Intelligent processing method and system for secondary system of transformer substation

Abstract

The invention belongs to the technical field of substations, and particularly relates to an intelligent processing method and system for a substation secondary system. The processing method comprises the following steps: performing secondary drawing intelligent identification based on deep learning; automatically constructing a digital model based on an identification result; performing multi-dimensional loop analysis based on a digital model; the analysis result and the model data generated in the step S3 are published through a mobile micro-application based on an i state grid platform, and safe, efficient and immersive access and interaction of field personnel to the secondary system digital twin model are achieved. By fusing computer vision, deep learning, power system modeling and mobile internet technologies, intelligent analysis of a secondary drawing, automatic construction of a digital model, deep analysis of a multi-dimensional topological relation and immersive interaction of an analysis result at a mobile terminal are realized, and the design, operation and management efficiency of a secondary system is comprehensively improved.

Description

Technical Field

[0001] This invention belongs to the field of substation technology, specifically relating to intelligent processing methods and systems for substation secondary systems. Background Technology

[0002] With the deepening of smart grid construction, the complexity of substation secondary systems is increasing daily. Traditional secondary system design, operation, and management heavily rely on manual drawing interpretation, manual accounting, and on-site verification, resulting in inefficiency, error-proneness, and untimely information updates. Although existing technologies have emerged that rely on drawing recognition methods based on OCR or simple image processing, they generally suffer from poor adaptability, inability to understand electrical logic relationships, and the inability to directly use the recognition results for structured modeling. Furthermore, the construction of digital models of secondary systems (such as SPD models) is mostly a manual or semi-automatic process, which is time-consuming, labor-intensive, and disconnected from the real-time state of the physical world.

[0003] Currently, secondary operation and maintenance personnel mainly employ traditional methods such as inspections and scheduled maintenance. Accurate and complete as-built drawings are the fundamental guarantee for the orderly conduct of daily maintenance, troubleshooting, and technical upgrades in substation secondary systems. However, due to limitations in management models, archiving conditions, current operational status, and intelligent technology, substation secondary system drawings are often poorly archived and inconsistent with reality. This seriously affects the on-site operations of frontline teams and threatens the safe and stable operation of the power grid. The main issues are as follows: (1) Decentralized management model: As-built drawings are managed separately by various work teams, which restricts the unified source management and maintenance of drawings; (2) Poor archiving conditions: Some paper drawings are blurred or severely damaged due to long storage time or poor preservation conditions, which affects their use. (3) Inconvenient modification and maintenance: In order to ensure the consistency between drawings and reality, manual modification on as-built drawings is commonly used, which is not convenient for unified maintenance and identification, and is also very easy to cause chaos in the management of drawing versions; (4) Lack of intelligent means: The secondary maintenance process lacks technical support means such as source control, editing, convenient query, and digital analysis, which is not conducive to the efficient development of secondary maintenance work.

[0004] Given the above issues, efficiently and safely integrating massive amounts of secondary system data with on-site mobile maintenance operations presents a significant technical challenge. Therefore, there is an urgent need for an integrated solution capable of automatically converting physical drawings into digital models and performing in-depth analysis and applications. Summary of the Invention

[0005] The purpose of this invention is to provide a method and system for intelligent identification, digital construction, and mobile interconnection analysis of substation secondary systems based on the fusion of deep learning and multidimensional models, addressing the problems existing in the prior art. By integrating computer vision, deep learning, power system modeling, and mobile interconnection technologies, it achieves intelligent parsing of secondary drawings, automatic construction of digital models, in-depth analysis of multidimensional topological relationships, and immersive interaction of analysis results on mobile devices, comprehensively improving the design, operation, and management efficiency of secondary systems.

