at a higher level and with more consistent standards, ultimately improving the overall operation and maintenance quality and efficiency of smart substations.

[0140] like Figure 2 As shown, the present invention also provides a full lifecycle management system for optical fiber smart tags, comprising: The data parsing module 601 is used to extract structured data, including descriptions of logical nodes, logical devices, communication connections, and physical devices, based on the intelligent substation configuration file, which includes intelligent substation SCD files and SPCD files, through an XML parsing engine. It uses natural language processing technology to perform semantic analysis on unstructured text, combines the symbolic quantile regression method to quantify text features, and constructs a knowledge graph containing the relationship between devices, ports, optical cables, and virtual terminals. The tag generation module 602 is used to dynamically adjust the text size, QR code position and fault tolerance according to the optical cable type and pasting scenario based on the knowledge graph and generative adversarial network technology to generate optical cable smart tags adapted to different scenarios. The tag recognition module 603 is used to collect tag images of the optical cable on site based on the optical cable smart tag, and use multimodal fusion computer vision technology to convert them into structured tag recognition data containing optical cable type, connection relationship and transmission signal type; The anomaly warning module 604 is used to rely on the structured tag recognition data and the knowledge graph to check the consistency between the tag information and the expected connection relationship through a real-time data comparison algorithm, analyze potential anomalies, evaluate the safety impact, reliability impact and functional impact of the anomalies using a multi-objective evolutionary algorithm, and generate graded warning information and targeted handling suggestions. The system integration module 605 is used to build a modular system based on a microservice architecture, including a data parsing engine, an AI generation engine, an intelligent recognition engine, and a knowledge graph service. It supports data exchange and functional integration with other systems in the smart substation through standardized API interfaces, and realizes intelligent management of the entire lifecycle of optical cable tags from generation and recognition to operation and maintenance.

[0141] It should be noted that the above specific embodiments are merely illustrative examples of the present invention and not limitations thereof. Those skilled in the art, under the guidance of the present invention, can make many modifications and variations without departing from the spirit and scope of the claims, and these all fall within the protection scope of the present invention.

Claims

1. A method for full lifecycle management of optical fiber smart tags, characterized in that, Includes the following steps: Based on the intelligent substation configuration file, which includes intelligent substation SCD and SPCD files, the structured data including descriptions of logical nodes, logical devices, communication connections and physical devices is extracted through an XML parsing engine. Natural language processing technology is used to perform semantic analysis on unstructured text, and the text features are quantified by the symbolic quantile regression method to construct a knowledge graph containing the relationship between devices, ports, optical cables and virtual terminals. Based on the knowledge graph, generative adversarial network technology is used to dynamically adjust the text size, QR code position and fault tolerance according to the type of optical cable and the pasting scenario, so as to generate smart optical cable labels that are adapted to different scenarios. Based on the aforementioned smart optical cable tags, tag images of the optical cables on site are collected, and multimodal fusion computer vision technology is used to convert them into structured tag recognition data containing optical cable type, connection relationship and transmission signal type. Based on the structured tag recognition data and the knowledge graph, the consistency between tag information and expected connection relationships is checked through real-time data comparison algorithms, potential anomalies are analyzed, and the security impact, reliability impact, and functional impact of anomalies are evaluated using multi-objective evolutionary algorithms to generate graded early warning information and targeted handling suggestions. A modular system based on a microservice architecture is constructed, including a data parsing engine, an AI generation engine, an intelligent recognition engine, and a knowledge graph service. Through standardized API interfaces, it supports data exchange and functional integration with other systems in smart substations, realizing intelligent management of the entire lifecycle of optical cable tags from generation and recognition to operation and maintenance.

2. The method according to claim 1, characterized in that, The intelligent substation configuration file, based on intelligent substation SCD and SPCD files, extracts structured data including descriptions of logical nodes, logical devices, communication connections, and physical devices using an XML parsing engine. Natural language processing techniques are used to perform semantic analysis on the unstructured text, and symbolic quantile regression is used to quantify text features. This constructs a knowledge graph containing relationships between devices, ports, optical cables, and virtual terminals, including: Based on the intelligent substation configuration file, graph neural network technology is used to mine the implicit connection relationships between devices, identify the indirect associations across optical cables, and establish a multi-level graph structure association network with an entity layer containing physical devices, ports, optical cables and virtual terminals, a relationship layer containing connection relationships, transmission relationships and attribution relationships, and an attribute layer containing device models, port protocols and the number of optical fiber cores. By using reservoir computing technology to predict potential optical cable connection patterns based on historical connection patterns, and employing Lyapunov index analysis to dynamically adjust the reliability of the prediction strategy, missing optical cable usage labels and spare fiber core information are automatically completed to obtain the completed label information. Based on the completed tag information and historical operation and maintenance data including historical operation and maintenance records, alarm logs, optical cable test data and tag usage feedback, a multi-objective evolutionary algorithm with diverse decision space is used to perform multi-dimensional data fusion while considering data integrity, consistency and timeliness. Based on the uniformity metric of Hamming distance, connection relationship conflicts, outdated information and abnormal patterns are detected and corrected to obtain the verified and optimized knowledge graph.

