A digital and intelligent design management system for the entire process of power distribution network engineering based on multimodal fusion.
The multimodal integrated digital and intelligent design management system for the entire process of distribution network engineering solves the problems of data silos, model asynchrony, and inefficient acceptance in distribution network design. It achieves efficient data synchronization and automatic deviation verification, reduces construction rework and twin platform costs, and supports the digital and intelligent transformation of distribution network engineering.
Patent Information
- Application Number
- CN202611228373.3
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-08-13
- Publication Date
- 2026-11-17
AI Technical Summary
Existing power distribution network design tools lack visual online collaborative design capabilities, making it difficult to optimize line routes and equipment selection, difficult to synchronize 2D drawings and 3D models, resulting in high design change rates and high construction rework costs. Project acceptance relies on manual measurement, lacks drone real-scene data collection and AI fusion verification, and the deviation between design drawings and actual construction cannot be quantified. Digital asset libraries are not unified, data correlation is poor, and twin platforms are costly and difficult to implement.
The system adopts a multimodal fusion-based digital intelligent design and management system for the entire process of power distribution network engineering. This system includes multimodal survey data collaborative acquisition, multi-dimensional two-dimensional and three-dimensional linkage modeling, real-scene AI intelligent verification and digital asset library, static digital twin mapping of typical scenarios, and one-map four-state closed-loop scheduling and control. Through algorithms such as Beidou spatiotemporal registration, dynamic filtering, genetic optimization, hash bidirectional mapping, 3DPointNet+ViT dual-branch AI, scene lightweighting, and finite state machine, it achieves data unification, model synchronization, and automatic deviation verification.
It enables automatic identification of spatial interference in power lines, quantitative comparison of construction scenes with design drawings, reduces model surface area by 70%, and can build static twins of transformer substations and rural power grids without real-time measurement data. This improves the accuracy of survey data collection and AI recognition, shortens the design and acceptance cycle, and reduces the cost of twin deployment.
Smart Images

Figure CN123117939A_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the field of digital power distribution network technology, and in particular relates to a digital and intelligent design and management system for the entire process of power distribution network engineering based on multimodal fusion. Background Technology
[0003] In the power distribution network design phase, existing tools can only complete single-dimensional drawing, lacking the ability for visual online collaborative design. They cannot automatically perform multi-objective optimization and selection of line routes and power distribution equipment. Two-dimensional drawings and three-dimensional reality models can only be converted in one direction, and model modifications cannot be automatically synchronized. Issues such as line intersections and equipment installation space interference can only be checked manually one by one, resulting in a high design change rate and directly increasing subsequent construction rework costs. In the project acceptance and digital handover phase, current methods mainly rely on on-site manual measurement of key indicators such as tower tilt and cable sag for spot checks. There is a lack of automated verification methods combining drone real-scene data collection and AI fusion recognition. Deviations between design drawings and actual construction cannot be quantitatively determined. Survey, design, and acceptance data are stored in a scattered manner, lacking a standardized digital asset library linked by unique equipment identifiers. The digital handover results are characterized by chaotic formats and poor data correlation, making it difficult to support efficient subsequent operation and maintenance. While digital twin technology has been implemented in main grid scenarios, existing twin platforms generally rely on real-time operational measurement data streams, resulting in high computing power deployment costs. They lack differentiated lightweight data processing solutions for typical distribution network scenarios such as densely populated distribution areas and wide-area rural distribution networks. The high cost of modeling and rendering massive amounts of small distribution equipment makes large-scale application in municipal and county-level distribution network projects difficult. In summary, the industry currently lacks an integrated closed-loop technical solution that spans distribution network surveying and design, modeling linkage, real-world verification, digital asset accumulation, typical scenario twinning, and one-map-four-state data synchronization. Fragmented operation of various single-point tools and the inability to connect data across the entire chain severely restrict the effectiveness of digital and intelligent transformation of distribution network projects. This is also a key practical business challenge that the State Grid Pingdingshan Power Supply Company needs to address in this technology project. Summary of the Invention
[0004] The purpose of this invention is to provide a digital and intelligent design management system for the entire process of distribution network engineering based on multimodal fusion. This system automatically identifies spatial interference of lines, quantitatively compares the actual construction scene with the design drawings, sets up a logic to lock the editing permissions of major deviations and force rectification, and reduces the model surface area by 70% while retaining the complete electrical topology. It can build static twins of transformer substations and rural power grids without real-time measurement data, and completely solves the core defects of traditional technology such as data silos, model asynchrony, inefficient manual acceptance, high twin deployment cost, and lagging updates of four-state data.