[0006] The technical solution of this invention is: A smart processing method for substation secondary systems based on the fusion of deep learning and multidimensional models includes the following steps: S1: Intelligent Recognition of Secondary Drawings Based on Deep Learning: A hybrid deep learning model integrating YOLOv5 object detection and PaddleOCR text recognition is used to preprocess, divide, detect feature symbols, and recognize text in the drawing frame of the input secondary system CAD or PDF drawings; in response to the differences in drawing layout, a point-to-surface approach is used to analyze the drawing structure and extract electrical symbols and wiring information such as terminal block wiring, fiber optic wiring, device backplane, and circuit breaker pressure plate; S2: Automatic construction of digital model based on recognition results: Based on the recognition results of step S1, according to the SPD model structure defined in the "Logical Model Specification for Power Transmission and Transformation Engineering" (DL / T 2765-2024), the identified electrical components and connection relationships are automatically converted into a standardized, machine-readable structured data model through electrical logic analysis algorithms, completing the automatic conversion from secondary drawings to digital models (such as XML format); S3: Multi-dimensional loop analysis based on digital model: Quickly parse the digital model constructed in step S2, and adaptively and dynamically generate a visual view of the hierarchical distribution of equipment, the topology of cabinet equipment relationships, the topology of switch network, the information flow topology, and the full path of secondary loops based on the "equipment topology analysis" and "full path analysis" algorithms. S4: Mobile Internet-based Visual Interactive Application: The analysis results and model data generated in step S3 are published through a mobile micro-application based on the iGuowang platform. The mobile application adopts a UI interaction design optimized for operation and maintenance scenarios, and uses data encryption and secure channel technology to enable on-site personnel to access and interact with the digital twin model of the secondary system securely, efficiently, and immersively.

[0007] Preferably, in step S1, a few-sample training technique is used to optimize the deep learning model in order to solve the problem of insufficient labeled samples in engineering practice.

[0008] Preferably, in step S2, the structured data model is an SPD model conforming to the DL / T 2765-2024 standard, which realizes a standardized description of secondary loop information.

[0009] Preferably, in step S3, the generation of the visualization view supports user interactive operations, including zooming, rotating, clicking to highlight, and path tracing.

[0010] Preferably, in step S4, the data synchronization between the mobile terminal and the cloud adopts an incremental update and conflict resolution mechanism to ensure data consistency in a weak network environment.

[0011] A substation secondary system intelligent processing system based on the fusion of deep learning and multidimensional models, used to implement the above method, includes: The drawing intelligent recognition module is used to execute step S1, load the deep learning model, and complete the extraction of drawing information. Digital modeling engine: used to execute step S2, automatically building a digital model based on the recognition results and standard specifications; Multidimensional Analysis Service Module: Used to execute step S3, providing services such as topology analysis and path analysis and generating visualization views; Mobile Internet Application Module: Used to execute step S4, providing data access, visualization rendering and interactive interface for mobile terminals. The mobile internet application module is embedded in the iGuowang APP and provides services in the form of micro-applications. Central database: Used to store identification results, digitization models, and analysis results data.

[0012] Specifically, the mobile internet application module is embedded in the iGuowang APP and provides services in the form of a micro-application.

[0013] Specifically, the drawing intelligent recognition module is deployed on a server equipped with a GPU to accelerate the inference calculation of the deep learning model.

[0014] Traditional drawing management methods are inefficient and error-prone due to heavy reliance on manual operations. The technical solution provided in this application, by introducing digital drawing technology, enables rapid retrieval and sharing of drawing information, significantly improving work efficiency and supporting multi-user collaborative work, thus shortening project cycles. Simultaneously, this transformation reduces the printing, storage, and transportation costs of paper drawings, saving maintenance units substantial expenses in the long term and effectively preventing economic losses from lost or damaged drawings. Furthermore, digital drawings enhance version control of substation drawing data through the allocation of different permissions to different personnel, ensuring improved efficiency in on-site operation and maintenance work and reducing maintenance costs. This system promotes the digital transformation of the power grid, creates significant economic value for the power grid, and fosters its sustainable development.

