An underground space planning management method based on multi-source data fusion

Abstract

The application relates to the technical field of underground space planning management, and discloses an underground space planning management method based on multi-source data fusion, which comprises the following steps: a data access end synchronously analyzes CAD drawings, GIS geographic data, BIM three-dimensional models and real-time geological sensing data, extracts core attributes to generate unified semantic representation, constructs cross-format attribute mapping relationship and transmits to a central database, records operation tracks to form a standardized data set; when data conflicts occur, an adaptive strategy is dynamically generated, and is executed by a local preprocessing module of an edge node in a differentiated manner; after the conflicts are solved, the edge-center cooperatively configures a cache data set, calls a reinforcement learning algorithm to reconstruct a three-dimensional visual digital baseboard and performs incremental updating; an AI engine analyzes planning specifications to complete compliance detection, collects user feedback to update attribute mapping rules and incremental updating strategies. The application can realize accurate collaborative fusion and efficient dynamic updating of multi-source heterogeneous data to adapt to the whole-cycle planning management requirements of underground space.

Description

Technical Field

[0001] This application relates to the technical field of underground space planning and management, and in particular to a method for underground space planning and management based on multi-source data fusion. Background Technology

[0002] With the acceleration of urbanization, underground space, as an important resource for urban development, is facing increasingly sophisticated demands for planning and management. Current underground space planning and management requires the integration of multi-source information, including computer-aided design drawings, geographic information system (GIS) data, building information models (BIMs), and real-time geological sensor data. These data exhibit heterogeneity in format, coordinate reference, and attribute definitions, making data integration challenging. Existing technologies largely rely on basic format conversion for initial fusion, which is insufficient to meet the precision requirements of planning and management.

[0003] Existing multi-source data fusion methods lack deep adaptation to the attribute associations and geometric constraints of different data sources. Conflicts are prone to occur during data synchronization, and the conflict resolution strategies are not targeted enough. It is difficult to balance data consistency and efficiency during dynamic updates, and thus cannot provide reliable data support for the full-cycle planning review and dynamic management of underground space.

[0004] As can be seen from the above, how to achieve accurate collaborative integration and efficient dynamic updating of multi-source heterogeneous data to adapt to the needs of full-cycle planning and management of underground space still needs to be solved. Summary of the Invention

[0005] In order to achieve accurate collaborative fusion and efficient dynamic updating of multi-source heterogeneous data to meet the needs of underground space planning and management throughout its entire life cycle, this application provides an underground space planning and management method based on multi-source data fusion.

[0006] Firstly, this application provides a method for underground space planning and management based on multi-source data fusion, employing the following technical solution: A method for planning and managing underground space based on multi-source data fusion, comprising: At the data access end, data from CAD drawings, GIS geographic data, BIM 3D models, and real-time geological sensing data are synchronously parsed to extract the core attributes corresponding to underground space components and generate a unified semantic representation. The core attributes include burial depth, elevation, geological category, and component type. The semantic representation is based on graph neural networks to construct cross-format attribute mapping relationships and is transmitted to the central database through a distributed consistency synchronization mechanism. In this process, blockchain nodes are embedded to record data operation trajectories, forming a traceable standardized dataset. Based on the standardized dataset and underground space geometric constraint rules, when multi-source data conflict, a corresponding adaptation strategy is dynamically generated. The adaptation strategy is executed by the local preprocessing module of the edge node. The local preprocessing module of the edge node performs a fast response correction process for sudden geological data and a batch fusion verification process for periodic construction data. After the conflict is resolved, in the edge-center collaborative architecture, the edge node cache dataset is dynamically configured in combination with geological conditions and construction progress. The reinforcement learning algorithm is called to reconstruct the corresponding three-dimensional visualization digital base plate based on the cached data and standardized dataset. Incremental update operation is performed synchronously, updating only the data of the changed components. Based on the aforementioned 3D visualized digital base, the AI ​​engine analyzes the planning specifications to generate structured review rules, completing compliance checks for pipeline conflicts and safety clearances; user feedback data is collected during the detection and operation process, and the attribute mapping rules and incremental update strategies corresponding to the graph neural network are updated based on the user feedback data.

[0007] Optionally, the method further includes: When constructing cross-format attribute mapping relationships in graph neural networks, mapping layers are divided according to the type of underground space components, and each mapping layer corresponds to an independent feature extraction link. The feature extraction link is bound to the core attributes of the corresponding underground space components, and the mapping layer results are associated with the data operation trajectory recorded by the blockchain node and stored, and synchronously transmitted to the central database along with the standardized dataset.

[0008] Optionally, before dynamically generating the adaptation strategy, the method further includes: Based on the mapping layering results, corresponding conflict determination thresholds are configured for different component types. When the local preprocessing module of the edge node executes the adaptation strategy, it completes the correction and verification based on the conflict determination thresholds of the corresponding component types. The correction results and verification results are synchronously written to the blockchain node and supplemented to the data operation trajectory.

[0009] Optionally, format standardization preprocessing is performed on the parsed CAD drawings, GIS geographic data, and BIM 3D model respectively. Layer component information is extracted from the CAD drawings and converted into vector data format. Spatial reference system identifiers are added to the GIS geographic data and coordinate deviations are corrected. The BIM 3D model is decomposed into component-level geometric units and structural attributes are extracted. Perform unified coordinate registration on CAD drawings, GIS geographic data and BIM 3D models, and use feature point matching algorithms to locate corresponding components in CAD drawings and BIM models, and establish spatial location association between GIS geographic data and BIM models. The coordinate-registered CAD drawings, GIS geographic data, and BIM 3D models are input into the graph neural network to supplement the layer affiliation, spatial zoning, and structural level attributes of components, optimize the construction accuracy of cross-format attribute mapping relationships, and simultaneously write the format conversion parameters and coordinate registration results into the blockchain node.

[0010] Optionally, the method further includes: Based on the coordinate registration results, the geometric parameter consistency of the vector data of the CAD drawings and the geometric units of the BIM model is checked. The check includes component dimensions, burial depth error, and elevation deviation, and a unified check tolerance range is set. If there is a deviation in the verification, the geometric correction algorithm is called to synchronously adjust the vector data of the CAD drawing and the geometric units of the BIM model. Among them, the spatial location information of the corresponding components is updated synchronously in the GIS geographic data. After the correction is completed, the geometric verification results and correction parameters are added to the standardized dataset, the attribute mapping weights of the graph neural network are updated synchronously, and the correction process log is written to the blockchain node.

