Heavy haul railway multi-modal power supply data governance and model construction system

Abstract

The application relates to a heavy-load railway multi-modal power supply data management and model construction system. The system comprises a data classification module, which is used for classifying multi-modal power supply original data based on modal types to obtain modal data corresponding to each modal type; a semantic analysis module, which is used for performing format conversion on each modal data to obtain standardized data, performing semantic analysis on each standardized data to obtain semantic labels, and associating the standardized data with the semantic labels; an association alignment module, which is used for associating and aligning each standardized data in multiple standard dimensions to obtain regularized data; and a model construction module, which is used for performing multi-modal feature fusion and semantic association modeling based on each regularized data and the associated semantic labels to obtain a fused semantic feature set, and constructing a business knowledge model based on the fused semantic feature set. The scheme of the application can improve the availability of multi-modal data.

Description

Technical Field

[0001] This application relates to the technical field of data governance, and in particular to a data governance and model building system for multimodal power supply in heavy-haul railways. Background Technology

[0002] With the continuous increase in heavy-haul railway freight volume and the continuous expansion of line coverage, the complexity of power supply systems such as power distribution and overhead contact lines has increased significantly, and the amount of multimodal power supply data (electrical quantity data, equipment status data, environmental perception data, etc.) has surged. This places higher demands on the accuracy of power supply data governance, cross-modal fusion capabilities, and the scientific nature of business model construction.

[0003] In related technologies, the power supply data management of traditional heavy-haul railways often adopts a modular independent storage and simple and organized mode; however, due to inconsistent formats, missing semantics and high redundancy, the usability of multimodal data is insufficient.

[0004] Therefore, how to tap the value of multimodal power supply data and improve its usability is an urgent problem to be solved. Summary of the Invention

[0005] Therefore, it is necessary to provide a data governance and model building system for multimodal power supply in heavy-haul railways that can improve the usability of data, addressing the aforementioned technical issues.

[0006] In a first aspect, this application provides a data governance and model building system for multimodal power supply in heavy-haul railways. The system includes a data classification module, a semantic parsing module, an association and alignment module, and a model building module, wherein:

[0007] The data classification module is used to classify the raw power supply data of multiple modes based on the mode type, and obtain the mode data corresponding to each mode type;

[0008] The semantic parsing module is used to convert the format of each modal data to obtain standardized data, perform semantic parsing on each standardized data to obtain semantic tags, and associate the standardized data with the semantic tags.

[0009] The association and alignment module is used to associate and align the standardized data in multiple standard dimensions to obtain regularized data.

[0010] The model building module is used to perform multimodal feature fusion and semantic association modeling based on the regularized data and the associated semantic tags to obtain a fused semantic feature set, and to build a business knowledge model based on the fused semantic feature set; the business knowledge model is used to build a digital twin model of the power supply system in the target area.

[0011] In one embodiment, the data classification module, when classifying the raw power supply data of multiple modes based on modal type to obtain modal data corresponding to each modal type, is specifically used for:

[0012] Identify structured and unstructured data in the raw power supply data;

[0013] Extract core fields from the structured data and annotate the source of the core fields to obtain structured labeled data;

[0014] The unstructured data is filtered for validity based on its data type to obtain unstructured categorized data.

[0015] The structured labeled data and the unstructured classified data are grouped and integrated according to the modality type to obtain the modality data corresponding to each modality type.

[0016] In one embodiment, the model building module, when performing multimodal feature fusion and semantic association modeling based on the regularized data and the associated semantic tags to obtain a fused semantic feature set, is specifically used for:

[0017] Multimodal feature extraction is performed on each of the regularized data to obtain a basic feature set; the basic feature set includes equipment attribute features, operating status features, fault characterization features, and environmental correlation features;

[0018] The basic feature set is subjected to feature normalization and noise reduction to obtain a purified feature set;

[0019] Based on business relationships, construct a feature association matrix, and generate feature association data based on the feature association matrix;

[0020] The feature-related data is fused with the purification feature set to obtain a preliminary fused feature set;

[0021] Redundancy is removed from the preliminary fused feature set to obtain the fused semantic feature set.

[0022] In one embodiment, the model building module, when building a business knowledge model based on the fused semantic feature set, is specifically used for:

[0023] Element extraction is performed on the fused semantic feature set to obtain multiple core knowledge elements;

[0024] The core knowledge elements are classified to determine the business knowledge elements corresponding to each business type.

[0025] Based on the business knowledge elements corresponding to each of the aforementioned business types, as well as the hierarchical relationships, logical rules, and associations among the knowledge elements, knowledge modeling is performed to obtain the knowledge model corresponding to each of the aforementioned business types.

[0026] The business knowledge model is constructed by integrating the knowledge models corresponding to each of the aforementioned business types.

[0027] In one embodiment, the model building module, when building a digital twin model of the power supply system within the target area based on the business knowledge model, is specifically used for:

[0028] Obtain geographical distribution information, equipment ledger information, and basic data of each device of the power supply system within the target area;

[0029] From the business knowledge model, attribute data of each device is extracted, and knowledge mapping data of each device is generated based on the attribute data of each device; the attribute data includes entity attributes, structural features, and business association rules.

[0030] The basic data of the same device is associated with the knowledge mapping data to generate device knowledge matching data for each device;

[0031] Based on the device knowledge matching data of each device, a three-dimensional geometric framework model of the power supply system in the target area is constructed.

[0032] The digital twin model of the power supply system within the target area is obtained by configuring the parameters of the three-dimensional geometric framework model according to the business knowledge model; the parameter configuration includes the configuration of the business knowledge elements and the configuration of the logical rules.

[0033] In one embodiment, the model building module constructs a three-dimensional geometric framework model of the power supply system within the target area based on the device knowledge matching data of each device, specifically for:

[0034] Extract the geometric structure information of each device from the device knowledge matching data of each device;

[0035] The geometric structure information of each device is parsed to determine the geometric constraint data of each device; the geometric constraint data includes the structure type, spatial assembly logic, and positional constraint relationship with adjacent devices.

[0036] Based on the geometric constraint data of each device, a three-dimensional sub-model of each device is generated;

[0037] Based on the equipment knowledge matching data of each device, the three-dimensional sub-models of each device are assembled to obtain a three-dimensional geometric framework model of the power supply system in the target area.

[0038] In one embodiment, when the model building module configures the parameters of the three-dimensional geometric framework model according to the business knowledge model to obtain the digital twin model of the power supply system in the target area, it is specifically used for:

[0039] From the business knowledge model, knowledge configuration data for each device in the power supply system within the target area is extracted; the knowledge configuration data includes business knowledge elements corresponding to device attributes, operating logic rules, and fault judgment criteria.

[0040] From the three-dimensional geometric framework model, extract the entity identification data of each device. The entity identification data includes the device's identification information, geometric feature parameters, and spatial location information.

[0041] The knowledge configuration data of each device is associated with the entity identification data, and the knowledge configuration data is filled into the entity nodes of the corresponding devices in the three-dimensional geometric frame model to obtain the parameter configuration model;

[0042] The parameter configuration model is instantiated and encapsulated to generate the instantiated digital twin model.

[0043] In one embodiment, the system further includes an update module for:

[0044] Based on the aforementioned business knowledge model, semantic parsing, feature extraction, and state recognition are performed on the target power supply data to obtain state parsing data; the target power supply data is multimodal power supply data collected for the power supply system of the target area.

[0045] Determine the target device corresponding to the state parsing data in the digital twin model;

[0046] The state parsing data is used to update the state of the entity node corresponding to the target device in the digital twin model to obtain the updated digital twin model;

[0047] The updated digital twin model is subjected to operational logic deduction to generate state deduction data of associated devices, and the state deduction data is associated with the update information of the digital twin model.

[0048] In one embodiment, the business knowledge model includes state determination rule knowledge and semantic rule knowledge; the update module, when performing semantic parsing, feature extraction, and state recognition on the target power supply data based on the business knowledge model to obtain state parsing data, is specifically used for:

[0049] Using the semantic rule knowledge, modal classification and semantic annotation are performed on the target power supply data to generate semantically annotated data;

[0050] Feature extraction is performed on the semantically labeled data to obtain real-time feature data;

[0051] The state determination rules are used to compare and analyze the real-time feature data to obtain the state recognition result;

[0052] The status recognition result is associated with the corresponding device identifier, time information and spatial location information to obtain the status parsing data.

[0053] In a second aspect, this application also provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps of the system described in any of the first aspects above.

[0054] The aforementioned system for multimodal power supply data governance and model building in heavy-haul railways comprehensively collects raw multimodal power supply data related to power distribution and overhead contact lines in heavy-haul railways. It then performs modal identification and classification processing on the collected raw data, generating a modal classification dataset containing modal data of various modal types. This system organizes and categorizes the disorganized raw data according to modal type, avoiding management chaos caused by mixed storage of different modal data. Secondly, addressing the format differences between different modal data, the system performs a unified cross-modal format conversion on the modal classification dataset, unifying the encoding rules, storage formats, and statistical standards for various types of data. This eliminates format barriers between different modal data, providing a foundation for comparison and integration of previously heterogeneous multimodal data, overcoming the limitations of traditional multimodal data analysis. Simultaneously, the system performs semantic parsing on the unified format data, uncovering the business meaning behind various data types, adding corresponding semantic tags to the data, clarifying the relationship between the data and power supply services, and generating standardized data with a unified format and semantic tags. This ensures that the data not only has a regular format but also clear business semantics. Then, by performing multi-dimensional correlation and alignment on standardized data, and based on semantic tags and business logic, an intrinsic relationship between different modal data is established, resulting in high-quality, well-organized data. This allows the originally independent multimodal data to form an organic whole, breaking the limitation of traditional submodal data lacking deep correlation.

[0055] Secondly, by conducting multimodal feature fusion and semantic association modeling based on high-quality, well-organized data, the core features of different modalities are integrated, and deep semantic relationships between various data types are mined. The resulting fused semantic feature set fully presents the operating rules of the power supply system, equipment status changes, and the influence mechanisms of various factors, breaking through the limitations of traditional single-modal feature analysis. Based on the fused semantic feature set, a business knowledge model for the power supply field is constructed, deeply integrating data features with power supply business scenarios and operation and maintenance experience to form a systematic business model that fits the actual power supply situation of heavy-haul railways. This avoids the drawbacks of traditional business models lacking data support and being disconnected from actual business operations. Attached Figure Description

[0056] To more clearly illustrate the technical solutions in the embodiments of this application or related technologies, the drawings used in the description of the embodiments of this application or related technologies will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0057] Figure 1 This is a schematic diagram of the architecture of a multimodal power supply data governance and model building system for heavy-haul railways in one embodiment;

[0058] Figure 2 This is a schematic diagram of the execution flow of the data classification module in one embodiment;

[0059] Figure 3 This is a flowchart illustrating the process of constructing a fused semantic feature set in one embodiment;

[0060] Figure 4 This is a schematic diagram of the process of building a business knowledge model in one embodiment;

[0061] Figure 5 This is a schematic diagram illustrating the process of constructing a digital twin model in one embodiment;

[0062] Figure 6 This is a schematic diagram illustrating the process of constructing a three-dimensional geometric framework model in one embodiment;

[0063] Figure 7 This is a flowchart illustrating parameter configuration in one embodiment;

[0064] Figure 8 This is a schematic diagram of the execution flow of the update module in one embodiment;

[0065] Figure 9 This is a flowchart illustrating the process of determining state parsing data in one embodiment;

[0066] Figure 10This is an internal structural diagram of a computer device in one embodiment. Detailed Implementation

[0067] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0068] It should be noted that the terms "first," "second," etc., used in this application can be used to describe various elements, but these elements are not limited by these terms. These terms are only used to distinguish the first element from the second element. The terms "comprising" and "having," and any variations thereof, used in this application, are intended to cover non-exclusive inclusion. The term "multiple" used in this application refers to two or more. The term "and / or" used in this application refers to one of the embodiments, or any combination of multiple embodiments.