[0005] To achieve the above objectives, the present invention provides the following technical solution: The digital and intelligent design management system for the entire process of power distribution network engineering based on multimodal fusion includes: a multimodal survey data collaborative acquisition subsystem, a multi-dimensional two-dimensional and three-dimensional linkage modeling subsystem, a real-scene AI intelligent verification and digital asset library subsystem, a typical scenario static digital twin mapping subsystem, and a one-map four-state closed-loop scheduling and control subsystem. The multimodal exploration data collaborative acquisition subsystem incorporates a BeiDou spatiotemporal registration transformation matrix, a dynamic sliding filter noise reduction algorithm, a multi-objective genetic optimization fitness function, and an RBAC hierarchical permission transfer state machine. It is used to unify coordinate benchmarks, reduce noise and clean the four-modal data (laser point cloud, aerial image, GIS coordinates, and equipment ledger), automatically generate the optimal power distribution network design scheme, and transfer it online in a hierarchical manner. The multi-dimensional two-dimensional and three-dimensional linkage modeling subsystem has a built-in hierarchical hash parameter mapping index table and an AABB bounding box two-level spatial conflict detection algorithm to establish a two-way real-time mapping relationship between two-dimensional primitive parameters and three-dimensional mesh, and automatically identify engineering spatial interference defects. The real-scene AI intelligent verification and digital asset library subsystem has a built-in dual-branch fusion recognition network and a map-real deviation quantification judgment function. It uses the device's unique ID to connect multi-modal data throughout the entire process to build a hybrid storage digital asset library. The typical scenario static digital twin mapping subsystem has a built-in scenario adaptive noise reduction weight allocation algorithm and equipment geometry lightweight downsampling rules. Based on the differentiated data processing logic of transformer area and rural power grid, it generates a lightweight static digital twin without real-time measurement dependence. The one-map four-state closed-loop scheduling and control subsystem incorporates a finite state machine for the engineering phase and an incremental differential graph model synchronization algorithm, automatically driving the four-state data flow of planning / design / construction / operation, and outputting standardized graph models to the provincial power grid one-map platform.
[0006] Preferably, the RBAC hierarchical permission flow state machine adopts multi-level dynamic authorization control logic. When a new project participation role is added, the corresponding data read / write, model modification, and asset export permissions are automatically matched. Moreover, each data modification and model export operation generates an immutable traceability log with timestamp, operator, and operation content.
[0007] Preferably, the dual-branch fusion recognition network is a 3DPointNet+ViT dual-branch fusion recognition network, and the image-real deviation quantization judgment function... , in, The absolute deviation value represents the absolute difference between the measured characteristic parameters of the actual construction and the theoretical design parameters. It is used to quantify the degree of deviation between the actual construction scene and the design drawings. For real-world measurement features, physical indicators are extracted from drone point clouds and aerial images by a 3DPointNet+ViT dual-branch fusion AI network, including key geometric parameters of the power distribution network such as tower tilt angle, conductor sag height, tower spacing, and equipment installation offset. Standard design parameters, stored in a two- or three-dimensional linkage model, are used to design standard theoretical quantities and to preset compliant theoretical values for distribution network engineering drawings. These values are derived from distribution network design specifications and project design schemes.
[0008] Preferably, two levels of deviation thresholds are set. , ,when > The system automatically locks the modeling subsystem's editing interface, builds a hybrid storage asset library with the device's unique ID as the primary key, stores structured parameters in a relational database, and stores 3D models / real-scene images in object storage. A three-level caching mechanism ensures that the asset archiving response time is ≤3 seconds.
[0009] Preferably, the scene adaptive noise reduction weight allocation algorithm has a built-in scene recognition trigger logic, which automatically distinguishes between two types of scenes, namely transformer area and rural power grid, and loads the corresponding noise reduction weight coefficients according to the engineering ledger; the equipment geometry lightweight downsampling rule distinguishes the different surface compression ratios of four types of equipment, namely main transformer, pole tower, branch box and cable, retains the complete and fine geometry of the main transformer, and greatly simplifies the surface of the cable, thus dynamically balancing the twin rendering efficiency and the equipment visualization accuracy.