[0015] The beneficial effects of this invention are: 1. High degree of automation: It achieves full-process automation from "drawings" to "data" to "application," greatly reducing manual intervention and time spent on manual drawing processing, thus improving work efficiency in substation engineering design, construction, and operation and maintenance. Simultaneously, drawing recognition technology can more accurately identify information on drawings, reducing the error rate caused by human factors and thereby improving project quality. 2. Deep intelligence: Employing advanced deep learning models, it not only recognizes graphic text but also understands electrical logic relationships with high accuracy. 3. Standardization and normalization: The output results strictly adhere to industry standards (DL / T 2765-2024), facilitating integration and promotion. 4. Multidimensional analysis: It provides multi-dimensional, panoramic analytical views, including physical connections, information flow, and network topology, offering strong insights. 5. Mobile application: By empowering mobile terminals through the national-level platform (iGuowang), it achieves precise delivery of knowledge to the field, demonstrating high practical value. Attached Figure Description

[0016] Figure 1 This is a schematic diagram of the drawing recognition process described in this invention.

[0017] Figure 2 This is a schematic diagram of the SPD model structure of the present invention; Figure 3 This is the logical deployment diagram of iGuo.net. Detailed Implementation

[0018] The technical solution of the present invention will be described in detail below with reference to the accompanying drawings and specific embodiments.

[0019] Example 1 This embodiment provides an intelligent processing method for substation secondary systems based on the fusion of deep learning and multidimensional models, including the following steps: S1: Intelligent Recognition of Secondary Drawings Based on Deep Learning: A hybrid deep learning model integrating YOLOv5 object detection and PaddleOCR text recognition is used to preprocess, divide, detect feature symbols, and recognize text in the drawing frame of the input secondary system CAD or PDF drawings; in response to the differences in drawing layout, a point-to-surface approach is used to analyze the drawing structure and extract electrical symbols and wiring information such as terminal block wiring, fiber optic wiring, device backplane, and circuit breaker pressure plate; This study analyzes the commonalities and differences in computer vision representation of CAD and PDF drawings for secondary systems. Drawings are classified using methods such as drawing area segmentation, feature symbol detection, and text recognition within drawing frames. For different types of drawings, a convolutional neural network is constructed, employing artificial intelligence techniques such as object detection, morphological analysis, and text recognition to identify and extract information such as electrical symbols and wiring connections.

[0020] To address the differences in drawing layouts, a point-to-surface approach is used to analyze and extract objects from secondary drawings. This enables the identification and extraction of design drawing information such as terminal block wiring, fiber optic wiring, device backplanes, and accessory information. It also enables the identification and extraction of physical information such as actual secondary external circuit wiring, switchgear circuit breakers, and pressure plates. This lays the underlying technical foundation for building digital models of secondary system equipment accessory asset information and cable wiring ledger information. Figure 1 As shown.

[0021] S2: Automatic Construction of Digital Model Based on Recognition Results: Based on the recognition results of step S1, and according to the SPD model structure defined in the "Logical Model Specification for Power Transmission and Transformation Engineering" (DL / T 2765-2024), as follows... Figure 2 As shown, through electrical logic analysis algorithms, the identified electrical components and connection relationships are automatically converted into standardized, machine-readable structured data models, completing the automatic conversion from secondary drawings to digital models (such as XML format); based on the drawing recognition results, according to the hierarchical relationship, model element definition, attribute values, etc. of the power transmission and transformation logic model specification file structure, the electrical logic analysis algorithm realizes the automatic modeling of primary and secondary equipment, cabinets, accessories, and the entire station circuit, completing the XML structured expression of text symbols and circuit logic in secondary drawings, realizing the conversion from secondary drawings to digital models; S3: Multi-dimensional loop analysis based on digital model: The digital model constructed in step S2 is quickly analyzed. Based on the "equipment topology analysis" and "full path analysis" algorithms, the system adaptively and dynamically generates a visual view of the hierarchical distribution of equipment, the topology of cabinet equipment relationships, the network topology of switches, the information flow topology, and the full path of secondary loops. The quick analysis specifically refers to the analysis of loop-related information such as the hierarchical structure of substations, bays, small rooms, cabinets, equipment, components, and terminals, as well as the wiring of fiber jumpers within the cabinet, terminal block wiring, optical cables, electrical cables, and cable core wiring between cabinets, based on the digital model. The generation of the visual view supports user interactive operations, including zooming, rotation, highlighting, and path tracing.