3. The method according to claim 2, characterized in that, The quantification of text features using the combined signed quantile regression method includes: For the unstructured text containing cabinet name, port description, and virtual terminal signal, use a power industry professional thesaurus and rule base to perform word segmentation and part-of-speech tagging, and perform symbolic processing to map professional terms and descriptions into symbolic representations to generate symbolic text; Based on the symbolized text, a quantile-based regression model is established to map the semantic features of the text to different quantiles, analyze the correlation strength between symbols, calculate the conditional quantiles of different features to determine the importance of symbols in optical cable tags, and determine the importance weights in optical cable tags. Based on the aforementioned importance weights, the following are extracted: optical cable types including trunk optical cable, pigtail cable, and single-mode / multimode optical fiber; connection relationships of signal flow and logical relationships between devices; transmission signal types including GOOSE signal, SV sample value, and MMS message; priority information including critical protection signal and non-critical monitoring signal; and quantized text features are obtained.

4. The method according to claim 1, characterized in that, The generation of fiber optic smart tags adapted to different scenarios includes: The knowledge graph is used to analyze the physical space constraints, environmental factors and viewing conditions of different application scenarios, including narrow cabinet spaces, outdoor environments and high-frequency viewing areas. The reservoir computing technology is used to learn from historical tag usage data to predict tag usage environment parameters and generate an initial tag template candidate set including compact tags, waterproof and weather-resistant tags and wear-resistant tags. Based on the initial candidate set of label templates, label content data containing fiber optic cable ID and connection information, and target scene parameters, generative adversarial network technology is used to generate label design schemes according to fiber optic cable type and pasting scene. A discriminator evaluates the information integrity, readability and recognition rate of the label design based on historical label samples. Content constraints are added to ensure that key information is clearly visible, QR code constraints are added to adjust the size, position and error tolerance according to the usage environment, spatial constraints are added to optimize the overall layout according to the pasting position, and visual constraints are added to ensure the contrast between text and background, thus generating an optimized label layout scheme. Based on the optimized label layout scheme and historical label usage data including label durability and scanning success rate, a reinforcement learning model is constructed. The Lyapunov index is used to analyze the impact of parameter adjustments on label performance, adaptively optimizing QR code encoding density, error correction level, and UV resistance encoding capability, adaptively optimizing text font size, thickness, and contrast, and adaptively optimizing material parameters including label thickness, adhesive type, and protective layer processing, to obtain the optical cable smart label.

5. The method according to claim 4, characterized in that, Generative adversarial network (GAN) technology is used to optimize label design, including: Using a large number of historical tag samples containing successful cases, failed cases, and expert annotations, along with corresponding usage effect evaluations, as training data, a generative adversarial network model was trained, which included a generator responsible for generating tag design schemes and a discriminator for evaluating whether the generated tag designs met the expected standards. Content constraints were added to ensure that key information was clearly visible, QR code constraints to automatically adjust the size and position of the QR code and the fault tolerance rate according to the usage environment, spatial constraints to optimize the overall layout according to the available space of the pasting position, and visual constraints to ensure the contrast between text and background to improve readability, thus forming a GAN model under constraints. Based on the GAN model under the aforementioned constraints, special adjustments were made to different types of ODF tags that simplify information display and highlight port numbers and optical cable routes, device connection tags that emphasize information at both ends of the equipment and clearly indicate signal types, and tail cable tags that optimize information layout under small size to ensure recognizability. The resolution limitations of printing equipment, the characteristics of commonly used label materials with different ink absorption properties, possible deformation during the pasting process, and visual changes during the aging process were all taken into account to generate a variety of label design schemes. Based on the various label design schemes, through multiple rounds of optimization and iteration, the optimized label layout scheme is obtained by simultaneously considering functional requirements to meet label recognition and information transmission requirements, aesthetic enhancement to improve visual effects, and practicality to ensure ease of use.