[0069] The heavy-haul railway multimodal power supply data governance and model building system provided in this application embodiment can be configured on a computer device. The computer device can be, but is not limited to, various personal computers, laptops, and servers; the server can be an independent physical server, a server cluster or distributed system composed of multiple physical servers, or a cloud server providing cloud computing services.

[0070] In one exemplary embodiment, such as Figure 1 As shown, a data governance and model building system for multimodal power supply in heavy-haul railways is provided. This system specifically includes a data classification module 10, a semantic parsing module 20, an association and alignment module 30, and a model building module 40, wherein:

[0071] The data classification module 10 is used to classify the raw power supply data of multiple modes based on the mode type, and obtain the mode data corresponding to each mode type.

[0072] In this embodiment, a modal classification dataset is generated by initiating a multimodal power supply raw data acquisition and classification processing flow. During the data acquisition process, a fixed acquisition algorithm and acquisition link are used to comprehensively collect multimodal power supply raw data related to power distribution and overhead contact lines in heavy-haul railways. This includes structured data such as power distribution equipment ledgers and operation monitoring parameters; unstructured image and video data such as overhead contact line detection images and substation monitoring videos; unstructured text data such as power supply technical documents and production management procedures; and environmental data such as ambient temperature, humidity, and wind force. All collected data are labeled with the acquisition time, acquisition location, and acquisition terminal identifier.

[0073] After data collection, a fixed modality recognition algorithm was used to extract the core features of various types of raw power supply data, thereby distinguishing between structured and unstructured modalities. Structured data was classified based on fixed fields and quantifiable features, while unstructured data was classified based on the absence of a fixed format and features that could not be directly quantified. A fixed classification algorithm was then used to divide structured data into two subcategories: equipment ledgers and operating parameters, and unstructured data into three subcategories: images, videos, and text. Each data category was labeled with a modality identifier and category label, and invalid and damaged data were removed to complete the classification, resulting in modal data corresponding to each modality type. All classified data was then integrated to generate a clearly structured and well-defined modality classification dataset, laying the foundation for subsequent format standardization and semantic analysis.

[0074] The semantic parsing module 20 is used to convert the format of each modal data to obtain standardized data, perform semantic parsing on each standardized data to obtain semantic tags, and associate the standardized data with the semantic tags.

[0075] In this embodiment, based on the modality classification dataset, a cross-modality format unification conversion and semantic parsing process is initiated to generate standardized data with unified format and semantic labels. During the format conversion process, a fixed cross-modality format conversion algorithm is used to perform targeted format conversion for data of different modalities and categories. This converts equipment ledgers and operating parameters in structured data into a preset structured format, standardizing field names and data types. Unstructured image data is converted into image files with unified resolution and encoding format; video data is converted into video files with unified frame rate and encoding; and text data is converted into plain text format with unified character encoding, thus eliminating format differences between different modalities.

[0076] After converting the format of each data in the modality classification dataset to obtain standardized data, a fixed semantic parsing algorithm is used. Based on the semantic rules of the power supply field preset by the system, semantic parsing is performed on the converted standardized data to add accurate semantic labels to each type of standardized data. For example, standardized data of equipment ledgers are labeled with equipment identification and parameter meaning, standardized data of operating parameters are labeled with parameter type and operating indicators, standardized data of images and videos are labeled with shooting location, equipment components and operating status, and standardized data of text are labeled with core semantics and business scenarios.

[0077] After completing semantic parsing, the format uniformity and semantic accuracy of various standardized data are checked, and format deviations and semantic annotation errors are corrected. All data are further integrated to generate standardized data with uniform format, clear semantics, and complete labels, providing support for subsequent association alignment and redundancy cleaning.

[0078] The correlation and alignment module 30 is used to correlate and align standardized data across multiple standard dimensions to obtain regularized data.

[0079] In this embodiment, a multi-dimensional association alignment and redundancy cleansing process is initiated for various types of standardized data to generate high-quality, well-organized data. During the data association alignment process, a multi-dimensional association alignment algorithm is employed to integrate standardized data from four standard dimensions: device identification, acquisition time, spatial location, and business scenario. Association integration means linking multimodal data collected from the same device, at the same time, and at the same location one-to-one, establishing inherent relationships between the data. For example, the operating parameters, detection images, and environmental data of a certain overhead contact line support are linked and bound together to clarify the business relationship logic between the data.

[0080] After the data association and alignment are completed, a redundancy cleaning algorithm is used to thoroughly clean the standardized data after association. The data cleanliness is divided into three dimensions: removing duplicate data, removing invalid data, and removing abnormal data.

[0081] Remove duplicate data: Delete duplicate parameters and images collected by the same device at the same time; Remove invalid data: Delete blurry images, videos with signal interruptions, and incomplete text and ledgers; Remove abnormal data: Replace abnormal values ​​in operating parameters with the average of adjacent data, and delete redundant information unrelated to power supply services.

[0082] After clarifying the data, verify the correlation and completeness of the data, correct correlation deviations and clean up omissions, supplement missing core data, classify and organize the data according to equipment type and collection time, and generate high-quality, well-organized data that is closely related, complete, non-redundant and free of anomalies, providing core data support for subsequent feature fusion and modeling.

[0083] The model building module 40 is used to perform multimodal feature fusion and semantic association modeling based on various regularized data and associated semantic tags, to obtain a fused semantic feature set, and to build a business knowledge model based on the fused semantic feature set.

[0084] In this embodiment, high-quality, well-organized data is used to initiate multimodal feature fusion and semantic association modeling to construct a business knowledge model for the power supply field. During the feature fusion and semantic association modeling process, a multimodal feature fusion algorithm is employed to extract multi-dimensional basic features such as equipment attributes, operating status, fault characterization, and environmental associations from the high-quality, well-organized data. After normalization and denoising, a cross-modal semantic association algorithm is used to mine the inherent business associations between features of different modalities and perform deep fusion, resulting in a fused semantic feature set that takes into account both multimodal information and business logic.

[0085] In the process of semantic association modeling, knowledge modeling algorithms are used to process the fused semantic feature set and abstract core knowledge elements such as equipment entities, business processes, operational constraints, and relationships in the power supply field. After classifying and sorting the core knowledge elements, knowledge sub-modules such as equipment entities, operation control, and fault handling are constructed. After further integrating the knowledge sub-modules, a business knowledge model for the power supply field is generated.

[0086] Business knowledge models can be used to construct digital twin models of power supply systems within a target area. In the process of constructing a digital twin model of the power supply system in a target area, the geographical distribution and ledger information of the overhead contact line power supply equipment within the target area are obtained. The business knowledge model is then matched with actual equipment data, and parametric modeling technology is used to construct a three-dimensional geometric framework for each piece of equipment in the power supply system within the target area. Business knowledge elements are then associated and bound to the three-dimensional geometric framework, completing the instantiation of the digital twin model. Furthermore, a real-time interface channel is established between the model and a multimodal data acquisition terminal, mapping real-time power supply data to the corresponding nodes of each piece of equipment in the digital twin model. This achieves dynamic mapping between the model and the actual equipment, ensuring that the model state is synchronized with the actual equipment state, and providing model support for power supply equipment operation and maintenance and fault early warning.

[0087] The aforementioned system for multimodal power supply data governance and model building in heavy-haul railways comprehensively collects raw multimodal power supply data related to power distribution and overhead contact lines in heavy-haul railways. It then performs modal identification and classification processing on the collected raw data, generating a modal classification dataset containing modal data of various modal types. This system organizes and categorizes the disorganized raw data according to modal type, avoiding management chaos caused by mixed storage of different modal data. Secondly, addressing the format differences between different modal data, the system performs a unified cross-modal format conversion on the modal classification dataset, unifying the encoding rules, storage formats, and statistical standards for various types of data. This eliminates format barriers between different modal data, providing a foundation for comparison and integration of previously heterogeneous multimodal data, overcoming the limitations of traditional multimodal data analysis. Simultaneously, semantic parsing is performed on the unified format data to uncover the business meaning behind various data types. Corresponding semantic tags are added to the data, clarifying the relationship between the data and power supply services. The resulting standardized data, with a unified format and semantic tags, not only possesses a regular format but also clear business semantics. Then, by performing multi-dimensional correlation and alignment on standardized data, and based on semantic tags and business logic, an intrinsic relationship between different modal data is established, resulting in high-quality, well-organized data. This allows the originally independent multimodal data to form an organic whole, breaking the limitation of traditional submodal data lacking deep correlation.

[0088] Secondly, by conducting multimodal feature fusion and semantic association modeling based on high-quality, well-organized data, the core features of different modalities are integrated, and deep semantic relationships between various data types are mined. The resulting fused semantic feature set fully presents the operating rules of the power supply system, equipment status changes, and the influence mechanisms of various factors, breaking through the limitations of traditional single-modal feature analysis. Based on the fused semantic feature set, a business knowledge model for the power supply field is constructed, deeply integrating data features with power supply business scenarios and operation and maintenance experience to form a systematic business model that fits the actual power supply situation of heavy-haul railways. This avoids the drawbacks of traditional business models lacking data support and being disconnected from actual business operations.

[0089] In one embodiment, reference Figure 2 The data classification module 10, in classifying the raw power supply data of multiple modes based on modal type to obtain the modal data corresponding to each modal type, specifically includes steps 11-13, wherein:

[0090] Step 11: Identify structured and unstructured data in the raw power supply data.

[0091] Specifically, during the data collection process, acquisition algorithms and acquisition links are employed. For the heavy-haul railway power distribution equipment ledger, core information such as equipment model, installation location, commissioning time, maintenance records, and parameter specifications of the power distribution equipment is collected through the acquisition interface of the equipment management terminal. For the catenary operation monitoring data, operating parameters such as catenary conductor height, pull-out value, wear amount, temperature, and vibration frequency are collected in real time through the acquisition module of the catenary monitoring terminal. For catenary detection images and videos, images and real-time operating videos of key components such as catenary conductors, insulators, and cantilever arms are captured at fixed intervals through high-definition acquisition equipment deployed along the line, and the shooting time and shooting location are recorded. For substation monitoring videos, video data such as the operating status of equipment and personnel operations within the substation are continuously collected through monitoring acquisition equipment deployed within the substation. For power supply technical documents, text data such as power supply equipment operation and maintenance manuals, fault handling plans, and technical transformation reports are collected through the document acquisition interface. For production management procedures, text data such as heavy-haul railway power supply safety regulations, equipment operation specifications, and operation and maintenance management details are collected. All collected data is stored in its original format, recording the collection time, collection location, and collection terminal identification to ensure that the original data of multimodal power supply is complete and traceable, laying the foundation for subsequent data mode determination and classification processing.