[0022] S4: Visual Interactive Application Based on Mobile Internet+: The analysis results and model data generated in step S3 are published through a mobile micro-application based on the iGuoWang platform. The mobile application adopts a UI interaction design optimized for operation and maintenance scenarios and uses data encryption and secure channel technology to enable on-site personnel to securely, efficiently, and immersively access and interact with the digital twin model of the secondary system. The iGuoWang platform is the underlying foundational platform for mobile interactive applications facing the Internet, providing functions such as a unified mobile portal, mobile application store, mobile device and application management, etc., to achieve unified access to Internet mobile terminals. The iGuoWang platform provides operational support for connected applications, including application interfaces, network channels, and security controls, as well as management support, including application management and monitoring analysis.

[0023] Specifically, step S1, the methods for identifying drawing types, drawing styles, electrical symbols, and electrical wiring for multiple types of drawings, include the following steps: First, based on digital image filtering and morphological feature recognition, secondary drawing structure analysis and coordinate extraction are performed. The commonalities and differences of electrical secondary drawings of different qualities in computer vision are analyzed. Image processing tools such as OpenCV are applied to perform morphological preprocessing on the drawings. The effect of optimizing drawing quality through noise reduction filtering, histogram equalization and other means is verified. A drawing classification method is designed by dividing the drawing area, detecting feature symbols and recognizing text in the drawing frame. A point-to-surface combined drawing recognition technology solution is studied for the differences in drawing layout.

[0024] Then, an artificial intelligence detection model is trained based on a deep learning framework for feature-based target detection. The YOLOv5 target detection algorithm and the PaddleOCR text recognition algorithm are integrated to build a deep learning neural network. Typical electrical secondary drawings are processed according to the target classification as a deep learning training set. Different training parameters are used to train and test recognition models at different levels. Multi-scale training methods and image pyramid sampling techniques are employed to improve the accuracy of primitive symbol recognition. The performance differences of detection models under different models and frameworks are compared and analyzed. The fusion technology of different detection models in the drawing recognition environment is studied. The detection model is iterated and optimized repeatedly using a large amount of engineering dataset. Considering the limited training samples in engineering implementation, model training methods and model optimization techniques in small-sample scenarios are studied. The hybrid deep learning model in step S1 is optimized using a small-sample training strategy that combines pre-trained model fine-tuning with synthetic data augmentation to solve the problem of insufficient labeled samples in engineering practice. The synthetic data is generated by procedurally combining electrical symbol templates and applying a style transfer network.

[0025] Few-Shot Learning employs a hybrid strategy based on fine-tuning of pre-trained models and synthetic data augmentation, as follows: 1. Pretraining + Transfer Learning: First, object detection models (such as the backbone network of YOLOv5) and OCR models are pre-trained on large general datasets (such as ImageNet and COCO) to enable them to obtain powerful general feature extraction capabilities.

[0026] Then, the pre-trained model is fine-tuned using a limited set of labeled electrical symbols and drawing text samples. A smaller learning rate and hierarchical learning rate strategy are employed during fine-tuning to protect the general features acquired during pre-training from being compromised, while adapting to the unique characteristics of electrical drawings.

[0027] 2. Synthetic Data Augmentation: Synthesis based on schematic templates: Create a library of common electrical symbols (circuit breakers, relays, terminals, etc.) and drawing frame templates. Through a programmatic method, randomly combine these symbols, connecting lines, and background drawing frames to generate a large number of realistic composite drawings and automatically generate corresponding annotation files (such as XML and JSON).