6. The method according to claim 1, characterized in that, The conversion into structured tag identification data, which includes fiber optic cable type, connection relationship, and transmission signal type, includes: Based on the aforementioned optical cable smart tag, the tag image is collected by a mobile terminal, combined with supplementary environmental data obtained in areas with tag damage, obstructed view, and high-density optical cable, and the light intensity, collection angle, and collection distance are recorded simultaneously. The image enhancement algorithms of adaptive histogram equalization, nonlocal mean filtering, and super-resolution reconstruction are applied, and the geometric correction algorithm and illumination compensation algorithm are applied to generate a standardized tag image with uniform brightness, contrast, viewing angle, and resolution. Based on the standardized label image, the QR code region is located and extracted through gradient analysis, morphological verification, and localization point detection. A super-resolution reconstruction technique combining residual learning network and perceptual loss function is used to process low-resolution or blurry QR codes. Edge repair techniques that preserve structural integrity, context-aware filling, and topology are used to process partially worn or broken QR codes. An attention mechanism that combines spatial attention to focus on the effective region, channel attention to dynamically adjust feature weights, and multi-scale fusion to capture local and global features is used. QR code recognition is achieved through adaptive binarization, perspective correction, and error correction enhancement decoding to obtain the QR code recognition result. Based on the QR code recognition results, a multi-objective evolutionary algorithm with decision space diversity is introduced, which encodes key parameters and strategies of the recognition process into decision vectors. An initial solution set with diversity is generated based on the uniformity metric of Hamming distance. Multiple complementary recognition strategies are obtained through a two-layer selection process using Pareto advantage and decision space diversity. Simultaneously, optical character recognition is performed on the text information on the label image. A multi-modal fusion technique, employing information consistency verification, complementary information integration, and redundant information utilization, is used to fuse the QR code information with the text recognition results. Natural language processing technology is then used to convert this data into a standardized data structure, resulting in structured label recognition data containing information about the optical cable type, connection relationship, and transmission signal type.

7. The method according to claim 6, characterized in that, The process of acquiring tag images of the optical cables based on the smart tags, and converting them into structured tag recognition data containing optical cable type, connection relationship, and transmission signal type using multimodal fusion computer vision technology, also includes: The key parameters and strategies in the identification process are encoded into decision vectors, the Hamming distance between each decision vector in the solution set is calculated, and a density-based diversity evaluation function is designed to use the diversity index as an additional optimization objective. Based on the diversity evaluation function, an initial solution set with diversity is generated. New candidate solutions are generated through mutation and crossover to explore the decision space. A two-layer selection is performed based on Pareto advantage and then based on the diversity of the decision space. The diversity weight is dynamically adjusted according to the convergence of the optimization process, the learning rate is adjusted according to the sensitivity of different parameters, and an adaptive adjustment mechanism is used to dynamically balance exploration and utilization to obtain a variety of complementary recognition strategies. Based on the aforementioned complementary recognition strategies, a strategy that focuses more on illumination compensation is selected in strong light interference environments, a more aggressive image restoration strategy is selected in severe damage situations, and a simplified strategy with higher computational efficiency is selected when computational resources are limited. The recognition method most suitable for the current situation is dynamically selected or combined according to the environmental conditions detected in real time, thereby enhancing the diversity of the decision space.

8. The method according to claim 1, characterized in that, The process involves using the structured tag recognition data and the knowledge graph, employing a real-time data comparison algorithm to check the consistency between tag information and expected connection relationships, analyzing potential anomalies, and utilizing a multi-objective evolutionary algorithm to assess the security, reliability, and functional impacts of the anomalies, generating tiered early warning information and targeted handling suggestions, including: Based on the structured label recognition data and the knowledge graph, multi-dimensional feature matching is used to compare topological relationships and logical functions. A fuzzy matching strategy of character similarity calculation, semantic similarity evaluation and structural similarity analysis is adopted. Design document data, historical scanning records and related optical cable data are integrated for multi-source data collaborative verification. The effective prediction time method of reservoir computer is used to learn from historical anomaly cases and configure anomaly patterns including mismatch, label error, unauthorized change and potential risk. The Lyapunov index is calculated to evaluate the time scale and range of anomaly impact. The case of physical connection not conforming to design drawings is detected to obtain anomaly detection results. Based on the anomaly detection results and tagged operation and maintenance history data including fault history, maintenance records, and performance trends, a multi-objective evolutionary algorithm is used to assess the severity and potential impact of the anomaly from multiple perspectives, including the degree of security threat, reliability impact, functional impact (assessing the scope of functional loss), and economic impact (assessing repair costs and resources). A balance is sought across multiple assessment dimensions, taking into account load conditions, seasonal factors, and the contextual sensitivity of concurrent events. Based on the severity of the assessment, tiered warning information is generated, including emergency warnings, important warnings, general warnings, and attention prompts. Case reasoning based on similar anomalies is retrieved from a historical case library, and rule reasoning based on application domain expert rules is used to provide targeted handling suggestions, including emergency response plans, temporary mitigation plans, fundamental solutions, and prevention strategies. Based on the execution results of the targeted handling suggestions and the actual connection status confirmed on-site, a change source identification and multi-level change confirmation process is established, including planned changes, on-site discovery and fault handling. Graph neural network technology is used to add, modify or delete entity nodes and connection relationships affected by changes using an incremental update strategy. The direct impact, indirect impact and redundancy assessment of network topology changes are re-evaluated. The connection relationship data in the knowledge graph is dynamically updated and the change version history is maintained to achieve dynamic association and anomaly early warning.