[0011] Optionally, the method further includes: Construct an attribute source dictionary for three types of data: computer-aided design drawings, geographic data from geographic information systems, and building information models. Clarify the correspondence between synonymous attributes in different data sources. The correspondence between synonymous attributes includes the mapping rules between layer names of computer-aided design drawings and component categories of building information models, and between land use codes of geographic information systems and geological category attributes. Based on the attribute homology dictionary, the component attributes in the standardized dataset are deduplicated and merged, core attribute fields are retained and homology attribute notes are added to form a unified attribute system. Add timestamps to the attributes of each type of component to record the time nodes of data generation, modification and synchronization, build the data evolution trajectory based on the time series, and associate it with the operation trajectory of blockchain nodes; The attribute homology dictionary, deduplication and fusion results, and time series trajectories are synchronously added to the standardized dataset; when performing incremental update operations, the three types of data that have changed within the specified time interval are filtered based on the timestamp identifier of the attribute data, and the changed components are associated according to the attribute homology relationship to form a changed data set; The topology relationship reconstruction algorithm is called to update the component topology association of the 3D visualized digital base plate, and to verify the spatial accessibility and geometric constraint compatibility between the changed component and the surrounding components. The topology reconstruction results are fed back to the AI ​​engine to optimize the structured review rule judgment logic, and the attribute mapping rules of the graph neural network are updated synchronously. The topology reconstruction parameters and the optimized review rules are written into the blockchain node to supplement the data evolution trajectory.

[0012] Optionally, the method further includes: After the unified semantic representation is generated, a multi-dimensional dynamic prediction model for geological parameters is constructed based on the time series features of historical geological datasets and real-time geological sensing data through a long short-term memory network, and the prediction results of confidence intervals for short-term key geological parameters are output. The confidence interval prediction results are cross-dimensionally correlated and mapped with the burial depth, elevation and geological category attributes of underground space components in the standardized dataset. The dynamic risk score of each underground space component is calculated, and risk warning indicators are generated. The risk warning indicators include geological stability risk level, potential conflict probability and predicted impact range. The geological stability risk level includes low risk, medium risk and high risk. Before dynamically generating the adaptation strategy, the risk warning indicators are used as the core input parameters. The conflict judgment threshold is dynamically corrected by the adaptive threshold adjustment algorithm based on the threshold function optimized by the genetic algorithm. The conflict judgment threshold of high-risk components is appropriately lowered, and the priority queue scheduling mechanism of the edge nodes is triggered to prioritize the execution of the fast response correction process for high-risk components. At the same time, the execution status of the correction process is fed back to the central database in real time. The generation process of risk warning indicators, dynamic threshold adjustment parameters, and priority queue scheduling results are synchronously recorded to blockchain nodes and stored in association with data operation trajectories.

[0013] Secondly, this application provides an underground space planning and management system based on multi-source data fusion, which adopts the following technical solution: An underground space planning and management system based on multi-source data fusion includes: The multi-source data semantic fusion module synchronously parses data from CAD drawings, GIS geographic data, BIM 3D models, and real-time geological sensing data at the data access end. It extracts the core attributes corresponding to underground space components and generates a unified semantic representation. The core attributes include burial depth, elevation, geological category, and component type. The semantic representation is based on a graph neural network to construct cross-format attribute mapping relationships and is transmitted to the central database through a distributed consistency synchronization mechanism. The synchronization process embeds blockchain nodes to record data operation trajectories, forming a traceable standardized dataset. The dynamic conflict adaptation execution module, based on the standardized dataset and underground space geometric constraint rules, dynamically generates corresponding adaptation strategies when multi-source data conflict. The adaptation strategies are executed by the edge node local preprocessing module. The edge node local preprocessing module performs a fast response correction process for sudden geological data and a batch fusion verification process for periodic construction data. After conflict resolution, the 3D digital base plate reconstruction module dynamically configures the cached dataset of edge nodes in the edge-center collaborative architecture, taking into account geological conditions and construction progress. It then calls reinforcement learning algorithms to reconstruct the corresponding 3D visualized digital base plate based on the cached data and standardized dataset, and synchronously performs incremental update operations, updating only the data of changed components. The intelligent compliance review optimization module, based on the aforementioned 3D visualized digital base, uses an AI engine to analyze planning specifications and generate structured review rules to complete compliance checks for pipeline conflicts and safety clearances; it also collects user feedback data during the detection and operation process and updates the attribute mapping rules and incremental update strategies corresponding to the graph neural network based on the user feedback data.

[0014] Thirdly, this application provides an electronic device that adopts the following technical solution: An electronic device includes a processor in which a program for the underground space planning and management method based on multi-source data fusion, as described in any one of the preceding claims, is running.

[0015] Fourthly, this application provides a storage medium, which adopts the following technical solution: A storage medium storing a program for the underground space planning and management method based on multi-source data fusion as described in any one of the above.

[0016] In summary, this application includes at least one of the following beneficial technical effects: In practical applications, targeted preprocessing and unified coordinate registration of CAD drawings, GIS geographic data, and BIM 3D models, combined with the establishment of synonymous attribute correspondences using an attribute homology dictionary, and the construction of cross-format attribute mapping relationships based on graph neural networks, effectively solve the inconsistencies in format, coordinates, and attribute definitions of multi-source heterogeneous data. Simultaneously, by optimizing mapping accuracy through hierarchical feature extraction and geometric parameter consistency verification, and by using blockchain nodes to record the entire process, end-to-end traceability of data from parsing and fusion to storage is achieved. This ensures accurate collaboration of multi-source data at the semantic, geometric, and attribute levels, providing a unified and reliable data foundation for subsequent planning and management.

[0017] By leveraging an edge-center collaborative architecture and combining timestamp identifiers with attribute homology to filter changed data, incremental update operations are performed only on changed components, significantly reducing the redundant overhead of data synchronization and improving the efficiency of dynamic updates. Simultaneously, the spatial correlation of components is verified through topological relationship reconstruction, and conflict judgment thresholds and processing priorities are dynamically adjusted in conjunction with risk warning indicators, forming a closed-loop mechanism of fusion-update-verification-optimization. This mechanism can quickly respond to compliance testing needs during planning reviews and adapt to dynamic data changes during construction, comprehensively supporting the refined management of underground space throughout its entire lifecycle from planning and construction to operation. Attached Figure Description

[0018] Figure 1 This is a flowchart illustrating an underground space planning and management method based on multi-source data fusion, according to an exemplary embodiment.

[0019] Figure 2 This is a structural block diagram of an underground space planning and management system based on multi-source data fusion, according to an exemplary embodiment. Detailed Implementation

[0020] The embodiments of this application are described in detail below, and examples of the embodiments are shown in the accompanying drawings.

[0021] In the description of this specification, the references to "certain embodiments," "one embodiment," "some embodiments," "illustrative embodiment," "example," "specific example," or "some examples" refer to specific features, structures, materials, or characteristics described in connection with the described embodiment or example, which are included in at least one embodiment or example of this application. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0022] This application discloses an underground space planning and management method based on multi-source data fusion, referring to... Figure 1 ,include: The S100 synchronously analyzes data from CAD drawings, GIS geographic data, BIM 3D models, and real-time geological sensing data at the data access end, extracts the core attributes corresponding to underground space components, and generates a unified semantic representation. The core attributes include burial depth, elevation, geological category, and component type. The semantic representation is based on a graph neural network to construct cross-format attribute mapping relationships and is transmitted to the central database through a distributed consistency synchronization mechanism. The synchronization process embeds blockchain nodes to record data operation trajectories, forming a traceable standardized dataset.