[0010] Preferably, the finite state machine in the engineering phase has built-in cross-subsystem linkage triggering logic. When the incremental differential graph model synchronization task fails and reaches the retry limit, it automatically sends an exception flag back to the real-scene AI verification subsystem and suspends the process of adding new assets into the database. The finite state machine, digital asset database, and digital twin subsystem share the same globally unique ID index pool for the same device, realizing one-click traceability and association of planning, design, construction, operation and maintenance, and digital twin business data.
[0011] This invention also provides a method for intelligent digital design and management of the entire process of power distribution network engineering based on multimodal fusion, characterized by being executed by the aforementioned system and comprising the following steps: S1. Standardized Acquisition of Multimodal Exploration Data: Substitute the BeiDou spatiotemporal registration transformation matrix to unify the four-modal data coordinate system, and execute... Filtering out outlier sampling points based on fitness function After iteration to convergence, the optimal power distribution scheme is output. The RBAC state machine manages the cross-departmental flow of design results according to role-based hierarchical control and retains operation logs. S2, Multi-dimensional 2D and 3D linkage modeling: Based on the hierarchical hash index, a bidirectional real-time mapping is established. The AABB two-level conflict detection algorithm is used to automatically identify and grade various spatial interference defects, and the high-brightness positions of defects in the 2D and 3D interface are updated synchronously. S3. Multimodal Real-Scene AI Mandatory Verification and Digital Asset Archiving: Drone 5G Backhaul Real-Scene Multimodal Data Input Dual-Branch Fusion Network Adaptively Adjusts Feature Weights to Extract Engineering Indicators, Substituting Them into... Quantitative comparison; if a major safety deviation is identified, all model editing interfaces are locked and a rectification list is pushed. After the rectification is verified, the entire chain data is collected by device ID and stored in the hybrid storage digital asset library. S4. Lightweight Static Digital Twin Construction for Different Scenarios: The system automatically identifies the project type and loads the corresponding noise reduction weights. or Filter redundant data, perform differential geometric downsampling processing according to equipment classification, and generate an interactive static digital twin of the substation / rural network that does not depend on real-time measurement. S5. Fully Automatic Closed-Loop Synchronization of One Map and Four States: After the finite state machine detects the complete completion and archiving mark, it automatically initiates the incremental differential map module synchronization task. If the synchronization fails, the linked asset library will suspend the entry and retry. After the standard is met, the standardized fusion map module will be pushed to the provincial power grid one map platform to realize closed-loop management and control of distribution network data without human intervention throughout the entire life cycle.
[0012] Preferably, step S3 includes a closed-loop verification process: after major safety deviations are rectified, the system automatically re-calls the 3DPointNet+ViT dual-branch network to identify real-world data, only verifying data with a difference between the two comparisons that is less than [value missing]. Asset archiving operations should only be performed at certain times to prevent quality loopholes such as manual filling of rectification records without on-site verification.
[0013] In summary, the beneficial technical effects of this invention are as follows: This invention constructs a multimodal fusion closed-loop architecture with deep coupling of five subsystems, and is equipped with a complete set of innovative algorithms, including BeiDou spatiotemporal registration, dynamic filtering, genetic optimization, hash bidirectional mapping, 3DPointNet+ViT dual-branch AI, scene lightweighting, and finite state machine incremental synchronization. Pilot verification in the Pingdingshan power distribution network project shows that the accuracy of survey data acquisition and component AI recognition both reach over 95%, 2D and 3D models achieve bidirectional synchronization within 300ms, asset archiving response ≤3s, map and model incremental synchronization ≤10s, and cross-system interface availability ≥99%. It automatically identifies line spatial interference, quantitatively compares construction realities with design drawings, sets up mandatory rectification logic by locking editing permissions for major deviations, and reduces model surfaces by 70% while preserving the complete electrical topology. Static twins of transformer substations and rural power grids can be built without real-time measurement data, completely solving the core defects of traditional technologies such as data silos, model asynchrony, inefficient manual acceptance, high twin deployment costs, and delayed updates of four-state data. Attached Figure Description
[0014] The accompanying drawings are provided to further illustrate the invention and form part of the specification, but do not constitute a limitation thereof. In the drawings: Figure 1 This is a schematic diagram of the structure of the digital and intelligent design and management system for the entire process of power distribution network engineering based on multimodal fusion in this embodiment. Detailed Implementation
[0015] The present invention will be further described in detail below with reference to the accompanying drawings.