[0028] Style Transfer: Using style transfer networks (such as CycleGAN), the style of a small number of real drawing samples is applied to the synthetic drawing, making the synthetic data closer to the real drawing in terms of texture, noise, color, etc., and reducing the domain gap during model training.

[0029] 3. Advanced data augmentation techniques: Beyond traditional rotation, scaling, and cropping, it employs enhancement techniques such as CutMix, MixUp, and Mosaic. These techniques can mix multiple samples in a batch of data, greatly increasing data diversity and allowing the model to see more diverse feature combinations even with limited samples, effectively preventing overfitting.

[0030] Specific solutions for model fusion In step S1, the YOLOv5 object detection algorithm and the PaddleOCR text recognition algorithm are fused through cross-validation: the text information recognized by PaddleOCR is fed back as semantic features to the YOLOv5 classification module to optimize the final classification result of electrical symbols; simultaneously, the recognition result is verified and corrected by a post-processing module based on electrical domain knowledge rules. The technical challenge lies in how to deeply fuse the effective information from the two independent models (YOLOv5 object detection and PaddleOCR text recognition), rather than simply concatenating them, to improve overall recognition accuracy and contextual understanding. The solution adopted in this embodiment is to construct a collaborative fusion framework based on cross-validation and feedback optimization. Detailed implementation method: 1. Cross-Modal Information Passing: The object detection model (YOLOv5) first locates all candidate regions in the drawing (such as electrical symbols, drawing frames, and text labels).

[0032] For each detected text label region, its image patch is fed into the PaddleOCR model for recognition.

[0033] Key fusion step: The text result recognized by OCR (such as the device code "7BK20") is used as a high-level semantic feature and fed back to the object detection model. For example, this text feature can be concatenated with the visual features of the corresponding area and input into a classification head to more accurately determine the type of electrical symbol in the area (for example, identifying a symbol labeled "7BK20" as a "protective device" rather than a regular frame).

[0034] 2. Rule-based post-processing optimization: Establish a domain knowledge rule base for electrical drawings (such as "equipment codes usually start with numbers" and "terminal block numbering follows a specific sequence").

[0035] The initial recognition results of the OCR are verified and corrected. For example, if the OCR misidentifies "1T1" as "IT1", the post-processing rules will correct it to "1T1" based on the context (it is located next to a terminal symbol).

[0036] The system uses the detected symbol types and recognized text information to perform logical consistency checks. For example, if a region is detected as "relay coil," but the text next to it is "closing circuit," the system will mark that there may be a recognition conflict and initiate more complex reasoning or prompt manual review.

[0037] In step S2, the intelligent recognition and automatic digitization of secondary electrical drawings based on deep learning includes the following steps: First, collect and organize the design technical specifications related to electrical secondary drawings, analyze the distribution patterns of information such as graphics, text, and relationships in the drawings, and sort out the auxiliary role of factors such as graphic features, connecting line segment features, and electrical symbol morphological features in judging logical relationships. Then, study the logical reasoning algorithm for recognizing the electrical connection relationships represented by the electrical secondary drawings. Finally, utilize dynamic parameters and multi-level logical classification to improve the adaptability of the drawing recognition algorithm in drawings with different design styles.

[0038] Then, the methods for establishing relationships between different drawings, different electrical components, and different circuits were studied, and secondary drawings were analyzed in depth from multiple perspectives, including principle description, equipment description, and cable description. Based on the recognition results of signal description recognition, equipment description recognition, cable description recognition, and graphic frame text recognition, the syntax and semantics of the station-level physical model were studied according to the power transmission and transformation logic model specification. Based on the hierarchical relationship of its file structure, model element definition, attribute values, etc., information such as all secondary equipment, cabinets, optical cable circuits between equipment cabinets, cable circuits between equipment cabinets, and cabinet terminal block diagrams in the substation were modeled and analyzed. The XML structured expression of text symbols and circuit logic in secondary drawings was completed, realizing the automatic conversion of secondary drawings into digital models. The structured data model is an SPD model conforming to the DL / T 2765-2024 standard, realizing the standardized description of secondary circuit information.