[0092] For the collected multimodal power supply raw data, a modality type determination process is initiated to clearly distinguish between structured and unstructured data. During the modality type determination process, a modality determination algorithm is used to first extract the core features of each type of raw power supply data. Structured data features are defined as having fixed fields, a fixed format, and being directly quantifiable, presenting the data in a regular table or field format. Unstructured data features are defined as having no fixed fields, no uniform format, and being unable to be directly quantified, presenting the data in a disordered and diverse form. For example, the ledgers of heavy-haul railway substation equipment and the catenary operation monitoring data are structured data. The substation equipment ledgers contain fixed fields such as equipment identification and parameter specifications, while the catenary operation monitoring data contains quantifiable fields such as conductor height and wear, both conforming to the characteristics of structured data. However, images and videos have no fixed fields, and technical documents and management procedures are in plain text format with no uniform quantifiable format, both conforming to the characteristics of unstructured data. Therefore, the catenary inspection images and videos, substation monitoring videos, power supply technical documents, and production management procedures are unstructured data.

[0093] Based on the extraction results of the core features of the raw power supply data, the modal type of each type of raw power supply data is determined one by one. After the modal type determination is completed, the modal type identifier is marked for each type of raw power supply data. The raw power supply data is initially classified according to the modal type, and a modal determination result table is generated to clarify the modal type of each type of raw power supply data.

[0094] Step 12: Extract core fields from structured data and annotate the source of the core fields to obtain structured labeled data.

[0095] Specifically, a field extraction algorithm is used to extract pre-defined core fields from structured data such as power distribution equipment ledgers. These core fields include fixed fields such as equipment identification, equipment model, installation location, commissioning time, maintenance cycle, maintenance records, fault records, and parameter specifications. Redundant and invalid information is then removed from the extracted core fields. For example, for structured data such as overhead contact line operation monitoring data, core fields such as conductor height, pull-out value, wear amount, temperature, vibration frequency, monitoring time, monitoring location, and monitoring terminal identification are extracted, and the field data is organized in chronological order of monitoring time.

[0096] After extracting the core fields, each piece of structured data is clearly labeled with its source information. Specifically, the source for the power distribution equipment ledger is the equipment management terminal and corresponding substation identifier, while the source for the overhead contact line operation monitoring data is the corresponding monitoring terminal and monitoring section identifier. The labeling information is linked one-to-one with the data fields, ensuring that each piece of data can be traced back to its original collection source. The extracted field data and source labeling information are then integrated and categorized by data type to generate well-structured, clearly defined, and clearly sourced structured labeled data.

[0097] Step 13: Filter the unstructured data based on its data type to obtain unstructured categorized data.

[0098] Specifically, unstructured data is divided into three main categories based on its data format: image-based unstructured data, text-based unstructured data, and video-based unstructured data. For example, overhead contact line inspection images are image-based unstructured data; overhead contact line inspection videos and substation monitoring videos are video-based unstructured data; power supply technical documents and production management procedures are text-based unstructured data; and each type of unstructured data is assigned a unique type identifier.

[0099] After classifying the unstructured data, a content screening algorithm is used to perform targeted screening of each type of unstructured data based on its type. For example, for image-based unstructured data, blurry, obscured, and non-critical images are removed, while images clearly showing key components such as contact wires and insulators are retained. For video-based unstructured data, blurry, interrupted, and non-functional videos are removed, while continuous and clear monitoring and inspection videos are retained. For text-based unstructured data, repetitive, incomplete, and irrelevant text is removed, while complete technical content and management specifications are retained. After the content screening is completed, each type of unstructured data is labeled with a type identifier, screening results, and data source. The data is then categorized and organized by type, and all invalid data is removed to ensure the effectiveness and relevance of the unstructured data classification.

[0100] Step 14: Group and integrate the structured labeled data and unstructured classified data according to modality type to obtain the modality data corresponding to each modality type.

[0101] Specifically, in the data integration process, a group integration approach is adopted. First, the data is divided into two main groups according to modality: structured data and unstructured data. The structured data group contains only structured labeled data, which is further divided into a subgroup of power distribution equipment ledger data and a subgroup of catenary operation monitoring data according to the data source. The unstructured data group contains unstructured classified data, which is further divided into three subgroups according to data type: image, video, and text. Each subgroup is further classified and collected according to the collection source and collection time.

[0102] After grouping, an integration algorithm is used to link and integrate data of the same modality and subgroup, supplementing information such as data collection time, collection location, and source identifier to ensure logical coherence and clear correlation among the data groups. After integration, all data undergoes standardized formatting, labeling data modality type, data subgroup type, source information, and collection time. Data integrity and correlation are verified, duplicate and invalid data are removed, and a well-structured, clearly categorized, and complete modality classification dataset is generated.

[0103] In one embodiment, reference Figure 3 The model building module 40, when performing multimodal feature fusion and semantic association modeling based on various regularized data and associated semantic labels to obtain a fused semantic feature set, specifically includes steps 411-415, wherein:

[0104] Step 411: Perform multimodal feature extraction on each regularized data to obtain a basic feature set; the basic feature set includes equipment attribute features, operating status features, fault characterization features, and environmental correlation features.

[0105] Specifically, in the feature extraction process, a feature extraction algorithm is used to extract features from the regular data one by one according to four preset dimensions; the four dimensions include equipment attribute dimension, operating status dimension, fault characterization dimension, and environmental correlation dimension.

[0106] For the equipment attribute dimension, basic features such as the model, parameter specifications, installation location, commissioning time, and maintenance cycle of the power distribution equipment are extracted and directly extracted from the structured and regularized data and converted into feature vectors. For the operating status dimension, basic features such as the contact wire conductor height, pull-out value, wear amount, temperature, operating voltage and current of the power distribution equipment are extracted. Among them, the equipment operating status information is comprehensively extracted by combining structured monitoring data and unstructured video images.

[0107] For the fault characterization dimension, basic features such as contact wire damage, insulator aging, equipment noise, and abnormal parameters are extracted. Fault features can be identified from images and videos, abnormal parameter features can be extracted from monitoring data, and fault description features can be extracted from text data. For the environmental correlation dimension, basic features such as temperature, humidity, rain, snow, and wind force at the collection location are extracted, which can be combined with monitoring data and text records for comprehensive extraction. After feature extraction, the basic features of the four dimensions are organized by category, labeled with feature source, modal type, and business meaning, and invalid features are removed, thereby generating a basic feature set covering multiple dimensions of equipment, operation, fault, and environment.

[0108] Step 412: Perform feature normalization and noise reduction on the basic feature set to obtain a purified feature set.

[0109] Specifically, a normalization algorithm is used to standardize all features in the basic feature set, uniformly converting the numerical ranges of various basic features such as equipment attributes, operating status, fault characterization, and environmental associations to a fixed interval. Equipment parameter features can be scaled proportionally, while fault characterization features, being binary features, can be encoded according to unified rules to eliminate heterogeneity interference caused by differences in numerical ranges between different modal features. After normalization, a denoising algorithm is used to denoise the normalized basic features. For structured features, abnormal numerical features caused by monitoring errors are removed, and the average of adjacent features is used to replace outliers. For features derived from unstructured features, ambiguous and highly interfering features are removed, retaining only valid features. For example, for text-derived features, redundant and irrelevant feature words are removed. After denoising, the integrity and consistency of the features are checked, and deviations in the processing are corrected to ensure that all features are free of interference and anomalies, resulting in a clean feature set with a well-structured and pure set, which includes multiple clean features.

[0110] Step 413: Construct a feature association matrix based on business relationships, and generate feature association data based on the feature association matrix.

[0111] Specifically, for the purification feature set, a process for mining the intrinsic business associations of different modal features is initiated to construct a feature association matrix and generate feature association data. During the mining process, a cross-modal semantic association algorithm is used to first analyze the business semantics of various modal features, clarifying the intrinsic business relationships between the four basic dimensions of equipment attributes, operating status, fault characterization, and environmental association; for example, equipment model features are directly associated with operating parameter features, environmental temperature and humidity features are associated with equipment fault characterization features, and contact wire wear features are associated with conductor status features in the image.

[0112] Furthermore, based on the semantic parsing results, the correlation strength between basic features of different modalities is quantified. Correlation strength is assigned a value according to a fixed scoring standard: the stronger the correlation, the higher the score; no correlation results in a score of zero. A feature correlation matrix is ​​then constructed, with rows and columns representing various cleanup features, and matrix elements representing the correlation strength score between corresponding features. The correlation type and business logic basis for each pair of features are clearly labeled. The feature correlation matrix and correlation strength data are integrated, supplemented with feature semantic descriptions and correlation logic, to generate complete feature correlation data.

[0113] Step 414: Perform feature fusion between the feature association data and the cleaned feature set to obtain a preliminary fused feature set.

[0114] Specifically, for feature-related data and the cleaned feature set, a deep feature fusion process is initiated to form preliminary fused features that take into account both multimodal information and business logic. During the fusion process, a feature fusion algorithm is used, based on feature-related data, to integrate various features in the cleaned feature set according to the priority of correlation strength. Features with high correlation strength are assigned higher fusion weights, and features with low correlation strength are assigned corresponding weights, ensuring that the fused features align with business logic. Specifically, during fusion, equipment attribute features and corresponding operating status features are first initially integrated to form basic fused features for equipment operation; fault characterization features are then further integrated with environmental correlation features and basic fused features for equipment operation to explore the impact of environmental factors on equipment faults; unstructured features derived from images and videos are cross-modal integrated with structured features to supplement visual feature information. During the fusion process, the core information of various modal features is retained, feature redundancy and conflicts are eliminated, ensuring that the fused features not only cover multimodal data information but also conform to the business logic of heavy-haul railway power supply. After fusion, preliminary fused features are generated.

[0115] Step 415: Redundancy removal is performed on the preliminary fused feature set to obtain the fused semantic feature set.

[0116] Specifically, for the initial fusion features, a core feature screening process is initiated to remove redundant features irrelevant to power supply business, resulting in a fusion semantic feature set. First, feature screening criteria are set, focusing on three core scenarios: equipment operation and maintenance, fault early warning, and operation monitoring. Only features related to these three scenarios are retained, while features irrelevant to power supply business are removed.

[0117] During the screening process, a screening algorithm is employed to verify each feature in the initial fusion feature set one by one. Core business features such as equipment parameters, operating status, fault characteristics, and environmental correlations are retained, while redundant features unrelated to the operation, faults, and maintenance of power supply equipment are eliminated. For example, features derived from parameters irrelevant to the data acquisition terminal and environmental features unrelated to the power supply scenario are removed. After screening, the core features retained after screening are semantically correlated and integrated. The business semantics and corresponding application scenarios of each type of feature are labeled, the correlation and completeness between features are verified, screening biases are corrected, and it is ensured that all features are consistent with the business needs of the power supply field and are free of redundancy. All core features retained after screening are integrated to generate a fused semantic feature set that is structurally standardized, semantically clear, and business-aligned.

[0118] In one embodiment, reference Figure 4 The model building module 40, when constructing a business knowledge model based on the fused semantic feature set, specifically includes steps 421-424, wherein:

[0119] Step 421: Extract elements from the fused semantic feature set to obtain multiple core knowledge elements.

[0120] Specifically, a knowledge abstraction algorithm is used to extract the core knowledge elements of four business types one by one from the fused semantic feature set; the four business types include equipment entity elements, business process elements, operational constraint elements, and relational elements.