[0016] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.
[0017] Please see Figure 1 The present invention provides a technical solution: The present invention is a digital and intelligent design and management system for the entire process of power distribution network engineering based on multimodal fusion. The hardware deployment is divided into front-end field acquisition terminals (drones, surveying tablets, lidar), cloud application servers, and distributed hybrid storage clusters. The software consists of five subsystems connected in series to form an end-to-end closed-loop link, namely: multimodal exploration data collaborative acquisition subsystem, multi-dimensional two-dimensional and three-dimensional linkage modeling subsystem, real-scene AI intelligent verification and digital asset library subsystem, typical scene static digital twin mapping subsystem, and one map four-state closed-loop scheduling and control subsystem.
[0018] The five subsystems share a global unique device ID index pool and a unified distribution network metadata standard library. The subsystems can communicate with each other in both directions, which solves the industry pain points of traditional distribution network design, modeling, acceptance, and twin tools being independent and fragmented, data not being interoperable, one-way conversion of 2D and 3D models, low efficiency of manual acceptance, and reliance on manual copying and updating of "one map with four states". It is suitable for the implementation of typical distribution network projects in Pingdingshan local transformer areas and rural power grids, and supports the dynamic data supply of a single map for the provincial power grid.
[0019] This system interfaces with various data acquisition terminals, including field mapping tablets, LiDAR, and drones. The software incorporates a BeiDou spatiotemporal registration transformation matrix and a dynamic sliding filter noise reduction algorithm. Multi-objective genetic fitness function The RBAC hierarchical permission flow state machine is implemented as follows: Unified coordinate transformation of multi-source and multi-modal data: Based on the BeiDou spatiotemporal registration transformation matrix, the three-dimensional coordinates of laser point clouds, the pixel coordinates of aerial images, the topographic coordinates of GIS, and the text parameters of equipment ledgers are unified to the 2000 National Geodetic Coordinate System, eliminating the problem of misalignment of coordinate references for multi-device acquisition; Data noise reduction and cleaning: The terminal uploads the raw data. The system automatically adjusts the dynamic sliding filter function based on ambient light and vegetation shading. Remove outliers caused by drift and occlusion, and output a standardized survey dataset. The accuracy rate of exploration data collection is consistently ≥95%; Optimizing Smart Distribution Network Solutions: Import multi-objective genetic optimization operators and iteratively calculate fitness. 10 consecutive generations Fluctuation less than the minimum value The iteration ends when the time is right, and the optimal route and transformer equipment selection scheme that takes into account cost, power supply distance and terrain constraints are output. RBAC hierarchical access control: The system has built-in multi-level dynamic authorization logic, which automatically assigns data read / write, model editing, and asset export permissions according to four roles: designers, reviewers, construction managers, and maintenance personnel. Every modification and export operation automatically generates an immutable traceability log with timestamp, operator ID, and operation content, and generates a standardized survey and design data package with a globally unique equipment ID, which is pushed to the downstream 2D and 3D modeling subsystems.
[0020] Multi-dimensional 2D / 3D linkage modeling subsystem: This subsystem receives standardized data packets output by the acquisition subsystem and has a built-in hierarchical hash parameter mapping index table and an AABB bounding box two-level spatial conflict detection algorithm. Specific implementation process: Multidimensional metadata binding: Read engineering attributes, equipment models, and GIS terrain metadata from data packets and store them in a unified distribution network metadata standard library; Two-dimensional and three-dimensional bidirectional real-time mapping construction: establishing two-dimensional drawing primitive parameters using a hierarchical hash index table. With 3D device mesh Bidirectional key-value mapping ↔ Modifying parameters on one end will automatically synchronize all associated elements on the other end within 300ms, completely solving the problem of traditional CAD only generating 3D in one direction and not synchronizing after modification. Two-level automatic spatial conflict detection: The first step is coarse-grained AABB bounding box pre-screening to quickly delineate areas with spatial interference; the second step is fine-grained 3D mesh intersection judgment to automatically distinguish three types of defects: insufficient conductor safety distance, equipment operating space encroachment, and tower corridor obstruction. The 3D coordinate points of the defects are highlighted synchronously on the 2D and 3D visualization interface, matched with the corresponding distribution network design specifications and rectification standards, and the defect rectification dataset is output.