[0039] Multi-dimensional loop analysis based on digital model: The digital model constructed in step S2 is quickly analyzed. Based on the "equipment topology analysis" and "full path analysis" algorithms, the system adaptively and dynamically generates a visual view of the hierarchical distribution of equipment, the topology of cabinet equipment relationships, the network topology of switches, the information flow topology, and the full path of secondary loops. The quick analysis specifically refers to the analysis of loop-related information such as the hierarchical structure of substations, bays, small rooms, cabinets, equipment, components, terminals, etc., the internal fiber jumpers, terminal block wiring, and the inter-cabinet optical cables, electrical cables, and cable core wiring based on the digital model.

[0040] Step S3, Multi-dimensional Loop Analysis Based on Digital Model: First, the method for constructing a secondary loop information dictionary is studied. Secondary loops contain a large number of proper nouns, such as equipment ledger information, installation unit information, equipment board terminal configuration information, cable connection information, and loop function information. Existing open-domain Chinese dictionaries cannot meet the needs of constructing a knowledge graph system for substation secondary loops. Therefore, this study employs techniques such as unsupervised learning-based topic word recognition and new word discovery in the main equipment domain based on a basic vocabulary to construct a dictionary for secondary loop information. In implementation, for new secondary loop text data, a vocabulary-based word segmentation model is first used to segment it. Then, a binary classifier is used to determine whether a segmented word is a new word. If so, it is added to the topic dictionary. This approach allows for the construction of a relatively complete and automatically updated secondary loop topic dictionary.

[0041] Then, the study uses D2R technology to map data from the secondary loop relational database to RDF (Resource Description Frame). Knowledge is extracted from data with different structures using entity recognition and event extraction techniques, forming knowledge (structured data) which is then stored in a knowledge graph, thus creating a secondary loop information structured dataset. Secondary loop information text has multiple expression methods, including object-oriented descriptions and connection relationship expressions. Based on standardized open-in / open-out interface expression rules, a bidirectional long short-term memory network model is proposed to model text sentences. Template matching, deep learning, and other methods are used to identify core entities within the secondary loop information.

[0042] Furthermore, for the physical circuit model, it enables rapid analysis of circuit-related information such as the hierarchical structure of substations, bays, small rooms, cabinets, equipment, components, and terminals, as well as fiber optic jumpers, terminal block wiring, optical cables, electrical cables, and cable core wiring between panels; for the information circuit model, it enables rapid analysis of virtual circuits and mapped port information at the entire station process layer; based on the digital design results of physical circuits and information circuits, it realizes the correlation and coupling after multi-dimensional information analysis, laying the data access foundation for the panoramic visualization display of the secondary system.

[0043] Based on multi-dimensional information fusion, an automatic generation algorithm for visual views such as network topology, device topology, information flow, and physical loops is designed. The algorithm supports automatic identification and extraction of key elements (such as devices, connection lines, ports, etc.), and dynamically generates clear and intuitive visual views based on their attributes (such as type, status, location, etc.) and interrelationships.

[0044] Finally, for the network topology view, the algorithm can display the network connections between devices, including the connection status and paths of network devices such as switches and secondary devices. The device topology view focuses on displaying the internal structure and composition of devices, including the hierarchical structure of devices and the connection relationships between various components. Through interactive methods such as zooming and rotating, users can gain a deeper understanding of the internal structure and layout of devices. The information flow view focuses on the transmission and flow of information, including data collection, processing, transmission, and storage. Through the visual view, users can clearly see the flow and transmission process of information, as well as the correlations and dependencies between various links. The physical full-loop view combines the characteristics of physical connections and information flow, displaying the physical connection relationships and information transmission paths of the entire system.

[0045] In step S4, in the mobile internet-based visual interactive application, the data synchronization between the mobile terminal and the cloud adopts an incremental update and conflict resolution mechanism to ensure data consistency in a weak network environment.