[0023] The specific execution process of S100 includes the following four steps: Step 1: Establish a multi-source data synchronization access channel and start the parsing engine to perform parallel synchronous parsing of CAD drawings, GIS geographic data, BIM 3D models, and real-time geological sensor data. Different parsing logic is adopted based on the characteristics of different data sources: CAD drawings are analyzed focusing on layer structure, line type annotations, and component vector information, removing redundant drawing aids and retaining core graphic data directly related to underground space components; GIS geographic data is analyzed first for spatial reference system parameters, land use codes, and associated geographic attribute information, correcting coordinate drift issues that may occur during data acquisition; BIM 3D models are broken down to the component level, analyzing the geometric shape, spatial location, and associated attribute parameters of each component to ensure the integrity of component information; real-time geological sensing data is analyzed for monitoring points, acquisition time, and corresponding monitoring values, filtering out abnormal fluctuations to ensure the validity of real-time data.

[0024] Step 2: Based on the parsed basic data, extract the core attributes of underground space components. Extract the component's burial depth, elevation, and component type information from CAD drawings and BIM 3D models. The burial depth is calculated using the component's bottom elevation and the ground surface datum, while the elevation is calibrated using the altitude datum from GIS geographic data. Extract geological category attributes from real-time geological sensing data and GIS geographic data, and classify and organize the data according to geological classification standards to ensure consistent geological category division. Simultaneously, perform preliminary verification of the extracted attribute data, removing missing and outlier values, and completing ambiguous attributes through cross-comparison with different data sources to form a preliminary attribute dataset.

[0025] Step 3: Construct a unified semantic representation and establish cross-format attribute mapping relationships based on a graph neural network. First, identify the differences in attribute representations across different data sources, such as the synonymous attribute correspondence between "water supply pipeline" in CAD drawings and "water supply components" in BIM models, and "water supply network" in GIS data, to clarify attribute mapping rules. Then, input the preliminary attribute dataset into the graph neural network. Through the network's feature learning capabilities, mine the inherent relationships between attributes in different data formats, construct a cross-format attribute mapping model, and uniformly convert the attributes of heterogeneous data into standard semantic representations, eliminating data semantic barriers and generating a unified semantic representation set of attributes.

[0026] Step 4: The standardized attribute data and associated graphical information are transmitted to the central database through a distributed consistency synchronization mechanism. During synchronization, multi-node data verification logic is employed to ensure data consistency between each access node and the central database, preventing data loss or tampering during transmission. Simultaneously, blockchain nodes are embedded in the synchronization process to record data operation trajectories in real time, including key information such as data source, parsing parameters, attribute mapping rules, transmission time, and operator. Each operation record generates a unique hash value, linking with previous records to form an immutable traceability chain, ultimately integrating to form a complete and traceable standardized dataset.

[0027] It should be noted that by targeted parsing, attribute extraction, and semantic unification, the format and semantic barriers between CAD, GIS, BIM, and geological sensing data have been broken down, allowing the originally scattered and heterogeneous underground space data to form a unified standard and traceable dataset.

[0028] S200, based on standardized datasets and underground space geometric constraints, dynamically generates corresponding adaptation strategies when multi-source data conflict. The adaptation strategies are executed by the local preprocessing module of the edge nodes. Specifically, the local preprocessing module of the edge nodes performs a fast response correction process for sudden geological data and a batch fusion verification process for periodic construction data.

[0029] The S200 execution process specifically includes the following 5 steps: Step 1: Load the standardized dataset and underground space geometric constraint rules to build a conflict detection framework. First, import the standardized dataset generated by S100 into the conflict detection module. At the same time, call the preset underground space geometric constraint rule library. The rule library covers the core constraints of industry standards and project requirements, including the minimum clearance between different types of components, burial depth limits, allowable elevation deviation range, and spatial location avoidance requirements. Then, establish the association mapping between the dataset and the constraint rules, clarify the geometric constraint standards corresponding to each type of underground space component, provide a unified basis for subsequent conflict judgment, and ensure the standardization of the detection logic.

[0030] Step 2 involves a comprehensive conflict scan and type determination of the standardized dataset. A component-by-component, cross-data source comparison logic is employed to focus on detecting two core conflict types: first, geometric conflicts, such as inconsistencies in the burial depth and elevation values ​​of the same component across different data sources, or overlapping positions of pipelines and structures in three-dimensional space with spacing less than the constraint threshold; second, attribute conflicts, such as discrepancies between the geological categories of the same area labeled in GIS data and real-time geological sensing data, or deviations in component type definitions between CAD drawings and BIM models.

[0031] At the same time, the characteristics of conflicting data are distinguished, and sudden geological data (such as temporarily monitored stratum subsidence and geological anomaly change data) and periodic construction data (such as BIM model update data and CAD drawing revisions submitted regularly according to the construction progress) are marked, laying the foundation for differentiated processing.

[0032] Step 3: Based on conflict type, data characteristics, and geometric constraint rules, dynamically generate targeted adaptation strategies. For sudden geological data conflicts, combine the importance of real-time monitoring points and the impact range of geological anomalies to generate rapid correction strategies, clarifying correction priorities, numerical adjustment ranges, and verification standards to ensure that the strategies can quickly respond to geological changes and avoid affecting subsequent planning judgments. For periodic construction data conflicts, combine the scale of batch data and component correlation to generate batch fusion verification strategies, formulate batch verification processes, data fusion rules, and conflict resolution priorities (e.g., using the component accuracy of the BIM model as the standard to calibrate the corresponding data in the CAD drawings), and clarify the unified attribute standards of the fused data to ensure the efficiency and accuracy of batch processing.

[0033] Step 4: The edge node local preprocessing module executes the adaptation strategy and implements the differentiated conflict resolution process. For sudden geological data, the edge node local preprocessing module initiates a rapid response correction process. Without waiting for instructions from the central database, it directly calls the locally cached constraint rules and compares them with historical data to correct conflicting data in real time. After correction, a secondary verification is immediately performed to confirm that the data meets the geometric constraint requirements, and the correction parameters and process are recorded. For periodic construction data, the edge node local preprocessing module executes a batch fusion verification process, fusing data in batches according to component type. After each batch of fusion is completed, the consistency between the fused data and the geometric constraint rules is automatically verified, and a batch verification log is generated, marking the unresolved conflict points and their causes.

[0034] Step 5: Synchronize conflict resolution results and process data, and update the standardized dataset. Synchronize the corrected emergency geological data and the batch-fused and verified construction data back to the standardized dataset, overwriting the original conflict data. Simultaneously, write the conflict detection results, adaptation strategy content, execution process logs, and final correction parameters to the blockchain node, supplementing the data operation trajectory to ensure the entire conflict handling process is traceable, providing a basis for subsequent data traceability and problem investigation.