[0021] Real-Scene AI Intelligent Verification and Digital Asset Repository Subsystem: This subsystem connects to the 5G drone real-time transmission channel and incorporates a 3DPointNet+ViT dual-branch fusion recognition network and an image-real-world deviation quantification judgment function. Supported by a hybrid distributed storage cluster, implementation method: Multimodal real-scene feature fusion extraction: Drones equipped with LiDAR and high-definition cameras collect point cloud and image data from the construction site, which are simultaneously transmitted to a dual-branch network; in outdoor backlighting and tree-occluded scenes, the weight of point cloud depth features is automatically increased while the weight of image texture is reduced; the AI recognition accuracy of towers and cable components is ≥95%, and measured parameters are output. ; Determination of Deviation Between Drawing and Actual Design: Retrieving Built-in Design Parameters from the 2D / 3D Model Calculate the deviation ; like ≤ If a general deviation is detected, a pop-up notification will be sent to the front end to prompt rectification, while the model remains editable; like > If a major safety deviation is detected, all editing interfaces of the 2D / 3D modeling subsystem will be automatically locked, and the defect coordinates, deviation values, and standard limit rectification list will be continuously pushed. Standardized digital asset storage: Using the unique ID of the global device as the primary key, a hybrid storage architecture is built. Structured electrical parameters are stored in a relational database, while 3D mesh models and aerial real-scene images are stored in object storage. Built-in three-level cache scheduling logic ensures that the asset archiving response time is ≤3s. Survey, design, real-scene verification, and completion acceptance data are all associated and collected to form a standardized digital asset library, which synchronizes data with the downstream digital twin subsystem.
[0022] A typical scenario static digital twin mapping subsystem. This subsystem reads full-link data from the digital asset library, and incorporates a scenario-adaptive noise reduction weight allocation algorithm and device geometry lightweight downsampling rules. It is compatible with both transformer substations and rural power grids. Implementation process: Automatic engineering scene recognition: Reads engineering ledger tags in the asset database, automatically distinguishes between urban power distribution areas and rural power distribution networks, and applies noise reduction weights accordingly. , ( ); The system filters massive amounts of redundant data from household meters and branch lines, while the rural power grid retains the original parameters of mountainous terrain and disaster relief lines. Differentiated and lightweight equipment processing: Differentiated surface downsampling is performed on four types of equipment: main transformer, tower, branch box, and cable; the main transformer retains a complete and fine geometric mesh, and the cable greatly simplifies redundant surfaces, reducing the number of model surfaces by 70% while fully preserving the electrical topology connectivity; Static digital twin generation: Based on the equipment geometry-electrical static mapping rules, lightweight models are bound to asset library parameters to generate an interactive static digital twin that does not require real-time SCADA measurement data streams. This is used for distribution network planning simulation, as-built review, and operation and maintenance simulation, significantly reducing the computing power deployment cost required for dynamic twins.
[0023] The four-state closed-loop scheduling and control subsystem serves as the central scheduling hub for the entire system. It incorporates a finite state machine for the engineering phase and an incremental differential graph synchronization algorithm, and interfaces with the provincial power grid's unified scheduling platform. The implementation logic is as follows: Four-state finite state machine flow: Set the trigger conditions for four-level state jumps: preliminary design review, construction commencement, completion acceptance, and operation and maintenance online; after the digital asset library marks complete completion archive information, it automatically triggers the data push from the construction state to the operation state without the need for manual synchronization.
[0024] Incremental differential fragmented synchronous transmission: only the changed device facets are extracted and parameters are updated, instead of retransmitting the entire model; it is transmitted in parallel by fragment according to device topology, and if a fragment is lost, only the corresponding fragment is retransmitted individually; the synchronization latency is ≤10s, with supporting exponential backoff retransmission fault tolerance logic, and the availability of cross-system service interfaces is ≥99%; Cross-subsystem anomaly linkage: When the incremental synchronization retry reaches the set limit, the anomaly flag is automatically sent back to the AI verification subsystem, and the process of adding digital assets into the database is temporarily suspended; the finite state machine, asset database, and digital twin subsystem share the same device ID index pool, realizing one-click traceability of planning, design, construction, operation and maintenance, and digital twin business data. Standardized data output: The synchronized four-state fusion map is formatted and pushed according to the State Grid "One Map for Power Grid" data interface specification to achieve dynamic and continuous data source updates on the provincial platform.