9. The method according to claim 1, characterized in that, The modular system built on a microservice architecture includes a data parsing engine, an AI generation engine, an intelligent recognition engine, and a knowledge graph service. It supports data exchange and functional integration with other systems in the smart substation through standardized API interfaces, enabling intelligent management of the entire lifecycle of optical fiber tags from generation and recognition to operation and maintenance. The design includes a mobile terminal equipped with a GPU to support AI inference, a high-definition camera with image stabilization and fill light, an RFID identification module for auxiliary identification, and a hardware system that supports label printing equipment for various materials. By using edge computing technology to optimize the terminal's processing capabilities to support AI inference and image processing, the design ensures the availability of data processing and label identification in weak network or offline environments, resulting in a usable hardware system. Based on the available hardware system, a core module is developed, including a data parsing engine for processing the configuration files of the smart substation using NLP and GNN technologies, an AI generation engine for label optimization and generation using GAN and reinforcement learning technologies, an intelligent recognition engine for multimodal parsing using CV and multimodal technologies, and a knowledge graph service for storing relationships and historical data. Containerization technology is used to achieve loose coupling and flexible deployment between modules, forming a microservice architecture that supports independent development, deployment, and expansion. Based on the aforementioned microservice architecture, a layered data storage strategy is designed for different types of data to achieve separation of hot and cold data. A multi-level security mechanism is constructed, including data encryption to protect sensitive information, access control to restrict operation permissions, and operation audit records for system use. Standardized API interfaces are developed to support data exchange and functional integration with other substation systems such as SCADA systems and asset management systems. A flexible plug-in mechanism that allows third parties to develop extended functions is designed to achieve full lifecycle management.

10. A full lifecycle management system for optical fiber smart tags, characterized in that, include: The data parsing module is used to extract structured data, including descriptions of logical nodes, logical devices, communication connections, and physical devices, based on the intelligent substation configuration files, including SCD and SPCD files. It uses an XML parsing engine to extract structured data, including descriptions of logical nodes, logical devices, communication connections, and physical devices. It uses natural language processing technology to perform semantic analysis on unstructured text and combines the symbolic quantile regression method to quantify text features, and constructs a knowledge graph containing the relationships between devices, ports, optical cables, and virtual terminals. The tag generation module is used to dynamically adjust the text size, QR code position and fault tolerance based on the knowledge graph and generative adversarial network technology according to the optical cable type and the pasting scenario, so as to generate smart optical cable tags that are adapted to different scenarios. The tag recognition module is used to collect tag images of the optical cable on site based on the optical cable smart tag, and use multimodal fusion computer vision technology to convert them into structured tag recognition data containing optical cable type, connection relationship and transmission signal type; The anomaly warning module is used to check the consistency between the tag information and the expected connection relationship by relying on the structured tag recognition data and the knowledge graph through real-time data comparison algorithm, analyze potential anomalies, evaluate the safety impact, reliability impact and functional impact of the anomalies using multi-objective evolution algorithm, and generate graded warning information and targeted handling suggestions. The system integration module is used to build a modular system based on a microservice architecture, including a data parsing engine, an AI generation engine, an intelligent recognition engine, and a knowledge graph service. It supports data exchange and functional integration with other systems in smart substations through standardized API interfaces, enabling intelligent management of the entire lifecycle of optical cable tags from generation and recognition to operation and maintenance.

Images