[0121] The equipment entity elements focus on core equipment such as power distribution equipment and overhead contact line equipment, abstracting core information such as equipment model, parameter specifications, installation location, and component composition; the business process elements revolve around core businesses such as power supply equipment operation and maintenance, fault handling, and operation monitoring, abstracting key content such as process steps, operating standards, and execution sequence; the operation constraint elements combine the operating rules of power supply equipment, abstracting constraints such as equipment operating parameter thresholds, environmental adaptability range, and safe operation requirements; the relationship elements explore the inherent connections between equipment, between equipment and business, and between business and environment, abstracting related information such as equipment linkage relationships, business connection relationships, and environmental impact relationships.

[0122] After extracting various core knowledge elements, all core knowledge elements are preprocessed. Preprocessing includes, but is not limited to, labeling the source of the elements, their business meaning and corresponding feature associations, eliminating invalid and redundant knowledge elements, and classifying and organizing them according to element type, thereby generating a domain knowledge element library covering equipment, business, constraints and associations.

[0123] Step 422: Classify multiple core knowledge elements and determine the business knowledge elements corresponding to each business type.

[0124] Specifically, the domain knowledge element base is classified and sorted out, and knowledge sub-modules corresponding to equipment entity knowledge, operation and control knowledge, fault handling knowledge, safety management knowledge, and environmental adaptation knowledge are divided, generating knowledge module division results.

[0125] During the classification process, a classification and sorting algorithm is adopted. First, the core business modules in the field of heavy-haul railway power supply are identified. Then, the business attributes and application scenarios of knowledge elements are combined to classify all core knowledge elements in the domain knowledge element library.

[0126] The core knowledge elements related to the physical equipment are integrated into an equipment entity knowledge submodule, which includes elements such as equipment parameters, component composition, and installation information. Knowledge elements related to equipment operation monitoring and maintenance management are integrated into an operation control knowledge submodule, which includes elements such as operation monitoring standards, maintenance procedures, and parameter adjustment. Knowledge elements related to fault identification and handling are integrated into a fault handling knowledge submodule, which includes elements such as fault characteristics, handling steps, and repair standards. Knowledge elements related to safe operation and risk prevention are integrated into a safety management knowledge submodule, which includes elements such as safety procedures, risk points, and prevention and control measures. Knowledge elements related to the impact of environmental factors on power supply equipment are integrated into an environmental adaptation knowledge submodule, which includes elements such as environmental parameter requirements and adaptation adjustment measures.

[0127] After classification, each knowledge submodule is labeled with a module identifier, core elements, and business scope. The preliminary relationships between modules are further clarified, and a clear knowledge module division result is generated, providing a basis for subsequent submodule knowledge modeling.

[0128] Step 423: Based on the business knowledge elements corresponding to each business type, as well as the hierarchical relationships, logical rules and associations between the knowledge elements, perform knowledge modeling to obtain the knowledge model corresponding to each business type.

[0129] Specifically, knowledge modeling is performed on each knowledge sub-module to construct the hierarchical relationships, logical rules, and association links of knowledge elements within the module, thereby generating a knowledge model for the sub-module.

[0130] In this embodiment of the invention, knowledge modeling is performed on each knowledge submodule to construct a complete relational structure among the knowledge elements within each submodule. During the modeling process, a knowledge modeling algorithm is employed. For each knowledge submodule, the hierarchical relationship of the core knowledge elements within the module is first clarified. For each knowledge submodule, a hierarchical structure of core element-secondary element-related element is established, defining the subordinate relationships of each level of elements. For example, in the equipment entity knowledge submodule, the equipment model is the core element, parameter specifications and installation location are secondary elements, and component composition and maintenance cycle are related elements.

[0131] Further develop logical rules among the core knowledge elements within each knowledge submodule, and clarify the constraints and triggering logic between the elements; for example, in the operation control knowledge submodule, set logical rules to trigger monitoring and early warning when the equipment operating parameters exceed the threshold; in the fault handling knowledge submodule, set corresponding logical rules for fault characterization and handling steps.

[0132] Further, the connections between the core knowledge elements within each knowledge submodule are established, linking related core knowledge elements one by one to form a complete knowledge model within the module. Upon completion of modeling each submodule, the rationality of the hierarchical relationships, the rigor of the logical rules, and the completeness of the connection links are verified, and modeling deviations are corrected.

[0133] Step 424: Integrate the knowledge models corresponding to each business type to construct a business knowledge model.

[0134] Specifically, knowledge graph technology is used to integrate the knowledge models of each sub-module, establishing knowledge association interfaces and linkage logic between modules to generate a preliminary business knowledge model. During the integration process, a knowledge graph construction algorithm is used to build a unified knowledge graph framework based on the knowledge models corresponding to each sub-module, with each knowledge model serving as a core node. Furthermore, knowledge association interfaces between modules are established to clarify the relationships between each sub-module. For example, the equipment entity knowledge sub-module and the operation control knowledge sub-module need to establish an association interface through equipment parameter elements, and the fault handling knowledge sub-module and the environment adaptation knowledge sub-module need to establish an association interface through environmental impact elements.

[0135] Further develop the linkage logic between various knowledge models to ensure that when any core knowledge element in a knowledge model changes, the related elements in other knowledge models will also change accordingly. For example, after the equipment parameters are updated in the equipment entity's knowledge model, the monitoring threshold in the operation and management knowledge model can be adjusted synchronously, and the repair standards in the fault handling knowledge model can also be optimized accordingly.

[0136] After integration, the logical structure of the entire knowledge graph is sorted out to ensure smooth connections and reasonable linkage between knowledge models, ultimately resulting in a preliminary business knowledge model that covers all knowledge models.

[0137] Furthermore, the preliminary business knowledge model is validated to correct deviations in knowledge elements, supplement missing relationships, and optimize the rationality of logical links, thereby generating a standardized business knowledge model for the power supply field.

[0138] Specifically, for the preliminary business knowledge model, a knowledge verification and model optimization process is initiated to generate a standardized business knowledge model for the power supply field. During the verification process, a knowledge verification algorithm is used to check each knowledge element, hierarchical relationship, logical rule, and related link in the preliminary business knowledge model, identifying issues such as knowledge element deviations, missing relationships, and unreasonable logical links. For knowledge element deviations, the element information is corrected by comparing and integrating the semantic feature set with actual power supply business practices; for missing relationships, the interfaces and linkage logic between modules and elements are supplemented; for unreasonable logical links, the link paths and triggering rules are optimized to ensure that the logic conforms to the business processes in the power supply field.

[0139] After verification, the preliminary business knowledge model is standardized, unifying the description of knowledge elements, hierarchical division standards, and format of related links. The business basis for each knowledge element and logical rule is labeled, and the application scenarios and usage specifications of the model are clarified. Finally, the completeness, rigor, and standardization of the model are checked, and deviations are corrected and optimized to obtain a standardized, logically rigorous, business-aligned, and uniformly standardized business knowledge model for the power supply field.

[0140] In one embodiment, reference Figure 5 The model building module 40, when building a digital twin model of the power supply system within the target area based on the business knowledge model, specifically includes steps 510-550, wherein:

[0141] Step 510: Obtain the geographical distribution information, equipment ledger information, and basic data of each piece of equipment of the power supply system within the target area.

[0142] Specifically, during the data acquisition process, acquisition algorithms and acquisition links are employed. For the geographical distribution information of the overhead contact system within the target area, the geographic information acquisition module captures spatial coordinate information such as longitude, latitude, and altitude of various equipment, including overhead contact system supports and suspension devices, within the section, recording the spatial distance and relative positional relationships between the equipment. For the overhead contact system power supply equipment ledger information of a specified section, high-quality standardized data output by the association, alignment, and standardization module is retrieved to extract core ledger information such as equipment identification, equipment model, installation time, parameter specifications, maintenance records, and the section to which it belongs. After data acquisition, an integration algorithm is used to associate the geographical distribution information with the equipment ledger information one by one, binding a unique spatial coordinate and ledger information to each piece of equipment, and marking the data acquisition time and source, eliminating invalid data with missing spatial coordinates or ledger information. Furthermore, the collected data is classified and organized according to equipment type to generate basic equipment data covering all overhead contact system power supply equipment within the target area, with complete information and clear associations.

[0143] Step 520: Extract attribute data for each device from the business knowledge model, and generate knowledge mapping data for each device based on the attribute data; the attribute data includes entity attributes, structural features, and business association rules.

[0144] Specifically, based on the business knowledge model in the power supply field, the entity attributes, structural features and business association rules of various equipment such as contact network support foundations, contact suspensions, support devices, and auxiliary equipment in the target area are extracted to generate equipment knowledge mapping data.

[0145] For the business knowledge model in the power supply field, a data extraction process for knowledge mapping of various equipment in the target area's overhead contact system is initiated to ensure that the extracted knowledge aligns with the actual equipment data within the target area. During the extraction process, a knowledge extraction algorithm is employed, focusing on four core equipment categories in the target area: overhead contact system support foundations, contact suspension, support devices, and auxiliary equipment. The core content of each type of equipment is extracted one by one from the business knowledge model in the power supply field.

[0146] For example, for support column foundations, extract the material, height, diameter, and load-bearing capacity from the equipment entity attributes; the foundation depth and fixing method from the structural features; the connection specifications with the contact suspension from the business association rules; and the maintenance association requirements. For contact suspensions, extract the conductor type, cross-sectional dimensions from the entity attributes; the suspension method, tension parameters from the structural features; and the linkage relationship with the support column foundation and support devices from the business association rules. For support devices and auxiliary equipment, extract the corresponding entity attributes, structural features, and business association rules according to the same logic. After extraction, associate the extracted data of various types of equipment with the equipment identifiers in the target area, and label the knowledge source and business meaning; further integrate all the extracted data to obtain clearly categorized and complete equipment knowledge mapping data.

[0147] Step 530: Associate the basic data and knowledge mapping data of the same device to generate device knowledge matching data for each device.

[0148] Specifically, for the basic data of equipment in a given area and the equipment knowledge mapping data, a one-to-one data matching and association process is initiated to achieve precise binding between business knowledge and actual equipment. During the matching process, a matching algorithm is used, with the equipment identifier as the sole matching criterion, to compare the equipment knowledge mapping data with the basic data of equipment in the given area one by one, ensuring that each actual device corresponds to a unique knowledge mapping content. In specific matching, the identifier of each device in the basic data of the target area is compared with the identifier of the corresponding device type in the equipment knowledge mapping data; after a match, the geographical distribution information and ledger information of the equipment are bound with the corresponding entity attributes, structural characteristics, and business association rules, thereby establishing a correspondence between actual equipment and business knowledge.

[0149] After matching is completed, the consistency of matching is checked, and data that fails to match or has incorrect association is removed. Matching deviations are corrected, and association records containing basic data and knowledge mapping data are generated for each device. The data are then classified and organized according to device type and spatial location, and finally, closely related and complete device knowledge matching data is obtained.

[0150] Step 540: Based on the equipment knowledge matching data of each device, construct a three-dimensional geometric framework model of the power supply system in the target area.

[0151] Specifically, for equipment knowledge matching data, the process of constructing a three-dimensional geometric framework model of the contact network power supply equipment in the target area is initiated to accurately restore the physical structure and spatial layout of the equipment. During the construction of the three-dimensional geometric framework model, a three-dimensional modeling algorithm is used, with equipment knowledge matching data as the core basis. First, the structural features and spatial coordinate information of each type of equipment are extracted. Specifically, the column foundations are constructed according to structural features such as material, height, and diameter, clarifying the ratio of burial depth to height above ground. The contact suspension is constructed according to structural features such as suspension method and conductor type, restoring the conductor's direction and tension state using spatial coordinates. Support devices and auxiliary equipment are constructed according to their own structural features, clarifying their connection positions with the column foundations and contact suspension.