[0025] The accompanying invention provides a digital and intelligent design and management method for the entire process of distribution network engineering based on multimodal fusion. This method is executed collaboratively by the aforementioned system hardware and software, and consists of five implementation steps, aligning with the implementation cycle of the Pingdingshan substation and rural power grid pilot projects in 2026. S1 Multimodal Exploration Data Standardization Collection Implementation During the initial site survey phase of the project, four-modal raw data were simultaneously collected using a surveying tablet, lidar, and drones. Substitute the BeiDou coordinate transformation matrix into a unified reference and execute. Noise reduction; Substitute into fitness function Iterative optimization until convergence is achieved, generating the optimal power distribution network layout scheme; the RBAC state machine provides hierarchical access to view and edit design results, retains operation traceability logs throughout the process, and outputs standardized exploration data packages to the modeling module.
[0026] S2 Multi-Dimensional Two- and Three-Dimensional Linkage Modeling Implementation Load exploration data packet metadata and construct a hierarchical hash bidirectional mapping. ↔ The AABB two-level conflict detection algorithm is used to traverse all line and equipment models, automatically identify spatial interference, and mark defect points on the two-dimensional and three-dimensional dual interfaces to complete the multi-dimensional modeling of the distribution network project and form a complete two-dimensional and three-dimensional model result that can be linked.
[0027] S3 Multimodal Real-Scene AI Mandatory Verification and Digital Asset Archiving Implementation During the construction and completion phase, drones transmit real-world multimodal data to the cloud via 5G; 3DPointNet+ViT dual-branch network adaptively adjusts feature weights for extraction. Substitute Compare with design parameters; if determined > Significant safety deviations will result in the locking of all model editing interfaces and the issuance of a rectification list. Upon completion of rectification, the system will automatically initiate a second real-world AI review, involving only a second comparison. Only then will it be allowed to collect all process data by device ID and store it in the hybrid storage digital asset library, thus eliminating the loopholes of manual false rectification and lack of real-world verification.
[0028] S4 Scenario-based Lightweight Static Digital Twin Construction and Implementation The system automatically identifies the project as a transformer substation / rural power grid and matches the corresponding noise reduction weights. Filter redundant data; perform differentiated geometric downsampling by main transformer, tower, and cable; bind electrical and terrain parameters of the asset library; generate lightweight static digital twin demonstration results of typical distribution network scenarios without real-time measurement dependence; and support the preparation of project technical reports and operation and maintenance simulation.
[0029] S5 One Map Four States Fully Automated Closed-Loop Synchronous Implementation After the finite state machine detects the completion and archiving mark of the asset library, it automatically starts the incremental differential segmentation synchronization task; if the synchronization fails multiple times, it will suspend the asset entry into the library and retry; after the synchronization is completed, it outputs the standardized fusion model to the provincial power grid one map platform, realizing the fully automatic closed-loop flow of data in the planning state, design state, construction state, and operation state, without the need for manual data export and copying throughout the process.
[0030] This system was piloted at the State Grid Pingdingshan Power Supply Company from January to December 2026. The entire process was verified based on the four research topics of the project, and the pilot covered two typical distribution network projects: urban transformer substations and rural power grids.
[0031] Relying on a complete set of innovative algorithms including BeiDou spatiotemporal registration, bidirectional hash modeling, 3DPointNet+ViT dual-modal AI verification, scenario-specific lightweight twins, and finite state machine four-state synchronization, the system demonstrates outstanding advantages in terms of technology, economy, and management. Compared with traditional distributed power grid design, acceptance, and operation and maintenance tools, the system achieves an accuracy rate of ≥95% for survey data and ≥95% for AI defect identification. It significantly reduces the design and acceptance cycle of single projects, automatically avoids spatial interference and major construction safety deviations, reduces model computing power consumption by 70%, and achieves ten-second-level map and model updates through incremental differential synchronization. It completely eliminates the lag problem of manual data copying through an automatic transfer mechanism. At the same time, it leaves traces of the entire operation and unifies data standards, eradicating long-standing industry pain points such as data silos, asynchronous drawings and models, missed inspections during acceptance, high costs of digital twin implementation, and lagging data sources for the "one map of the power grid".