[0035] Based on the execution steps of S200 described above, targeted conflict detection and differentiated adaptation effectively resolve potential geometric and attribute conflicts that may arise after multi-source data fusion. Simultaneously, leveraging the local processing capabilities of edge nodes, it achieves rapid response to sudden data releases and efficient batch processing of periodic data, balancing the timeliness and accuracy of conflict resolution. This step further optimizes the quality of the standardized dataset, ensuring that the data conforms to the geometric constraints of underground space while also adapting to the processing needs of different data types.

[0036] After conflict resolution, S300 dynamically configures the cached dataset of edge nodes in the edge-center collaborative architecture, taking into account geological conditions and construction progress. It then calls reinforcement learning algorithms to reconstruct the corresponding 3D visualized digital base plate based on the cached data and standardized dataset, and synchronously performs incremental update operations, updating only the data of changed components.

[0037] The S300 execution process includes the following five steps: Step 1: After resolving data conflicts, initiate the data scheduling mechanism of the edge-center collaborative architecture. First, synchronously read the standardized dataset after conflict resolution, and at the same time collect the geological condition data of the current underground space and the project construction schedule.

[0038] The geological condition data focuses on sorting out the regional stratigraphic distribution, rock and soil mechanical properties and potential geological risk areas. The construction schedule clearly defines the current construction stage, core construction parts and subsequent progress nodes. Based on these two types of information, data priorities are divided. For example, BIM component data of the current construction area and sensor-related data of geologically complex areas are set as high priority, while historical archived data of non-construction areas are set as low priority, providing a basis for edge node caching configuration.

[0039] Step 2: Dynamically configure the edge node cache dataset. Based on the storage capacity, computing power, and regional service range of the edge nodes, cache high-priority data as needed to ensure that the edge can quickly access core data and reduce frequent interactions with the central database. At the same time, establish a cache update mechanism to regularly adjust the cache content according to the progress of construction, changes in geological conditions, and data access frequency. Remove low-priority data that has not been accessed for a long time and add newly added high-priority data to balance cache efficiency and data timeliness. After the cache configuration is completed, synchronously feed back the cache list to the central database to achieve data synchronization and alignment between the edge and the center.

[0040] Step 3: Call the reinforcement learning algorithm and complete data preprocessing. Link the high-priority data in the edge node cache with the complete standardized dataset in the central database and import them into the reinforcement learning algorithm module, clarifying the algorithm optimization objective: Ensure that the topological relationships of the components in the 3D digital base plate are complete, the geometric accuracy meets the standards, and it can adapt to geological conditions and construction progress requirements; at the same time, set algorithm constraints, including underground space geometric constraint rules, component attribute association requirements and visualization presentation standards, to guide the algorithm to focus on the core optimization direction and avoid ineffective iterations.

[0041] Step 4: Reconstruct the 3D visualization digital baseboard based on reinforcement learning algorithms. The algorithm first iteratively learns and mines the optimal correlation between data, such as optimizing the spatial fit between BIM components and GIS geographic data, adjusting the adaptation logic between geological conditions and component layout, and outputting a data fusion optimization scheme. Then, based on the scheme, a basic framework for the 3D baseboard is constructed, and the component geometric information, core attribute data, and spatial correlations from various data sources are superimposed in sequence to form a preliminary 3D model. Subsequently, the model deviation is repeatedly corrected through the algorithm verification module to ensure that parameters such as component position, burial depth, and elevation match the actual geological conditions and construction progress, and finally, a complete 3D visualization digital baseboard is generated.

[0042] Step 5: Simultaneously execute incremental update operations. First, use the data comparison engine to compare the data differences between the reconstructed 3D base plate and the previous version of the base plate. Combine the timestamp of S100 and the conflict resolution record of S200 to accurately filter out the changed component data, including three categories: newly added components, components with modified attributes, and obsolete / deleted components. Only perform update operations on these changed components without changing the overall data architecture of the base plate. Newly added components are directly embedded into their corresponding spatial locations and associated with attributes. Modified components are updated synchronously with updated geometric parameters and attribute information. Deleted components are marked as obsolete and archived records are retained. During the update process, the topological compatibility and geometric constraint compliance of the changed components with surrounding components are simultaneously verified to ensure the consistency of the updated base plate data. At the same time, the incremental update content, update time, and operation basis are synchronized to the central database and stored in the blockchain node to supplement the operation trajectory.

[0043] Based on the execution steps of S300 described above, dynamic configuration of the cache through edge-center collaboration ensures both rapid local access to core data and global data coordination through the central database, balancing data processing efficiency and integrity. The application of reinforcement learning algorithms ensures that the 3D visualized digital base plate can accurately adapt to geological conditions and construction progress, improving the reliability and practicality of the base plate. The incremental update mode only processes changed components, significantly reducing redundant overhead in data updates and avoiding the inefficiency caused by full updates. Simultaneously, through full-process verification and traceability, it ensures that the base plate data is always synchronized with the actual underground space status.

[0044] The S400, based on a 3D visualized digital base, uses an AI engine to analyze planning specifications and generate structured review rules to complete compliance checks on pipeline conflicts and safety clearances. It also collects user feedback data during the detection and operation process and updates the attribute mapping rules and incremental update strategies corresponding to the graph neural network based on the user feedback data.

[0045] The S400 execution process includes the following five steps: Step 1: Load the 3D visualization digital baseboard, start the AI ​​engine, and import the complete set of underground space planning specifications, including national industry standards, local management regulations, and project-specific technical requirements. The AI ​​engine uses the text parsing module to break down the specifications sentence by sentence, eliminating redundant descriptions and non-mandatory clauses, and accurately extracting core constraint parameters related to pipeline layout and safety clearances. These parameters include minimum horizontal and vertical clearance limits between different types of pipelines (water supply, drainage, gas), avoidance requirements between pipelines and structures, geological risk areas, and boundary conditions for determining violations. Simultaneously, the extracted constraint parameters are converted into computer-recognizable structured review rules, organized and archived in the format of pipeline type – constraint type – parameter value – judgment logic, forming a standardized review rule library to ensure clear basis for subsequent inspections.

[0046] Step 2: Based on structured review rules, conduct a full-dimensional compliance inspection of the 3D visualized digital baseboard. The AI ​​engine calls the 3D spatial analysis module to first perform a full-area scan of the pipeline components in the baseboard, locating the spatial position, direction, and related components of all pipelines; then, it compares the distance of each pipeline with surrounding pipelines, structures, and geological areas to verify whether it meets the safety clearance requirements, and simultaneously detects whether there are conflicts such as pipeline intersections, overlaps, or intrusions into prohibited construction areas. During the inspection process, a "precise positioning + hierarchical marking" logic is adopted, marking the specific violation type, corresponding standard clauses, and deviation values ​​for violation points, using different colors to highlight minor violations, general violations, and serious violations, generating a visualized inspection result list, and synchronously linking it to the specific location in the 3D baseboard for easy and intuitive viewing by users.