[0152] Furthermore, considering the spatial spacing and relative positions of the equipment, the 3D frames of various devices are pieced together according to the actual spatial layout to recreate the overall spatial distribution of the contact network power supply equipment in the target area. The equipment identification and structural parameters corresponding to each frame part are labeled. After construction, the rationality of the frame structure and the accuracy of the spatial layout are verified, and modeling deviations are corrected. Finally, a complete 3D geometric frame model with a realistic layout and clear parameters is obtained.

[0153] Step 550: Configure the parameters of the three-dimensional geometric framework model according to the business knowledge model to obtain a digital twin model of the power supply system in the target area; the parameter configuration includes the configuration of business knowledge elements and the configuration of logical rules.

[0154] Specifically, the knowledge elements and logical rules of the business knowledge model in the power supply field are associated and bound with the three-dimensional geometric framework data to complete the parameter configuration and instantiation encapsulation of the digital twin model, generating an instantiated digital twin model. This instantiated digital twin model is used to support the real-time mapping and linkage update of subsequent equipment status.

[0155] For the business knowledge model and 3D geometric framework data in the power supply field, a digital twin model parameter configuration and instantiation encapsulation process is initiated to generate an instantiated digital twin model that can support real-time mapping and linkage updates of equipment status. During the binding process, an association binding algorithm is used to associate and bind the knowledge elements and logical rules in the power supply field business knowledge model with the 3D geometric framework data one by one. Each part of the 3D geometric framework is bound to the corresponding equipment entity attributes, structural features, business association rules, and operational constraint logic. For example, the 3D framework of the support foundation is bound to knowledge elements such as its material and load-bearing capacity, and logical rules such as maintenance cycle and fault handling are also bound.

[0156] Furthermore, parameter configuration is performed, setting real-time model mapping parameters and linkage update frequency, clarifying the mapping relationship between the 3D geometric framework model and the actual equipment status, and setting linkage rules between knowledge elements and equipment operating status. After configuration, an instantiation encapsulation algorithm is used to encapsulate the associated and bound 3D geometric framework data, knowledge elements, and logical rules, and add model identifiers, segment information, and data update interfaces to obtain a complete instantiated digital twin model.

[0157] After encapsulation, verify the integrity of model parameters and the accuracy of association binding, correct encapsulation deviations, and ensure that the model can receive actual device operation data in real time, so as to realize real-time mapping and linkage update of device status.

[0158] In one embodiment, reference Figure 6 Model building module 40 constructs a three-dimensional geometric framework model of the power supply system within the target area based on the device knowledge matching data of each device, specifically including steps 541-544, wherein:

[0159] Step 541: Extract the geometric structure information of each device from the device knowledge matching data of each device.

[0160] Specifically, from the equipment knowledge matching data, the structural parameters, installation posture, and connection relationships of various overhead contact line power supply equipment are extracted to generate a set of equipment geometric information. For the equipment knowledge matching data, a process for extracting geometric information of the overhead contact line power supply equipment within the target area is initiated to ensure that the extracted information fully meets the requirements of 3D modeling. During the extraction of geometric information, an extraction algorithm is used, focusing on four main categories of equipment: overhead contact line support foundations, contact suspension, support devices, and auxiliary equipment. Three core categories of geometric information are extracted one by one from the equipment knowledge matching data.

[0161] The extraction of structural parameters covers the height, diameter, and burial depth of the support foundation; the cross-sectional dimensions and suspension length of the contact suspension conductors; the dimensions of the support brackets and the specifications of the connecting components; the external dimensions of the auxiliary equipment; and the installation interface parameters. The extraction of installation postures covers the vertical angle of the support foundation, the suspension tilt angle of the contact suspension, and the installation angle of the support equipment, clarifying the spatial posture parameters of various equipment. The extraction of connection relationships covers the connection positions between the support foundation and the support equipment, the connection nodes between the support equipment and the contact suspension, and the connection points between the auxiliary equipment and the main equipment, clarifying the connection methods and relative positions between each piece of equipment. After extraction, the geometric information of each type of equipment is associated with its identification, the source of the information and the meaning of the geometric parameters are labeled, and the information is categorized and organized by equipment type. Missing or incorrect geometric information is removed, ultimately resulting in a clear and complete set of equipment geometric information.

[0162] Step 542: Analyze the geometric structure information of each device to determine the geometric constraint data of each device.

[0163] Specifically, the geometric structure information of each device is analyzed to determine the geometric structure type, spatial assembly logic, and positional constraint relationship of each device with adjacent devices, thereby obtaining the set constraint data of each device; the geometric constraint data includes the structure type, spatial assembly logic, and positional constraint relationship with adjacent devices.

[0164] For the equipment geometric information set, a geometric information parsing and constraint relationship mining process is initiated to clarify the equipment's geometric structure and spatial assembly requirements. During the parsing process, an analytical algorithm is used to analyze each type of data in the equipment's geometric information set one by one to determine the geometric structure type of each type of equipment. Specifically, the support foundation is a column geometry, the contact suspension is a cable geometry, the support device is a frame geometry, and the auxiliary equipment is a small component geometry, further clarifying the core geometric features and composition of each type of structure.

[0165] Furthermore, the spatial assembly logic of various equipment is analyzed to clarify the assembly sequence and requirements. Specifically, the support foundation is assembled first, followed by the support device, then the contact suspension, and finally the additional equipment. The connection nodes and assembly standards for each step are clearly defined. At the same time, the positional constraints between adjacent equipment are explored to clarify the spacing constraints between the support foundation and the support device, the height constraints between the support device and the contact suspension, and the positional constraints between the additional equipment and the main equipment. The constraint parameters are quantified, and the constraint range is clearly defined.

[0166] After parsing, the geometric structure type, spatial assembly logic, and positional constraint relationships are integrated, and the constraint basis and parameter standards are marked, finally obtaining logically clear and well-defined geometric constraint data.

[0167] Step 543: Generate a 3D sub-model for each device based on the geometric constraint data of each device.

[0168] Specifically, based on geometric constraint data, parametric modeling technology is used to sequentially construct three-dimensional geometric sub-models of the overhead contact line support foundation, contact suspension, support device, and auxiliary equipment, generating three-dimensional sub-model data for each device. For the geometric constraint data, the three-dimensional geometric sub-model construction process for various overhead contact line power supply equipment is initiated to ensure that the three-dimensional geometric sub-models conform to the actual geometric characteristics of the equipment.

[0169] During the modeling process, a parametric modeling algorithm was employed, using geometric constraint data as the core basis to construct three-dimensional geometric sub-models sequentially according to equipment type. When constructing the three-dimensional sub-model of the support foundation, parameterizable variables such as height, diameter, and burial depth were set based on the column's geometric structure type and structural parameters. Combined with positional constraints, a column sub-model conforming to actual dimensions was constructed, clearly defining the ratio of above-ground to underground burial depth. When constructing the three-dimensional sub-model of the contact suspension, parameters such as conductor cross-sectional dimensions, suspension length, and inclination angle were set based on the cable's geometric structure type and suspension parameters. Combined with assembly logic, a cable sub-model conforming to the actual suspension state was constructed. When constructing the three-dimensional sub-models of support devices and additional equipment, frame-type and component-type sub-models were constructed respectively based on the corresponding geometric structure type and structural parameters, clearly defining the geometric characteristics of the connection interfaces.

[0170] After the three-dimensional geometric sub-models of each type of equipment are constructed, the consistency between the structure and geometric constraint data of the three-dimensional geometric sub-models is checked, the modeling deviations are corrected, and the equipment identifiers and geometric parameters corresponding to the sub-models are labeled. Finally, three-dimensional sub-models of each equipment with accurate structure and complete parameters are obtained.

[0171] Step 544: Based on the equipment knowledge matching data of each device, assemble the three-dimensional sub-models of each device to obtain the three-dimensional geometric framework model of the power supply system in the target area.

[0172] Specifically, the 3D sub-model data of each device are combined and assembled according to the spatial distribution relationship and assembly logic in the device knowledge matching data to restore the overall spatial layout of the contact network power supply equipment in the target area, thus obtaining the overall 3D assembly model data. During the assembly process, an assembly algorithm is used, based on the spatial distribution relationship in the device knowledge matching data and the spatial assembly logic in the geometric constraint data, to combine and assemble the devices according to a preset assembly sequence.

[0173] First, position the 3D sub-model of the support foundation according to the spatial coordinates in the equipment knowledge matching data, ensuring that the spacing meets the position constraints; then, assemble the 3D sub-model of the support device to the corresponding connection node of the support foundation, conforming to the installation posture constraints and position constraints; then, assemble the 3D sub-model of the contact suspension with the sub-model of the support device, ensuring that the suspension tilt angle and connection position meet the assembly logic; finally, assemble the 3D sub-model of the additional equipment to the connection part of the corresponding main equipment to complete the overall assembly.

[0174] During assembly, the assembly accuracy and spatial relationship of each 3D sub-model are checked in real time, and assembly deviations are corrected to ensure that the connection relationship and spatial layout of each equipment sub-model are consistent with the actual equipment. After assembly, the overall 3D assembly model data covering all overhead contact line power supply equipment in the target area is obtained, and the overall layout parameters and their correlation with each sub-model are marked.

[0175] Furthermore, the geometric accuracy of the overall 3D assembly model data is verified to correct structural deviations, refine equipment details, and optimize spatial layout rationality, resulting in the final 3D geometric framework data. For the overall 3D assembly model data, a geometric accuracy verification and model optimization process is initiated to generate standardized 3D geometric framework data. During the verification process, an accuracy verification algorithm is used to check the geometric accuracy, structural integrity, and spatial layout rationality of the overall 3D assembly model one by one. Geometric accuracy verification focuses on the dimensional parameters and connection positions of each equipment sub-model, comparing them with the equipment's geometric information set and geometric constraint data to correct dimensional deviations and connection misalignments. For example, structural integrity verification includes checking for missing sub-models and omitted details, as well as supplementing features such as equipment connection components and external details; spatial layout rationality verification includes checking whether the equipment spacing and installation posture conform to the actual scene, optimizing unreasonable spatial layouts for wall-mounted installations, and ensuring that the overall layout is consistent with the actual equipment distribution in the target area.

[0176] After verification, the model is optimized by simplifying redundant geometric details while retaining core structural features. Rendering efficiency is also improved, and geometric parameters, spatial coordinates, and device identifiers are annotated. Finally, the completeness and rationality of the optimized model are checked, optimization deviations are corrected, and a complete, accurate, and realistically aligned 3D geometric framework is obtained.

[0177] In one embodiment, reference Figure 7 The model building module 40, when configuring parameters of the three-dimensional geometric framework model based on the business knowledge model to obtain a digital twin model of the power supply system within the target area, specifically includes steps 551-554, wherein:

[0178] Step 551: Extract knowledge configuration data for each device in the power supply system within the target area from the business knowledge model; the knowledge configuration data includes business knowledge elements corresponding to device attributes, operating logic rules, and fault judgment criteria.