[0046] When mobile terminal devices are used for operation and maintenance work, in addition to the hardware itself needing to meet specific requirements and specifications, the application UI interaction should also be specifically designed, which is reflected in the following aspects: 1) Functional design: Investigate the usage needs of operation and maintenance personnel on mobile terminals, analyze and abstract the needs, and design from the application function level to ensure that the application meets basic functional usability; 2) User-friendly UI interaction design: Fully consider the characteristics and usage methods of operation and maintenance personnel, study the differences between mobile terminal UI and typical PC terminal UI screens, and optimize the design; 3) UI screen compatibility design: It is necessary to consider the differences between different terminal models (such as screen resolution, screen size, graphics rendering performance, etc.) and design compatibility solutions.

[0047] Due to the high mobility of mobile terminals, there are significant risks to data security. In order to eliminate these risks, in addition to optimizing management methods, necessary technical support is also required: 1) Data encryption algorithms: research data encryption algorithms, select appropriate encryption algorithms, and seek a balance between encryption strength and system performance; 2) Data transmission channel security: combine the iGuowang platform to study security solutions for traditional data channels.

[0048] In step S4, data synchronization between the mobile terminal and the cloud adopts an incremental update mechanism based on operation logs: the cloud monitors data changes and generates operation instruction logs; during mobile synchronization, the cloud only sends incremental operation instruction packets, and the mobile terminal updates local data by replaying the instructions; data version conflicts are detected through a version vector mechanism and resolved according to predefined business rules or manual intervention. The technical challenge of incremental updates in this embodiment is: how to ensure efficient and consistent synchronization of digital twin model data between the mobile terminal and the cloud in an unstable mobile network environment, while reducing traffic consumption and synchronization latency. The solution adopted in this embodiment is: using an incremental synchronization and conflict resolution mechanism based on operation logs and version vectors. Detailed implementation method: 1. Incremental data capture and packaging: In the cloud, changes (additions, deletions, modifications) to all model data records (such as a device or a connection line) in the central database are monitored and converted into atomic operation instructions (e.g., {operation:'update',id:'Device_A', field:'status', value:'offline'}), which are then stored in the operation log in chronological order.

[0050] Each time a mobile device initiates a synchronization request, it carries a version identifier for the local data (such as the timestamp or sequence number of the last synchronization).

[0051] 2. Intelligent differential synchronization: After receiving the synchronization request from the mobile device, the cloud compares the mobile device's version identifier with the cloud's operation logs, and only packages all operation instructions generated after that point in time into an incremental update package (Delta Package) and sends it to the mobile device.

[0052] After receiving the incremental packets, the mobile device replays these operation commands sequentially to update the local data to the latest state. This method transmits very little data, making it ideal for weak network environments.

[0053] 3. Conflict Detection and Resolution: Conflict detection: Employs a version vector mechanism. A version vector is maintained for each data record, recording the number of modifications made to it by each client (including cloud and multiple mobile devices). When two clients simultaneously modify the same data, the version vector can quickly detect the conflict.

[0054] Conflict resolution: Strategy 1 (Automatic Resolution): Define rules such as "Last Write Wins" or automatic merging strategies based on business logic (e.g., the latest device status update takes precedence).

[0055] Strategy 2 (Manual Resolution): When the automatic resolution strategy cannot handle the situation (such as two users modifying two important attributes of the same device), the system saves the conflict information (such as values ​​of different versions) and prompts the on-site operation and maintenance personnel through the UI interface, allowing them to manually choose which version to keep or merge.

[0056] Example 2 This embodiment provides an intelligent processing system for substation secondary systems based on the fusion of deep learning and multidimensional models, used to implement the method described in the embodiment, including: The drawing intelligent recognition module is used to execute step S1, load the deep learning model, and complete the extraction of drawing information. Digital modeling engine: used to execute step S2, automatically building a digital model based on the recognition results and standard specifications; Multidimensional Analysis Service Module: Used to execute step S3, providing services such as topology analysis and path analysis and generating visualization views; Mobile Internet Application Module: Used to execute step S4, providing data access, visualization rendering, and interactive interface for the mobile terminal. The mobile internet application module is embedded in the iGuowang APP in the form of a micro-application. Providing services such as Figure 3 The diagram below shows the logical deployment of the national grid. Central database: Used to store identification results, digitization models, and analysis results data.