[0047] Step 3: Collect user feedback data during the testing and operation process to establish a multi-dimensional feedback collection mechanism. When viewing the test results, users can confirm, correct, or reject marked violations, such as identifying false conflicts misjudged by the AI ​​engine, supplementing special scenario requirements not covered by the specifications, and adjusting the violation level classification standards. Simultaneously, the user's operation trajectory is recorded, including modifications to the test results, new review suggestions, parameter adjustment records, and operation times, which are then categorized and organized into valid feedback data. Invalid feedback (such as operational errors or duplicate feedback) is filtered out, and valid feedback is labeled with feedback type, corresponding component information, and core requirements, forming a structured feedback dataset.

[0048] Step 4: Based on valid feedback data, iteratively update the attribute mapping rules and incremental update strategy of the graph neural network. For user-corrected misjudgment cases, extract the attribute features and spatial relationships of the corresponding components, feed them back into the graph neural network, optimize the accuracy of cross-format attribute mapping, adjust the core parameters of conflict determination, and reduce subsequent similar misjudgments. For special specification requirements supplemented by users, integrate them into the structured review rule base, and synchronously update the attribute association logic of the graph neural network to ensure that the new constraints can adapt to multi-source data fusion scenarios. Regarding the incremental update strategy, combine user feedback on "frequently modified component types" and "key detection areas" to adjust the priority of incremental updates. Set shorter update cycles for core area components and easily non-compliant pipelines that users are concerned about, while optimizing the identification algorithm for changed components to improve update accuracy. After the update is completed, synchronously write the feedback data, update parameters, and effect verification results into the blockchain node to form a closed-loop record.

[0049] Step 5: Verify the update effect and synchronize it to the entire process. The updated graph neural network attribute mapping rules, incremental update strategies, and review rule base are reapplied to the 3D visualization digital baseboard for secondary detection. The differences between the before and after detection results are compared to confirm whether the false positive rate has decreased and whether the violation identification is more accurate. Simultaneously, the updated rules and strategies are synchronized to the edge-center collaborative architecture to ensure that the subsequent data fusion, dynamic updates, and compliance detection processes all use the optimized logic, achieving continuous iteration of the technical solution.

[0050] Based on the S400 execution steps described above, an AI engine transforms abstract planning specifications into implementable structured review rules, efficiently completing compliance checks for pipeline conflicts and safety clearances. This significantly reduces the workload of manual review and improves the accuracy and efficiency of the checks. Simultaneously, a closed-loop optimization mechanism is built based on user feedback, allowing the graph neural network's attribute mapping rules and incremental update strategies to continuously adapt to real-world application scenarios, constantly correcting technical deviations and supplementing specific needs, ensuring the entire methodology keeps pace with project progress and specification updates.

[0051] Based on the solutions implemented from S100 to S400, a case study is used to illustrate this approach. Taking an underground integrated utility tunnel project in a core urban area (including multiple pipelines such as water supply, drainage, electricity, and communications, encompassing the entire lifecycle of planning, construction, and operation) as an example, the precise fusion of multi-source heterogeneous data is achieved through multi-stage collaboration. In the initial stage of the project, the data access point simultaneously analyzes CAD pipeline drawings provided by the design unit, GIS geospatial data from the municipal department, BIM utility tunnel component models from the construction unit, and real-time geological sensor data deployed along the route. Redundant annotations in the CAD drawings are removed to extract core vector information such as pipeline direction and diameter. GIS data corrects coordinate drift and supplements the distribution of surrounding roads and underground structures. The BIM model is disassembled to the component level, such as compartments and supports, and installation elevations are extracted. Geological sensor data filters out outliers and retains settlement and soil stress monitoring results. A semantic mapping is established through a graph neural network, unifying the CAD "water supply main pipe," BIM "water supply compartment pipeline," and GIS "water supply network" into a standard semantic. This is then transmitted synchronously to the central database via distributed consistency, with blockchain data embedded to record the operation trajectory.

[0052] When a discrepancy is detected between the elevation of the utility tunnel support in the BIM model and the GIS topographic data, or when sudden geological sensor data shows local settlement anomalies, the local preprocessing module at the edge node responds quickly: it performs immediate correction on the settlement data and calibrates the burial depth of the utility tunnel components in the corresponding area; for elevation conflicts between BIM and GIS, it performs batch fusion verification according to geometric constraint rules, calibrates the data based on the accuracy of BIM components, and finally forms a unified and traceable standardized dataset.

[0053] The project has entered the main structure construction phase. The edge-center collaborative architecture, considering current geological conditions (local silty clay soil prone to settlement) and construction progress (completion of the eastern section's main structure pouring this month), dynamically configures the cached datasets at edge nodes. The BIM model of the eastern section's utility tunnel, real-time geological sensor data from the surrounding area, and GIS topographic data of the construction area are set as high-priority caches to reduce interaction latency with the central database. When the reinforcement learning algorithm is invoked, the cached data is linked with the full standardized dataset to optimize the spatial fit between the BIM utility tunnel components and the GIS topography. Combined with geological settlement data, the model's support structure parameters are adjusted to reconstruct a 3D visualized digital base that accurately matches the current construction status.

[0054] When the construction unit completes the laying of power compartment pipelines in the eastern section of the utility tunnel and submits the updated BIM model, the system uses a data comparison engine to accurately filter out the data of newly added power pipelines, adjusted support positions, and other changed components. Incremental updates are performed only on these parts without changing the overall base structure. The system also verifies the topological compatibility of the newly added pipelines with the water supply compartment pipelines. The entire update process is written to the blockchain, which avoids the redundant overhead of a full update and ensures that the base data is synchronized with the construction progress in real time.

[0055] During the planning review phase, the AI ​​engine analyzes the corresponding technical specifications for urban integrated utility tunnel projects, generating structured review rules such as pipeline spacing and safety clearance. Based on a 3D baseboard, it automatically detects issues such as insufficient clearance between power and communication compartments and some pipelines encroaching on geological risk areas, highlighting these issues with different colors and associating them with corresponding specification clauses. Reviewers discovered that the AI ​​misjudged the pipeline clearance at corners (due to the irregular structure of the utility tunnel). After feedback, the system extracted the features of this scene to feed back into the graph neural network, optimizing the attribute mapping rules and review logic, and correcting the misjudged parameters.

[0056] As the project enters the operational phase, real-time geological sensing data is continuously integrated, and edge nodes dynamically adjust their caching and update frequencies. Intensive updates are provided for key areas such as utility tunnel entrances and exits and geologically vulnerable sections. The AI ​​engine regularly conducts safety inspections based on the updated 3D base plate. Combined with pipeline maintenance records provided by operations personnel, the priority of incremental updates and conflict judgment criteria are continuously optimized.