[0179] Specifically, knowledge content corresponding to equipment attribute parameters, operational logic rules, and fault judgment criteria in the power supply business knowledge model is extracted to generate knowledge configuration data. For the power supply business knowledge model, a knowledge configuration data extraction process is initiated to ensure that the extracted knowledge content aligns with the parameter configuration requirements of the digital twin model. During the knowledge configuration data extraction process, a knowledge extraction algorithm is employed, focusing on the core requirements of the digital twin model's parameter configuration, and extracting the three core knowledge categories—equipment attributes, operational logic rules, and fault judgment criteria—from the power supply business knowledge model one by one.

[0180] Equipment attribute parameters are directly extracted from the equipment entity knowledge submodule, covering core attributes such as material, size, load-bearing capacity, and installation parameters of contact wire support foundations, contact suspensions, support devices, and auxiliary equipment. Operational logic rules are comprehensively extracted from the operation control knowledge submodule and the environment adaptation knowledge submodule, covering equipment operating parameter thresholds, start-stop control logic, and linkage operation requirements. Fault judgment criteria are extracted from the fault handling knowledge submodule, covering various equipment fault characteristics, fault judgment parameters, and judgment logic. After extraction, the extracted knowledge configuration data is categorized and organized by equipment type, and the corresponding equipment type, business meaning, and application scenario are labeled. Knowledge content irrelevant to model parameter configuration is removed, and deviations in the extraction process are corrected, ultimately resulting in knowledge configuration data that is clearly structured, complete in content, and closely matches configuration requirements.

[0181] Step 552: Extract entity identification data for each device from the 3D geometric framework model; entity identification data includes device identification information, geometric feature parameters, and spatial location information.

[0182] Specifically, the identification information, geometric feature parameters, and spatial location information of each device entity in the 3D geometric framework data are extracted to generate model entity identification data. For the 3D geometric framework data, the model entity identification data extraction process is initiated to clarify the core information of each device entity in the 3D geometric model. During the entity identification data extraction process, an extraction algorithm is used to extract entity identification data one by one from the 3D geometric framework data, ensuring that the information and knowledge configuration data can be accurately correlated.

[0183] The equipment entity identification information includes a unique identifier for each piece of equipment and an equipment type identifier. The unique identifier of each equipment is consistent with the equipment identifier in the equipment knowledge matching data to ensure the uniqueness of the association. The geometric feature parameters include the core parameters such as the size, shape, and installation posture of each equipment's 3D geometric model, corresponding to the geometric detail parameters in the 3D geometric framework data, fully preserving the equipment's geometric features. The spatial location information includes the longitude, latitude, and altitude coordinates of each equipment entity, as well as its relative positional relationship with adjacent equipment, reconstructing the actual spatial layout of the equipment.

[0184] After extraction, the entity identification data is associated with the device entity identification one by one, categorized and organized according to device type, and the information source and corresponding geometric frame node are marked. Invalid data with missing identification or location information is removed, and finally, complete and clearly associated model entity identification data is obtained.

[0185] Step 553: Associate the knowledge configuration data of each device with the entity identification data, and populate the knowledge configuration data into the entity nodes of the corresponding devices in the 3D geometric frame model to obtain the parameter configuration model.

[0186] Specifically, a one-to-one correspondence is established between knowledge configuration data and model entity identification data to create a mapping relationship between business knowledge and the 3D geometric model, generating knowledge model binding data. For both knowledge configuration data and model entity identification data, a one-to-one correspondence process is initiated to establish a stable mapping relationship between business knowledge and the 3D geometric model. During the association process, an association algorithm is used, with the device entity identification as the sole basis for association, to compare and associate the knowledge configuration data with the model entity identification data one by one.

[0187] In the specific association process, the unique identifier of each device in the entity identification data is compared with the identifier of the corresponding device type in the knowledge configuration data. After the comparison is consistent, the knowledge configuration content such as the device attribute parameters, operating logic rules, and fault judgment criteria of the device is bound with the corresponding device geometric feature parameters and spatial location information to clarify the one-to-one correspondence between business knowledge and three-dimensional geometric model entities.

[0188] After the association is completed, the consistency and completeness of the association are checked, and invalid data with association errors and missing associations are removed. The association deviations are corrected, and an association record containing knowledge configuration and model identifier is generated for each device. The records are then classified and organized according to the spatial location of the devices, and finally, the knowledge model binding data with close association and complete data is obtained.

[0189] Step 554: Instantiate and encapsulate the parameter configuration model to generate an instantiated digital twin model.

[0190] Specifically, based on the knowledge model binding data, the parameters and rules in the knowledge configuration data are populated into the corresponding entity nodes of the 3D geometric framework model to complete the model parameter configuration and generate a parameter configuration model. For the knowledge model binding data, the 3D geometric framework model parameter configuration process is initiated to achieve deep integration of knowledge parameters and model entities. During the configuration process, a parameter configuration algorithm is adopted, using the knowledge model binding data as the core basis, to populate the various parameters and rules in the knowledge configuration data into the corresponding entity nodes of the 3D geometric framework model one by one according to their correlation relationships.

[0191] During the specific filling process, the equipment attribute parameters are filled into the geometric model nodes of the corresponding equipment entity to achieve the binding of attribute parameters and geometric features; the operation logic rules are filled into the control nodes of the model to clarify the logical constraints of equipment operation; and the fault judgment criteria are filled into the monitoring nodes of the model to provide a basis for subsequent fault identification.

[0192] During the population process, the accuracy of the population needs to be checked in real time to ensure that the parameters and rules corresponding to each model entity node are complete and correct, avoiding incorrect or missing parameters. After the population is completed, the model is initially validated to correct parameter configuration deviations and ensure that the parameters, model entities, and business logic are highly consistent, ultimately resulting in a parameter configuration model with a complete structure and all parameters.

[0193] Furthermore, the parameter configuration model is instantiated and encapsulated, and the geometric data, knowledge data and correlation logic of the model are integrated to generate an instantiated digital twin model that can support real-time mapping and linkage updates of device status.

[0194] During the encapsulation process, an instantiation encapsulation algorithm is employed. First, all data of the parameter configuration model is integrated, including 3D geometric data, knowledge configuration data, and their interrelationships, clarifying the storage paths and calling relationships for each type of data. Then, basic model information is added, including core information such as model identifier, the heavy-haul railway section to which it belongs, generation time, and data update interface. Real-time mapping parameters and linkage update frequency are set for the model, clarifying the interface between the model and actual equipment operation data, ensuring that the model can receive actual equipment operation data in real time and complete status updates.

[0195] During the encapsulation process, the model data storage structure is optimized, redundant data is simplified, and model operating efficiency is improved, while all core data and logical relationships are retained. After encapsulation, the instantiated digital twin model is comprehensively verified to check the model's geometric accuracy, parameter completeness, and the rationality of the associated logic. Encapsulation deviations are corrected to ensure that the model can accurately reproduce the physical structure, operating status, and business logic of the contact network power supply equipment in the target area. The final result is an instantiated digital twin model, providing reliable model support for subsequent equipment operation and maintenance and fault early warning.

[0196] In one embodiment, reference Figure 8 The heavy-haul railway multimodal power supply data governance and model building system also includes an update module, which is specifically used to execute steps 610-640, wherein:

[0197] Step 610: Based on the business knowledge model, perform semantic parsing, feature extraction and state recognition on the target power supply data to obtain state parsing data; the target power supply data is multimodal power supply data collected for the power supply system of the target area.

[0198] First, a real-time data interface channel is established between the instantiated digital twin model and the multimodal data acquisition terminal, creating a real-time data transmission link and generating data transmission link configuration data. By initiating the real-time interface process between the instantiated digital twin model and the multimodal data acquisition terminal, stable and efficient data transmission is ensured. Furthermore, based on a business knowledge model in the power supply field, a real-time data parsing engine is constructed to perform semantic parsing, feature extraction, and status recognition on the real-time multimodal power supply data transmitted from the acquisition terminal, generating real-time status parsing data.

[0199] During the docking process, a docking algorithm is employed to first clarify the data receiving interface of the instantiated digital twin model and the data sending interface of the multimodal data acquisition terminal, unifying the interface communication protocol and data transmission format to eliminate interface incompatibility issues. Furthermore, a bidirectional real-time data docking channel is established. One end of the data docking channel connects to the receiving interface of the instantiated digital twin model, and the other end connects to the multimodal data acquisition terminals distributed on the overhead contact line power supply equipment in the target area. These multimodal data acquisition terminals include, but are not limited to, various acquisition terminals such as overhead contact line monitoring terminals, substation monitoring terminals, and environmental monitoring terminals.

[0200] After the data connection channel is established, a real-time data transmission link is set, with a fixed transmission frequency, data frame structure, and retransmission mechanism. The retransmission mechanism is configured to trigger retransmission immediately upon link interruption or data loss, ensuring no data loss or delay. Link parameters are also configured, specifying transmission bandwidth, data encryption methods, and link authentication requirements. After link configuration, transmission link configuration data containing interface parameters, transmission rules, and link identifiers is generated, labeling the corresponding devices and acquisition terminals to provide link support for subsequent real-time data transmission.

[0201] Furthermore, based on the knowledge elements and logical rules in the power supply business knowledge model, a real-time data parsing engine is constructed. The data parsing engine includes a semantic parsing module 20, a feature extraction module, and a state recognition module. During the parsing process, the parsing engine receives real-time data transmitted by the multimodal data acquisition terminal through the transmission link, including contact network operating parameters, equipment fault characteristics, environmental parameters, and multimodal data such as images and videos.

[0202] The semantic parsing module 20 parses the business semantics of various types of data, converting unstructured image, video, and text data into recognizable semantic information; the feature extraction module extracts the core features in the data, corresponding to the feature elements in the power supply business knowledge model; the status recognition module combines the fault judgment criteria and operation constraint rules in the business knowledge model to identify the real-time operating status of the equipment and distinguish between normal operation, abnormal warning, and fault status.

[0203] After parsing is completed, the semantic parsing results, core features and state recognition results are integrated, and the data collection time, collection terminal and corresponding device are labeled to finally obtain real-time state parsing data with clear structure and complete information.

[0204] Step 620: Determine the target device corresponding to the state resolution data in the digital twin model.

[0205] Furthermore, the status parsing data is associated and matched with the device entity nodes in the instantiated digital twin model to determine the model entity (target device) corresponding to each real-time status data, and generate status data matching results.

[0206] During the matching process, a matching algorithm is employed, using the device entity identifier as the sole matching criterion. The status parsing data is compared one-to-one with the device entity nodes in the instantiated digital twin model. Specifically, the device identifier is extracted from the status parsing data and compared with the identifiers of all device entity nodes in the instantiated digital twin model. Upon matching, the model entity node corresponding to the status parsing data is determined, clarifying the one-to-one correspondence between data and nodes.

[0207] For real-time data from multimodal fusion, data types are broken down. Equipment operating parameter data is matched to the corresponding entity's operating node, fault characterization data is matched to the corresponding entity's monitoring node, and environmental parameter data is matched to the corresponding entity's environmental association node. After matching, consistency is checked, data with matching errors or no corresponding model entity is removed, and matching deviations are corrected. The final result is a status data matching result containing data identifiers, model entity node identifiers (unique identifiers of the target device), and matching results.

[0208] Step 630: Update the state of the entity node corresponding to the target device in the digital twin model using the state parsing data to obtain the updated digital twin model.

[0209] Specifically, based on the state data matching results, the real-time state parsing data is mapped to the entity nodes corresponding to the instantiated digital twin model, and the operating parameters, state information and associated logic of the device in the model are updated to obtain the state-updated twin model.