[0057] The technical solution provided by this invention reduces the time spent manually processing drawings and improves the work efficiency in the design, construction, and operation and maintenance stages of substation engineering. Simultaneously, the drawing recognition technology can more accurately identify information on drawings, reducing the error rate caused by human factors, thereby improving project quality, ensuring the uniqueness of substation drawing file operations and transfers, enabling different operation and maintenance personnel to share substation drawing information, collaborate in real time, break down information barriers, and improve operation and maintenance efficiency within the substation.

[0058] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit them; although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications can still be made to the specific implementation of the present invention or equivalent substitutions can be made to some technical features without departing from the spirit of the technical solutions of the present invention, and all such modifications and substitutions should be covered within the scope of the technical solutions claimed in the present invention.

Claims

1. A method for intelligent processing of secondary systems in substations, characterized in that, Includes the following steps: S1: Intelligent recognition of secondary drawings based on deep learning: A hybrid deep learning model that integrates target detection algorithm and text recognition algorithm is used to preprocess secondary drawings, divide regions, detect feature symbols and recognize text in the drawing frame, and extract electrical symbols and wiring information; S2: Automatic construction of digital model based on recognition results: Based on the recognition results of step S1, and in accordance with the power transmission and transformation engineering logic model specifications, the identified information is automatically converted into a standardized structured data model through electrical logic analysis algorithms; S3: Multi-dimensional loop analysis based on digital model: The digital model constructed in step S2 is analyzed, and various visualization views are adaptively and dynamically generated based on topology analysis and path analysis algorithms; S4: Mobile Internet-based Visual Interactive Application: The analysis results and model data from step S3 are published and interacted with through a mobile platform micro-application. The mobile terminal adopts UI design and data security technology optimized for operation and maintenance scenarios.

2. The intelligent processing method according to claim 1, characterized in that, In step S1, the target detection algorithm is the YOLOv5 algorithm, and the OCR algorithm is the PaddleOCR algorithm.

3. The intelligent processing method according to claim 1, characterized in that, In step S1, the hybrid deep learning model is optimized and trained using few-shot training techniques.

4. The intelligent processing method according to claim 1, characterized in that, In step S2, the structured data model is an SPD model that conforms to the DL / T 2765-2024 standard.

5. The intelligent processing method according to claim 1, characterized in that, In step S3, the visualization view includes at least one of a device topology view, a network topology view, an information flow view, and a physical full loop view, and supports user interactive operation.

6. The intelligent processing method according to claim 1, characterized in that, In step S4, the mobile platform is the iGuowang platform, and the data synchronization between the mobile terminal and the cloud adopts an incremental update and conflict resolution mechanism.

7. A substation secondary system intelligent processing system for implementing the method of any one of claims 1-5, characterized in that, include: The drawing intelligent recognition module is used to load a deep learning model and extract drawing information. Digital modeling engine: used to automatically build digital models based on recognition results and standard specifications; Multidimensional Analysis Service Module: Used to provide topology analysis and path analysis services and generate visualization views; Mobile Internet Application Module: Used to provide data access, visualization rendering, and interactive interface for mobile terminals; Central database: Used to store identification results, digitization models, and analysis results data.

8. The intelligent processing system for substation secondary systems according to claim 7, characterized in that, The mobile internet application module is embedded in the iGuowang APP and provides services in the form of a micro-application.

9. The intelligent processing system for substation secondary systems according to claim 8, characterized in that, The drawing intelligent recognition module is deployed on a server equipped with a GPU to accelerate the inference calculation of the deep learning model.

Images