[0057] Throughout the process, data goes from initial integration and calibration in the planning stage to dynamic updates during the construction phase, and then to optimization and iteration during the operation phase, forming a complete closed loop that fully supports the refined management needs of the entire project lifecycle.

[0058] In this embodiment of the application, the method further includes: Step 1: After clarifying the differences in attribute descriptions from different data sources and defining preliminary attribute mapping rules, the mapping layers are divided according to the type of underground space components. Based on core attributes and actual planning and management needs, components are divided into three main categories: pipelines (water supply, drainage, power, and communication pipelines), structures (pipe gallery compartments, supports, maintenance wells, and pump rooms), and geological structures (strata and soil). The scope and classification standards of each layer are clearly defined to ensure that similar components are grouped into the same layer, avoiding cross-type confusion.

[0059] Step 2 involves configuring an independent feature extraction link for each mapping layer, with each link specifically bound to the core attributes of the corresponding layer's components. For example, the pipeline layer link focuses on extracting core attributes such as burial depth, pipe diameter, material, and orientation; the structure layer link extracts attributes such as elevation, size, structural type, and installation location; and the geology layer link extracts attributes such as geological category and geotechnical parameters. Each link operates independently without interfering with others, thus specifically enhancing the feature capture capability for the corresponding attributes.

[0060] Step 3: After completing the mapping layering and feature extraction link configuration, the mapping layering results (including layering standards, a list of components for each layer, and corresponding feature extraction link parameters) are associated and stored with the data operation trajectory recorded by the blockchain nodes. The parsing parameters and attribute mapping rules corresponding to each layer are simultaneously labeled to ensure that the configuration process and data source of each layer can be traced through the blockchain, forming traceable, associated data for each layer.

[0061] Step 4: When transmitting standardized data to the central database through the distributed consistency synchronization mechanism, the mapping layering result is transmitted along with the standardized dataset, and the layering identifier, link configuration information and associated traceability data in the central database are updated synchronously, so that the standardized dataset in the central database has layered structured characteristics.

[0062] By dividing the mapping layers according to component type and configuring independent feature extraction links, the accuracy and relevance of cross-format attribute mapping are improved, avoiding cross-interference of attributes of different types of components. At the same time, the mapping layer results are linked and traced with the blockchain operation trajectory, providing a layered data foundation for subsequent conflict detection and dynamic cache configuration according to component type. The structure of the standardized dataset is further optimized, making the whole process data processing more targeted.

[0063] In this embodiment of the application, before dynamically generating the adaptation strategy, the method further includes: After establishing the conflict detection framework and before dynamically generating adaptation strategies, the stored mapping layer results are retrieved from the central database to clarify the specific component types under each mapping layer, such as pipelines and structures, as well as the core attributes of the underground space components bound to each layer. Based on the geometric constraints of underground space, specific conflict judgment thresholds are configured for different component types. For pipeline components, the judgment thresholds are set around burial depth deviation and safety clearance, while for structure components, the judgment thresholds are set around elevation deviation and spatial overlap, ensuring that the threshold standards match the component's own attributes and industry specifications. The local preprocessing module of the edge nodes loads the conflict judgment thresholds corresponding to each component type. When executing the adaptation strategy to handle data conflicts, the type and corresponding mapping layer of the conflicting component are first identified, and then the matching conflict judgment thresholds are retrieved for correction and verification. The data deviation is judged according to the threshold standards to see if it is within the allowable range; if it exceeds the range, the adaptation strategy is used to complete the adjustment. After the correction and verification of the conflicting components are completed, core information such as the conflicting component type, the judgment threshold used, the correction and adjustment parameters, and the verification pass status are recorded. This information is synchronously written to the blockchain node, directly supplementing the corresponding data operation trajectory.

[0064] By combining the mapping and layering results to configure exclusive conflict judgment thresholds for different component types, the judgment and correction of multi-source data conflicts are more in line with the component attribute characteristics, improving the accuracy of conflict resolution. At the same time, the correction and verification results are synchronized into the chain, improving the full-process traceability link of data operation.

[0065] In this embodiment of the application, the method further includes: Step 1: Perform format standardization preprocessing on the CAD drawings, GIS geographic data, and BIM 3D model that have been initially analyzed in S100. For CAD drawings, focus on extracting layer information, component graphics, and annotation data related to underground space components, remove drawing auxiliary elements, and convert the extracted information into vector data format to ensure the editability and compatibility of the graphic data. For GIS geographic data, supplement the preset spatial reference system identifier (clarify the coordinate benchmark), and check and correct coordinate drift and deviation problems that occurred during data collection to ensure the accuracy of geospatial location. For BIM 3D model, decompose it to component-level geometric units, extract the structural attributes (such as material, size, load-bearing level, etc.) of each component, and form a component-level attribute list to lay the foundation for subsequent integration.

[0066] Step 2: After completing the format standardization preprocessing, perform a unified coordinate registration operation on the three types of data: Using a preset spatial reference system as the benchmark, call the feature point matching algorithm to filter out the corresponding feature components (such as the corner of the pipe gallery compartment, the center point of the maintenance well, etc.) in the CAD drawings and BIM 3D models. By comparing and calibrating the feature point coordinates, lock the correspondence between the same component in the two types of data; at the same time, associate the spatial location information in the GIS geographic data with the geometric coordinates of the components in the BIM 3D model to establish a spatial location mapping between the GIS geographic data and the BIM model, ensuring that the three types of data are in the same coordinate system and eliminating spatial location deviations.

[0067] Step 3: Input the coordinate-registered CAD drawings, GIS geographic data, and BIM 3D model into the graph neural network. Based on the original attribute extraction, supplement the layer affiliation of each component (clarifying the CAD layer to which the component belongs), spatial zoning (clarifying the underground space area where the component is located), and structural level attributes to enrich the component attribute dimensions. Utilize the feature learning capability of the graph neural network, combined with the supplemented attribute information, to optimize the construction accuracy of cross-format attribute mapping relationships and reduce the deviation of attribute mapping from different data sources. At the same time, the parameters in the format conversion process (such as vector format conversion rules) and the results of coordinate registration (such as feature point matching records and coordinate calibration parameters) are synchronously written into the blockchain node and stored in association with the data operation trajectory to ensure the traceability of this step.

[0068] By standardizing the format of three types of core data and performing unified coordinate registration, the format barriers and spatial location deviations of different data sources were eliminated. Combined with graph neural networks to supplement component attributes and optimize mapping accuracy, the accuracy of multi-source data fusion was further improved.

[0069] In this embodiment of the application, the method further includes: First, load the coordinate registration results completed earlier to clarify the correspondence between the vector data of the CAD drawings and the geometric units of the BIM model. In conjunction with the industry standards for underground space planning and management and the specific requirements of the project, set a unified range of verification tolerances (such as the allowable range of component size tolerances, burial depth and elevation deviations) to ensure that the verification standards meet the actual application requirements and provide a unified basis for subsequent consistency judgments.