[0210] For the status data matching results, the process of mapping real-time status parsing data to the instantiated digital twin model and updating the model is initiated to ensure that the model status is synchronized with the actual equipment. During the mapping process, a mapping algorithm is used to map the real-time status parsing data to the corresponding entity nodes of the instantiated digital twin model one by one, based on the status data matching results. Specifically, during the mapping, equipment operating parameter data is mapped to the parameter fields of the corresponding entity nodes, updating the operating parameters of the equipment in the model, such as voltage, current, and temperature; equipment status identification results are mapped to the status fields of the corresponding entity nodes, updating the normal, abnormal, and fault status information of the equipment in the model; and associated logic data is mapped to the associated links of the corresponding entity nodes, updating the association logic between the equipment and the environment in the model.

[0211] After mapping is complete, the model update process is initiated, synchronously updating the geometric display status and data relationships of the model according to fixed update rules. For example, when a device malfunctions, the corresponding entity node in the model displays a corresponding malfunction indicator, and the linkage logic of the associated device is updated synchronously. After the update is completed, the completeness and accuracy of the model update are verified, update deviations are corrected, and a digital twin model with updated status is obtained, ensuring that the model status is initially consistent with the actual device operating status.

[0212] Step 640: Perform a runtime logic deduction on the updated digital twin model to generate state deduction data for associated devices, and associate the state deduction data with the updated information of the digital twin model.

[0213] Specifically, the operational logic of the updated digital twin model is deduced, the rationality of the equipment's operating status is determined based on the business knowledge model in the power supply field, and state deduction data is generated. The state deduction data is synchronously associated with the model update information to achieve real-time linkage between the instantiated digital twin model and the actual power supply equipment, ensuring that the model's state is consistent with the actual state of the equipment.

[0214] For the updated twin model and the power supply domain business knowledge model, a model operation logic deduction and real-time linkage process is initiated to ensure that the model is completely consistent with the actual equipment state. During the deduction process, a logic deduction algorithm is used to perform operation logic deduction on the updated twin model based on the operation logic rules and fault judgment criteria in the power supply domain business knowledge model. This simulates the equipment operation process, determines the rationality of the current operating state of the equipment, and identifies potential state deviations and logical contradictions that may occur during the model update process.

[0215] After the simulation is completed, equipment status simulation data is generated, including the rationality judgment results of equipment operating status, deviation correction suggestions, and linkage response logic. Furthermore, the status simulation data is synchronously correlated with model update information. If the simulation finds discrepancies between the model status and business logic, the model parameters and status information are immediately corrected based on the simulation data. Simultaneously, a real-time linkage mechanism between the model and the actual power supply equipment is established. After the model status is updated, it is synchronously fed back to the data acquisition terminal. The acquisition terminal adjusts the acquisition frequency and focus according to the linkage logic, achieving bidirectional linkage between the instantiated digital twin model and the actual power supply equipment. After linkage is completed, the consistency between the model status and the actual equipment status is verified, linkage deviations are corrected, and it is ensured that the model status can accurately reflect the operating status of the actual power supply equipment in real time.

[0216] In one embodiment, reference Figure 9 The update module, based on the business knowledge model, performs semantic parsing, feature extraction, and state recognition on the target power supply data to obtain state parsing data. Specifically, this includes steps 611-614, where:

[0217] Step 611: Using semantic rule knowledge, perform modal classification and semantic annotation on the target power supply data to generate semantically annotated data.

[0218] Specifically, the business knowledge model includes state determination rule knowledge and semantic rule knowledge. It acquires real-time target power supply data transmitted from multimodal data acquisition terminals and generates real-time raw data; the target power supply data includes multiple types of data corresponding to equipment operating parameters, detection images, monitoring videos, and status alarm information.

[0219] For transmission link configuration data, a real-time power supply data acquisition process is initiated to ensure that the acquired data comprehensively covers various types of power supply-related information. During the acquisition of target power supply data, a receiving algorithm is employed to receive various real-time data transmitted from multi-modal data acquisition terminals distributed across the contact network power supply equipment in the target area via the established real-time data transmission link. Specifically, the system receives equipment operating parameter data, covering core operating parameters such as contact network conductor height, pull-out value, wear amount, temperature, and operating voltage, current, and power of power distribution equipment; it also receives detection image data, covering real-time images of key components such as contact network conductors, insulators, and support foundations; it receives monitoring video data, covering continuous video of equipment operating status within the substation and real-time operating scenes of equipment along the contact network; and it receives status alarm information data, covering various alarm information such as abnormal equipment operating parameters, component damage, and environmental exceedances, including alarm type and alarm trigger time. After acquisition, the data is categorized and organized by data type, with each type of data labeled with the acquisition terminal identifier, acquisition time, and acquisition location. Damaged or missing data generated during transmission is removed, resulting in comprehensive, complete, and traceable real-time raw data covering multiple types.

[0220] For the power supply business knowledge model and real-time raw data, a modal classification and semantic annotation process is initiated to achieve standardized processing of the real-time raw data. During feature extraction, classification and annotation algorithms are employed. First, based on the semantic rules in the power supply business knowledge model, the real-time raw data is modally classified. Equipment operating parameters are classified as structured data, detection images and monitoring videos are classified as unstructured image / video data, and status alarm information is classified as structured text data. A unique modal identifier is assigned to each type of data. Subsequently, based on the feature standards in the business knowledge model, targeted semantic annotation is performed on each type of data. Equipment operating parameters are annotated with parameter name, numerical range, and corresponding equipment component; detection images are annotated with the shooting location, equipment component name, and image clarity; monitoring videos are annotated with the range of equipment covered and the type of operating scenario; and status alarm information is annotated with the corresponding equipment, alarm reason, and alarm level.

[0221] After the annotation is completed, the modality identifiers and semantic annotation information are bound to the corresponding original data, classified and organized according to modality type, the accuracy of the annotation is checked, the annotation deviation is corrected, and finally semantically clear and well-categorized semantically annotated data is obtained.

[0222] Step 612: Extract features from the semantically labeled data to obtain real-time feature data.

[0223] Specifically, for semantically labeled data, a real-time feature extraction process is initiated, focusing on key information about equipment operation to ensure that the extracted features align with the requirements of status recognition. During feature extraction, a feature extraction algorithm is employed, combined with semantically labeled information, to extract targeted features from various types of semantically labeled data. From equipment operating parameter data, real-time numerical features and variation amplitude features of parameters such as voltage, current, and temperature are extracted, corresponding to core indicators of equipment operating status. From detection image data, visual features such as the integrity of equipment components, surface wear, and component connection status are extracted, corresponding to equipment fault characteristics. From monitoring video data, dynamic features such as equipment operation actions, component displacement, and personnel work status are extracted, corresponding to the real-time operating status of the equipment. From status alarm information data, features such as alarm type, alarm intensity, and alarm duration are extracted, corresponding to potential equipment anomalies. After extraction, various core features are associated with semantically labeled information and equipment identifiers. Redundant features irrelevant to equipment operating status, fault characteristics, and environmental changes are removed. Features are categorized and organized by type to generate clear and well-connected real-time feature data, providing core feature support for subsequent status recognition.

[0224] Step 613: Use state determination rules to compare and analyze real-time feature data to obtain state recognition results.

[0225] Specifically, for real-time feature data and the power supply business knowledge model, an equipment operation status identification process is initiated to accurately identify the equipment operation status and abnormal conditions. During the identification process, a comparative analysis algorithm is used to compare and analyze the real-time feature data with the status judgment rules in the power supply business knowledge model one by one. The equipment operation parameter features are compared with the operation parameter threshold rules in the model to determine whether the parameters are within the normal range; the fault characterization features in images and videos are compared with the fault judgment rules in the model to identify whether the equipment has component damage or other abnormal potential hazards; the feature change amplitude is compared with the change trend judgment rules in the model to predict the change trend of the equipment operation status and distinguish between three categories: normal operation, abnormal warning, and fault status.

[0226] After the comparative analysis is completed, the comparison results are integrated to clarify the current operating status level of the equipment, the specific types of abnormal hazards, the predicted trend of change, and the judgment basis and corresponding characteristics are marked. The comparison deviation is corrected, and finally a logically clear and clearly judged status identification result is obtained, which provides a basis for subsequent structured organization.

[0227] Step 614: Associate the status recognition result with the corresponding device identifier, time information and spatial location information to obtain status analysis data.

[0228] Specifically, for the status identification results, a structured processing flow is initiated to generate standardized real-time status analysis data. During this process, a structured processing algorithm is used to first extract core information from the status identification results, including equipment operating status, types of anomalies and potential hazards, trends, and judgment criteria. Then, the corresponding data information is associated, binding the status identification results with equipment identifiers to ensure that each identification result corresponds to a unique device; data acquisition time information is associated to clarify the specific time point corresponding to the status identification; and spatial location information is associated to clarify the longitude, latitude, altitude coordinates, and area of ​​the corresponding equipment. After association, all information is organized according to a fixed structure, setting a unified field format, and arranging the equipment identifier, time information, spatial location information, and core status identification content sequentially. Furthermore, redundant information is removed from the arranged data, and consistency of associations is checked and processing deviations are corrected to ensure a well-structured data structure, complete information, and clear associations. After processing, standardized real-time status analysis data is finally obtained.

[0229] In one embodiment, the state of the entity node corresponding to the target device in the digital twin model is updated using state parsing data to obtain an updated digital twin model. Specifically, this may include: extracting the model entity node identifier and the corresponding real-time state parsing data from the state data matching results to generate node state association data; based on the node state association data, locating the corresponding device entity node in the instantiated digital twin model, retrieving the current operating parameters and state information of the device entity node, and generating the node current state data; comparing the real-time state parsing data with the node current state data to determine the parameter items, state items, and association logic that need to be updated, and generating a node update list; according to the node update list, writing the parameters and state information from the real-time state parsing data into the corresponding model entity node, updating the node's operating logic and association relationships, and generating single-node update data; performing linkage verification on all completed single-node update data to ensure that the association logic between nodes is consistent with the power supply domain business knowledge model, and generating node linkage verification results; based on the node linkage verification results, integrating all updated node data with the overall model logic to generate a state-updated twin model.

[0230] Specifically, for the status data matching results, a node status association data extraction process is initiated to clarify the correspondence between model entity nodes and real-time status data, providing a basis for subsequent node positioning and updates. During the extraction process, an extraction algorithm is employed, focusing on the core information in the status data matching results. The model entity node identifier and corresponding real-time status parsing data are extracted one by one from each matching result, ensuring the extracted information is complete and accurately correlated. Specifically, the unique node ID and device type identifier in the model entity node identifier are extracted, ensuring consistency with the node identifier in the instantiated digital twin model. Equipment operating parameters, status identification results, anomaly / potential information, time information, and spatial location information are extracted from the corresponding real-time status parsing data, completely preserving the core content of the real-time status. After extraction, each model entity node identifier is bound to its corresponding real-time status parsing data, the correlation matching criteria are marked, and the data is categorized and organized according to node identifier. Correlation errors and missing information that occurred during the extraction process are removed, and extraction deviations are corrected. Finally, node status association data with a clear structure, close correlation, and complete information is obtained, ensuring that each model entity node has corresponding real-time status data support.