[0032] From the perspective of practical value, the pilot application significantly saves manpower hours in surveying and mapping and on-site acceptance, greatly reduces engineering rework losses, and standardized digital assets can be reused across projects; the system has built-in mandatory quality control logic, which restricts major construction hazards to closed-loop rectification from the procedural level, and is equipped with complete specifications and intellectual property output paths. The entire multimodal integrated full-process architecture has the conditions for batch replication and promotion of power distribution network projects throughout the province, and can provide stable and standardized core technical support for the closed-loop digital management and control of the provincial power distribution network with "one map and four states".
[0033] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.
[0034] Finally, it should be noted that the above descriptions are merely preferred embodiments of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.
Claims
1. A digital and intelligent design management system for the entire process of power distribution network engineering based on multimodal fusion, characterized in that: include: Multimodal exploration data collaborative acquisition subsystem, multi-dimensional two-dimensional and three-dimensional linkage modeling subsystem, real-scene AI intelligent verification and digital asset library subsystem, typical scene static digital twin mapping subsystem, one map four-state closed-loop scheduling and control subsystem; The multimodal exploration data collaborative acquisition subsystem incorporates a BeiDou spatiotemporal registration transformation matrix, a dynamic sliding filter noise reduction algorithm, a multi-objective genetic optimization fitness function, and an RBAC hierarchical permission transfer state machine. It is used to unify coordinate benchmarks, reduce noise and clean the four-modal data (laser point cloud, aerial image, GIS coordinates, and equipment ledger), automatically generate the optimal power distribution network design scheme, and transfer it online in a hierarchical manner. The multi-dimensional two-dimensional and three-dimensional linkage modeling subsystem has a built-in hierarchical hash parameter mapping index table and an AABB bounding box two-level spatial conflict detection algorithm to establish a two-way real-time mapping relationship between two-dimensional primitive parameters and three-dimensional mesh, and automatically identify engineering spatial interference defects. The real-scene AI intelligent verification and digital asset library subsystem has a built-in dual-branch fusion recognition network and a map-real deviation quantification judgment function. It uses the device's unique ID to connect the multi-modal data of the entire process to build a hybrid storage digital asset library. The typical scenario static digital twin mapping subsystem has a built-in scenario adaptive noise reduction weight allocation algorithm and equipment geometry lightweight downsampling rules. Based on the differentiated data processing logic of transformer area and rural power grid, it generates a lightweight static digital twin without real-time measurement dependence. The one-map four-state closed-loop scheduling and control subsystem incorporates a finite state machine for the engineering phase and an incremental differential graph model synchronization algorithm, automatically driving the four-state data flow of planning / design / construction / operation, and outputting standardized graph models to the provincial power grid one-map platform.
2. The intelligent digital design and management system for the entire process of power distribution network engineering based on multimodal fusion as described in claim 1, characterized in that, The RBAC hierarchical permission flow state machine adopts multi-level dynamic authorization control logic. When a new project participant role is added, the corresponding data read / write, model modification, and asset export permissions are automatically matched. Moreover, each data modification and model export operation generates an immutable traceability log with timestamps, operator information, and operation content.
3. The intelligent digital design and management system for the entire process of power distribution network engineering based on multimodal fusion as described in claim 1, characterized in that, The dual-branch fusion recognition network is a 3DPointNet+ViT dual-branch fusion recognition network, with an image-real deviation quantization judgment function. , in, The absolute deviation value represents the absolute difference between the measured characteristic parameters of the actual construction and the theoretical design parameters. It is used to quantify the degree of deviation between the actual construction scene and the design drawings. For real-world measurement features, physical indicators are extracted from drone point clouds and aerial images by a 3DPointNet+ViT dual-branch fusion AI network, including key geometric parameters of the power distribution network such as tower tilt angle, conductor sag height, tower spacing, and equipment installation offset. Standard design parameters, stored in a two- or three-dimensional linkage model, are used to design standard theoretical quantities and to preset compliant theoretical values for distribution network engineering drawings. These values are derived from distribution network design specifications and project design schemes.