[0070] Secondly, based on the set verification tolerance range, the geometric parameter consistency verification is carried out on the vector data of CAD drawings and the geometric units of BIM model. The focus is on comparing three core parameters of the same corresponding component in the two types of data: component size (such as pipe diameter, cabin length, width and height), burial depth error (the difference between the burial depth marked in CAD and the burial depth calculated in BIM model), and elevation deviation (the numerical difference between the top / bottom elevation of the component). The comparison and inspection are completed for each component, and the verification results (qualified or with deviation) and the specific value of the deviation are recorded.

[0071] Then, if the verification finds deviations exceeding the tolerance range, the preset geometric correction algorithm is immediately invoked to synchronously adjust the vector data of the CAD drawing and the geometric units of the BIM model: for dimensional deviations, the dimensional parameters of the corresponding components in the two types of data are corrected according to the design reference values; for burial depth and elevation deviations, the corresponding values ​​in CAD and BIM are synchronously calibrated in conjunction with the coordinate registration reference to ensure that the two types of data remain consistent after correction; at the same time, the spatial location information of the corresponding components in the GIS geographic data is updated synchronously to ensure that the spatial correlation of the three types of data is not disconnected.

[0072] Finally, after the correction is completed, the complete results of this geometric verification (including the list of verified components, pass / fail status, and deviation values) and the relevant parameters of the geometric correction algorithm (such as correction coefficients and adjustment logic) are added to the standardized dataset to improve the completeness of the dataset. The attribute mapping weights of the graph neural network are updated simultaneously, and the geometrically corrected parameters are integrated into the mapping model to optimize the accuracy of cross-format attribute mapping. Finally, the log of the entire verification and correction process (including operation time, operators, and correction details) is written to the blockchain node to supplement the data operation trajectory and ensure that this step is traceable.

[0073] By verifying and synchronously correcting the geometric parameters of CAD and BIM data, the geometric accuracy of the two types of core data is ensured to be consistent, avoiding errors in subsequent data fusion and conflict detection due to parameter deviations. At the same time, the mapping weights and blockchain records are updated synchronously, further improving the reliability and traceability of multi-source data fusion.

[0074] In this embodiment of the application, the method further includes: Step 1 involves constructing an attribute source dictionary specifically for three core data types: computer-aided design drawings, geographic information system (GIS) data, and building information model (BIM) data. This involves comprehensively analyzing the differences in the descriptions of component attributes across these three data sources, clarifying the correspondences of synonymous attributes. For example, the "water supply pipeline layer" in computer-aided design drawings corresponds to the "water supply component category" in BIM, and the "silty clay code" in GIS corresponds to the "silty clay stratum" in geological category attributes. All the rules governing the correspondence of synonymous attributes are compiled into a standardized attribute source dictionary, providing a unified basis for subsequent attribute fusion.

[0075] Step 2: Based on the constructed attribute homology dictionary, deduplication and fusion processing is performed on the component attributes in the existing standardized dataset. Each component's attribute information is compared one by one, duplicate attribute fields are removed, and core attributes such as burial depth, elevation, and geological category are retained. Simultaneously, homology attribute notes are added to each attribute to clarify synonymous expressions of the attribute in different data sources. Ultimately, a unified and non-redundant component attribute system is formed, eliminating the problem of inconsistent attribute representations across different data sources.

[0076] Step 3: Add a unique timestamp identifier to each type of component attribute, record in detail the key nodes such as the generation time, modification time, and synchronization time of each attribute data, and then construct the data evolution trajectory of each component attribute based on these time nodes. Link this trajectory with the existing data operation trajectory in the blockchain node to achieve traceability of the entire life cycle of the component attribute.

[0077] Step 4 involves simultaneously supplementing the constructed attribute homology dictionary, the complete result of attribute deduplication and fusion, and the time series trajectory of each component into the standardized dataset to further improve the attribute completeness and standardization of the dataset. When performing incremental update operations, based on the timestamp identifier of the component attributes, the three types of data that have been modified, added, or deleted within the specified time interval are accurately selected: computer-aided design drawings, geographic information system geographic data, and building information model data. Then, according to the corresponding rules in the attribute homology dictionary, the changed components in each type of data source are associated to form a centralized set of changed data, which facilitates efficient updates in the future.

[0078] Step 5: Invoke the topology relationship reconstruction algorithm. Based on the selected set of changed data, update the topology relationship of the components in the 3D visualized digital base plate. At the same time, focus on verifying the spatial accessibility (such as the connectivity between pipelines and maintenance wells) and geometric constraint compatibility (such as the matching degree of component spacing and burial depth) between the changed components and surrounding components to ensure that the topology relationship of the base plate is complete and accurate after the update.

[0079] Step 6: The results of the topology relationship reconstruction are fed back to the artificial intelligence engine, which optimizes the judgment logic of the existing structured review rules to make the review rules more in line with the actual situation of the component topology relationship. At the same time, the cross-format attribute mapping rules of the graph neural network are updated to integrate the topology relationship information into the mapping model and improve the accuracy of attribute mapping. Finally, all the relevant parameters of the topology relationship reconstruction and the optimized content of the artificial intelligence engine review rules are written into the blockchain node to supplement the corresponding component data evolution trajectory and improve the traceability link of the entire update and optimization process.

[0080] By constructing an attribute homology dictionary, attribute deduplication and fusion, and time series trajectory tracing, a unified standard for three types of core data attributes was achieved. At the same time, relying on topological relationship reconstruction and multi-module linkage optimization, the accuracy of incremental updates and the topological integrity of the three-dimensional base plate were improved, and the closed-loop mechanism of data fusion, updating and optimization was improved, further adapting to the needs of the accuracy of component attributes and topological correlation in the full-cycle planning and management of underground space.

[0081] In this embodiment of the application, the method further includes: Step 1: After generating a unified semantic representation, integrate the historical geological datasets accumulated in the project with the monitoring data collected by real-time geological sensors. Extract time series features from both types of data, focusing on extracting the changing patterns of core geological parameters such as stratum subsidence, soil stress, and groundwater level over time. Based on the extracted features, build a long short-term memory network. Through the network's temporal learning capability, construct a multi-dimensional dynamic prediction model for geological parameters. After model training and validation, output the confidence interval prediction results of short-term key geological parameters, clarifying the range of change and probability distribution of each geological parameter.

[0082] Step 2 involves cross-dimensional correlation mapping between the geological parameter confidence interval prediction results and the burial depth, elevation, and geological category attributes of each underground space component in the standardized dataset, establishing the correlation between geological parameter changes and component spatial attributes. Following a pre-set risk assessment algorithm, the degree of influence of geological changes on the components is quantitatively calculated based on the geological parameter prediction results, resulting in a dynamic risk score for each underground space component. Risk warning indicators are then generated based on these scores, clarifying the low, medium, and high risk levels of geological stability for each component, as well as the potential conflict probability of data conflicts arising from geological changes and the predicted range of geological risk impact on the component and surrounding area.