[0231] Furthermore, for the node status association data and instantiated digital twin model, the process of locating model entity nodes and retrieving their current status is initiated to clarify the current operating status of the nodes. During the location process, a node location algorithm is used, with the model entity node identifier in the node status association data as the unique location basis. The corresponding equipment entity nodes are located one by one in the instantiated digital twin model. The target node is locked by comparing the node identifier, covering model nodes corresponding to various types of equipment such as catenary support foundations, catenary suspensions, support devices, and auxiliary equipment. After the location is completed, a data retrieval algorithm is used to retrieve all currently stored operating parameters and status information of the equipment entity node. The operating parameters include core parameters such as equipment voltage, current, temperature, and dimensions, while the status information includes the current operating status of the equipment, associated logic, and historical update records. After retrieval, the retrieved node's current operating parameters and status information are associated with the corresponding node identifier and real-time status parsing data. The retrieval time and node location are marked, the integrity of the retrieved data is checked, data that failed to be retrieved or had missing information is removed, and retrieval deviations are corrected. Finally, the node's current status data containing node identifier, current operating parameters, and current status information is obtained, providing a reference for subsequent data comparison and node updates.

[0232] Furthermore, for node status-related data and current node status data, a data comparison, list generation, and single-node update process is initiated to achieve accurate updates of model nodes. During the comparison process, a comparison algorithm is used to compare the real-time status parsing data in the node status-related data with the operating parameters and status information in the current node status data one by one, clarifying the differences between the two. Specifically, the comparison involves comparing the numerical differences of equipment operating parameters, the types of equipment operating states, and the differences in the relationships of the associated logic, determining the parameters, status items, and associated logic items that need to be updated. Examples include changes in voltage values ​​in operating parameters, changes in equipment status from normal to abnormal, and adjustments to equipment linkage relationships in associated logic. Subsequently, a node update list is generated, clearly marking the node identifier, the specific items to be updated, the values ​​or statuses before and after the update, and categorized by node identifier. Based on the node update list, an update algorithm is used to write the parameters and status information from the real-time status parsing data one by one into the corresponding model entity nodes, overwriting the original old data, updating the node's operating logic and relationships, and ensuring that the node data is consistent with the real-time status. After the update is completed, the accuracy of the update is verified, the update deviation is corrected, and finally the single-node update data containing node identifier, update content and update time is obtained, which provides data support for subsequent linkage verification.

[0233] Furthermore, for all single-node updated data and the power supply domain business knowledge model, a node linkage verification and model integration process is initiated to generate a twin model after the status update. During the verification process, a linkage verification algorithm is used to comprehensively verify all updated single-node data, focusing on the correlation logic between nodes. The algorithm compares the updated linkage relationship between nodes with the operational logic rules and correlation link requirements in the power supply domain business knowledge model to verify whether it conforms to the business logic. Specifically, the verification checks whether the correlation logic between adjacent device nodes is smooth, whether the response logic of associated nodes after the equipment operating parameters are updated is reasonable, and whether the linkage rules between abnormal state nodes and associated nodes meet business requirements. Data with inconsistent linkage logic or incorrect associations is removed, verification deviations are corrected, and node linkage verification results are generated, identifying nodes that pass verification and those that need correction. Based on the verified node linkage verification results, an integration algorithm is used to integrate all updated single-node data, streamline the overall model logic, ensure smooth data association between nodes, ensure the overall logic conforms to the business needs of the power supply domain, optimize the model data storage structure, and retain core updated content and correlation logic. After integration, the overall integrity and rationality of the model are checked, integration deviations are corrected, and a twin model with complete structure, rigorous logic, and synchronized status with the actual equipment is generated, providing model support for subsequent operational logic deduction and real-time linkage.

[0234] It should be understood that although the steps in the flowcharts of the above embodiments are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the above embodiments may include multiple steps or multiple stages. These steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the steps or stages in other steps. It is understood that the steps in different embodiments can be freely combined as needed, and all non-contradictory solutions formed by such combinations are within the scope of protection of this application.

[0235] In one exemplary embodiment, a computer device is provided, which may be a terminal, and its internal structure diagram may be as follows: Figure 10As shown, the computer device includes a processor, memory, input / output interface, communication interface, display unit, and input device. The processor, memory, and input / output interface are connected via a system bus, and the communication interface, display unit, and input device are also connected to the system bus via the input / output interface. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The input / output interface is used for exchanging information between the processor and external devices. The communication interface is used for wired or wireless communication with external terminals; wireless communication can be achieved through Wi-Fi, mobile cellular networks, Near Field Communication (NFC), or other technologies. When the computer program is executed by the processor, it implements a data governance and model building system for multimodal power supply in heavy-haul railways. The display unit is used to form a visually visible image and can be a display screen, projection device, or virtual reality imaging device. The display screen can be an LCD screen or an e-ink screen. The input device of the computer device can be a touch layer covering the display screen, or buttons, trackballs, or touchpads set on the casing of the computer device, or external keyboards, touchpads, or mice, etc.

[0236] Those skilled in the art will understand that Figure 10 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0237] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of the relevant data must comply with relevant regulations.

[0238] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. Any references to memory, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile memory and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM). The databases involved in the embodiments provided in this application may include at least one type of relational database and non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the embodiments provided in this application may be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, quantum computing-based data processing logic devices, artificial intelligence (AI) processors, etc., and are not limited to these.

[0239] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this application.

[0240] The embodiments described above are merely illustrative of several implementation methods of this application, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of this patent application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this application should be determined by the appended claims.

Claims

1. A data governance and model building system for multimodal power supply in heavy-haul railways, characterized in that, The system includes a data classification module, a semantic parsing module, an association alignment module, and a model building module, wherein: The data classification module is used to classify the raw power supply data of multiple modes based on the mode type, and obtain the mode data corresponding to each mode type; The semantic parsing module is used to convert the format of each modal data to obtain standardized data, perform semantic parsing on each standardized data to obtain semantic tags, and associate the standardized data with the semantic tags. The association and alignment module is used to associate and align the standardized data in multiple standard dimensions to obtain regularized data. The model building module is used to perform multimodal feature fusion and semantic association modeling based on the regularized data and the associated semantic tags to obtain a fused semantic feature set, and to build a business knowledge model based on the fused semantic feature set; the business knowledge model is used to build a digital twin model of the power supply system in the target area.

2. The system according to claim 1, characterized in that, The data classification module, when classifying the raw power supply data of multiple modes based on modal type to obtain the modal data corresponding to each modal type, is specifically used for: Identify structured and unstructured data in the raw power supply data; Extract core fields from the structured data and annotate the source of the core fields to obtain structured labeled data; The unstructured data is filtered for validity based on its data type to obtain unstructured categorized data. The structured labeled data and the unstructured classified data are grouped and integrated according to the modality type to obtain the modality data corresponding to each modality type.

3. The system according to claim 2, characterized in that, The model building module, when performing multimodal feature fusion and semantic association modeling based on the regularized data and associated semantic labels to obtain a fused semantic feature set, is specifically used for: Multimodal feature extraction is performed on each of the regularized data to obtain a basic feature set; the basic feature set includes equipment attribute features, operating status features, fault characterization features, and environmental correlation features; The basic feature set is subjected to feature normalization and noise reduction to obtain a purified feature set; Based on business relationships, construct a feature association matrix, and generate feature association data based on the feature association matrix; The feature-related data is fused with the purification feature set to obtain a preliminary fused feature set; Redundancy is removed from the preliminary fused feature set to obtain the fused semantic feature set.

4. The system according to any one of claims 1 to 3, characterized in that, The model building module, when constructing a business knowledge model based on the fused semantic feature set, is specifically used for: Element extraction is performed on the fused semantic feature set to obtain multiple core knowledge elements; The core knowledge elements are classified to determine the business knowledge elements corresponding to each business type. Based on the business knowledge elements corresponding to each of the aforementioned business types, as well as the hierarchical relationships, logical rules, and associations among the knowledge elements, knowledge modeling is performed to obtain the knowledge model corresponding to each of the aforementioned business types. The business knowledge model is constructed by integrating the knowledge models corresponding to each of the aforementioned business types.

5. The system according to claim 4, characterized in that, The model building module, when constructing a digital twin model of the power supply system within the target area based on the business knowledge model, is specifically used for: Obtain geographical distribution information, equipment ledger information, and basic data of each device of the power supply system within the target area; From the business knowledge model, attribute data of each device is extracted, and knowledge mapping data of each device is generated based on the attribute data of each device; the attribute data includes entity attributes, structural features, and business association rules. The basic data of the same device is associated with the knowledge mapping data to generate device knowledge matching data for each device; Based on the device knowledge matching data of each device, a three-dimensional geometric framework model of the power supply system in the target area is constructed. The digital twin model of the power supply system within the target area is obtained by configuring the parameters of the three-dimensional geometric framework model according to the business knowledge model; the parameter configuration includes the configuration of the business knowledge elements and the configuration of the logical rules.

6. The system according to claim 5, characterized in that, The model building module, based on the device knowledge matching data of each device, constructs a three-dimensional geometric framework model of the power supply system within the target area, specifically for: Extract the geometric structure information of each device from the device knowledge matching data of each device; The geometric structure information of each device is parsed to determine the geometric constraint data of each device; the geometric constraint data includes the structure type, spatial assembly logic, and positional constraint relationship with adjacent devices. Based on the geometric constraint data of each device, a three-dimensional sub-model of each device is generated; Based on the equipment knowledge matching data of each device, the three-dimensional sub-models of each device are assembled to obtain a three-dimensional geometric framework model of the power supply system in the target area.

7. The system according to claim 5 or 6, characterized in that, The model building module, when configuring parameters of the three-dimensional geometric framework model based on the business knowledge model to obtain the digital twin model of the power supply system within the target area, is specifically used for: From the business knowledge model, knowledge configuration data for each device in the power supply system within the target area is extracted; the knowledge configuration data includes business knowledge elements corresponding to device attributes, operating logic rules, and fault judgment criteria. From the three-dimensional geometric framework model, extract the entity identification data of each device. The entity identification data includes the device's identification information, geometric feature parameters, and spatial location information. The knowledge configuration data of each device is associated with the entity identification data, and the knowledge configuration data is filled into the entity nodes of the corresponding devices in the three-dimensional geometric frame model to obtain the parameter configuration model; The parameter configuration model is instantiated and encapsulated to generate the instantiated digital twin model.

8. The system according to claim 7, characterized in that, The system also includes an update module for: Based on the business knowledge model, semantic parsing, feature extraction and state recognition are performed on the target power supply data to obtain state parsing data; The target power supply data is multimodal power supply data collected for the power supply system of the target area; Determine the target device corresponding to the state parsing data in the digital twin model; The state parsing data is used to update the state of the entity node corresponding to the target device in the digital twin model to obtain the updated digital twin model; The updated digital twin model is subjected to operational logic deduction to generate state deduction data of associated devices, and the state deduction data is associated with the update information of the digital twin model.

9. The system according to claim 8, characterized in that, The business knowledge model includes state determination rule knowledge and semantic rule knowledge; the update module, when performing semantic parsing, feature extraction, and state recognition on the target power supply data based on the business knowledge model to obtain state parsing data, is specifically used for: Using the semantic rule knowledge, modal classification and semantic annotation are performed on the target power supply data to generate semantically annotated data; Feature extraction is performed on the semantically labeled data to obtain real-time feature data; The state determination rules are used to compare and analyze the real-time feature data to obtain the state recognition result; The status recognition result is associated with the corresponding device identifier, time information and spatial location information to obtain the status parsing data.

10. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the system according to any one of claims 1 to 9.

Images