4. The intelligent digital design and management system for the entire process of power distribution network engineering based on multimodal fusion as described in claim 3, characterized in that, Set two levels of deviation thresholds , ,when > The system automatically locks the modeling subsystem's editing interface, builds a hybrid storage asset library with the device's unique ID as the primary key, stores structured parameters in a relational database, and stores 3D models / real-scene images in object storage. A three-level caching mechanism ensures that the asset archiving response time is ≤3 seconds.
5. The intelligent digital design and management system for the entire process of power distribution network engineering based on multimodal fusion as described in claim 1, characterized in that, The scene-adaptive noise reduction weight allocation algorithm has built-in scene recognition trigger logic. It automatically distinguishes between two types of scenes, namely transformer substations and rural power grids, and loads the corresponding noise reduction weight coefficients based on the engineering ledger. The equipment geometry lightweight downsampling rule distinguishes the different surface compression ratios of four types of equipment: main transformers, poles, branch boxes, and cables. The main transformer retains complete and detailed geometry, while the cable surface is greatly simplified, achieving a dynamic balance between twin rendering efficiency and equipment visualization accuracy.
6. The intelligent digital design and management system for the entire process of power distribution network engineering based on multimodal fusion as described in claim 1, characterized in that, The finite state machine in the engineering phase has built-in cross-subsystem linkage triggering logic. When the incremental differential graph model synchronization task fails and reaches the retry limit, it automatically sends an exception flag back to the real-scene AI verification subsystem and suspends the process of adding new assets into the database. The finite state machine, digital asset database, and digital twin subsystem share the same globally unique ID index pool for the same device, enabling one-click traceability and association of planning, design, construction, operation and maintenance, and digital twin business data.
7. A method for intelligent digital design and management of the entire process of power distribution network engineering based on multimodal fusion, characterized in that: Performed by the system of claim 1, the steps include: S1. Standardized Acquisition of Multimodal Exploration Data: Substitute the BeiDou spatiotemporal registration transformation matrix to unify the four-modal data coordinate system, and execute... Filtering out outlier sampling points based on fitness function After iteration to convergence, the optimal power distribution scheme is output. The RBAC state machine manages the cross-departmental flow of design results according to role-based hierarchical control and retains operation logs. S2, Multi-dimensional 2D and 3D linkage modeling: Based on the hierarchical hash index, a bidirectional real-time mapping is established. The AABB two-level conflict detection algorithm is used to automatically identify and grade various spatial interference defects, and the high-brightness positions of defects in the 2D and 3D interface are updated synchronously. S3. Multimodal Real-Scene AI Mandatory Verification and Digital Asset Archiving: Drone 5G Backhaul Real-Scene Multimodal Data Input Dual-Branch Fusion Network Adaptively Adjusts Feature Weights to Extract Engineering Indicators, Substituting Them into... Quantitative comparison; if a major safety deviation is determined, all model editing interfaces are locked and a rectification list is pushed. After the rectification is verified, the full-link data is collected by device ID and stored in the hybrid storage digital asset library. S4. Lightweight Static Digital Twin Construction for Different Scenarios: The system automatically identifies the project type and loads the corresponding noise reduction weights. or Filter redundant data, perform differential geometric downsampling processing according to equipment classification, and generate an interactive static digital twin of the substation / rural network that does not depend on real-time measurement. S5. Fully Automatic Closed-Loop Synchronization of One Map and Four States: After the finite state machine detects the complete completion and archiving mark, it automatically initiates the incremental differential map module synchronization task. If the synchronization fails, the linked asset library will suspend the entry and retry. After the standard is met, the standardized fusion map module will be pushed to the provincial power grid one map platform to realize closed-loop management and control of distribution network data without human intervention throughout the entire life cycle.
8. The method for intelligent digital design and management of the entire process of power distribution network engineering based on multimodal fusion as described in claim 7, characterized in that, In step S3, a closed-loop verification mandatory process is set up: after the rectification of major safety deviations is completed, the system automatically calls the 3DPointNet+ViT dual-branch network to re-identify the real-world data, and only the second comparison difference is lower than the threshold value. Asset archiving operations should only be allowed at certain times to prevent quality loopholes such as manual filling of rectification records without on-site verification.