[0083] Step 3: Before dynamically generating the multi-source data conflict adaptation strategy, the risk warning indicators are imported into the threshold adjustment module as core input parameters. Through the adaptive threshold adjustment algorithm, combined with the threshold function optimized by the genetic algorithm, the original conflict judgment thresholds of different component types are dynamically corrected. For components with high geological stability risk levels, their conflict judgment thresholds are appropriately lowered to improve the sensitivity of conflict detection. At the same time, the priority queue scheduling mechanism of edge nodes is triggered to include the conflict handling tasks of high-risk components in the priority execution sequence. A fast response correction process is executed for these components. During the correction process, the execution nodes and processing status of the process are fed back to the central database in real time, realizing the real-time monitoring of the high-risk component processing process by the center.

[0084] Step 4: The entire process of generating risk warning indicators, various parameters for dynamic threshold adjustment, results of priority queue scheduling, and execution status are all synchronously recorded to the blockchain node and stored in association with the original data operation trajectory to ensure that the data of the entire process of geological parameter prediction, risk assessment, threshold adjustment, and high-risk component handling is traceable and tamper-proof.

[0085] By constructing a dynamic prediction model for geological parameters using a long short-term memory network and generating risk warning indicators for components, the system enables early prediction of the impact of geological changes on underground space components. Simultaneously, based on the risk warning indicators, the system dynamically adjusts the conflict judgment threshold and prioritizes the handling of high-risk components, making data conflict handling more aligned with actual geological risk conditions and improving the pertinence and rationality of conflict resolution. Furthermore, by using blockchain, the system enables traceability of the entire process of geological risk linkage handling, further strengthening the ability of multi-source data fusion and underground space planning and management to adapt to dynamic geological changes.

[0086] This application discloses an underground space planning and management system based on multi-source data fusion, referring to... Figure 2 ,include: The multi-source data semantic fusion module 001 synchronously parses data from CAD drawings, GIS geographic data, BIM 3D models, and real-time geological sensing data at the data access end. It extracts the core attributes corresponding to underground space components and generates a unified semantic representation. The core attributes include burial depth, elevation, geological category, and component type. The semantic representation is based on a graph neural network to construct cross-format attribute mapping relationships and is transmitted to the central database through a distributed consistency synchronization mechanism. The synchronization process embeds blockchain nodes to record data operation trajectories, forming a traceable standardized dataset. The dynamic conflict adaptation execution module 002, based on the standardized dataset and underground space geometric constraint rules, dynamically generates corresponding adaptation strategies when multi-source data conflict. The adaptation strategies are executed by the edge node local preprocessing module. Specifically, the edge node local preprocessing module performs a fast response correction process for sudden geological data and a batch fusion verification process for periodic construction data. After conflict resolution, the 3D digital base plate reconstruction module 003 dynamically configures the edge node cache dataset in the edge-center collaborative architecture, combining geological conditions and construction progress. It calls the reinforcement learning algorithm to reconstruct the corresponding 3D visualized digital base plate based on the cached data and standardized dataset, and synchronously performs incremental update operations, updating only the data of changed components. The Intelligent Compliance Review Optimization Module 004, based on a 3D visualized digital base, uses an AI engine to analyze planning specifications and generate structured review rules to complete compliance checks for pipeline conflicts and safety clearances. It also collects user feedback data during the detection and operation process and updates the attribute mapping rules and incremental update strategies corresponding to the graph neural network based on the user feedback data.

[0087] This application also discloses an electronic device, including a processor, wherein the processor runs a program of the underground space planning and management method based on multi-source data fusion as described in any one of the above-mentioned embodiments.

[0088] This application also discloses a storage medium storing a program for the underground space planning and management method based on multi-source data fusion as described in any one of the above embodiments.

[0089] Although embodiments of this application have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting this application. Those skilled in the art can make changes, modifications, substitutions and variations to the above embodiments within the scope of this application.

Claims

1. A method for underground space planning and management based on multi-source data fusion, characterized in that, include: At the data access end, data from CAD drawings, GIS geographic data, BIM 3D models, and real-time geological sensing data are synchronously parsed to extract the core attributes corresponding to underground space components and generate a unified semantic representation. The core attributes include burial depth, elevation, geological category, and component type. The semantic representation is based on graph neural networks to construct cross-format attribute mapping relationships and is transmitted to the central database through a distributed consistency synchronization mechanism. In this process, blockchain nodes are embedded to record data operation trajectories, forming a traceable standardized dataset. Based on the standardized dataset and underground space geometric constraint rules, when multi-source data conflict, a corresponding adaptation strategy is dynamically generated. The adaptation strategy is executed by the local preprocessing module of the edge node. The local preprocessing module of the edge node performs a fast response correction process for sudden geological data and a batch fusion verification process for periodic construction data. After the conflict is resolved, in the edge-center collaborative architecture, the edge node cache dataset is dynamically configured in combination with geological conditions and construction progress. The reinforcement learning algorithm is called to reconstruct the corresponding three-dimensional visualization digital base plate based on the cached data and standardized dataset. Incremental update operation is performed synchronously, updating only the data of the changed components. Based on the aforementioned 3D visualized digital base, the AI ​​engine analyzes the planning specifications to generate structured review rules, completing compliance checks for pipeline conflicts and safety clearances; user feedback data is collected during the detection and operation process, and the attribute mapping rules and incremental update strategies corresponding to the graph neural network are updated based on the user feedback data.

2. The underground space planning and management method based on multi-source data fusion according to claim 1, characterized in that, The method also includes: When constructing cross-format attribute mapping relationships in graph neural networks, mapping layers are divided according to the type of underground space components, and each mapping layer corresponds to an independent feature extraction link. The feature extraction link is bound to the core attributes of the corresponding underground space components, and the mapping layer results are associated with the data operation trajectory recorded by the blockchain node and stored, and synchronously transmitted to the central database along with the standardized dataset.

3. The underground space planning and management method based on multi-source data fusion according to claim 2, characterized in that, Before dynamically generating the adaptation strategy, the method also includes: Based on the mapping layering results, corresponding conflict determination thresholds are configured for different component types. When the local preprocessing module of the edge node executes the adaptation strategy, it completes the correction and verification based on the conflict determination thresholds of the corresponding component types. The correction results and verification results are synchronously written to the blockchain node and supplemented to the data operation trajectory.

4. The underground space planning and management method based on multi-source data fusion according to claim 1, characterized in that, The method also includes: The parsed CAD drawings, GIS geographic data, and BIM 3D models are respectively subjected to format standardization preprocessing. The CAD drawings are extracted to extract layer component information and converted into vector data format. The GIS geographic data is supplemented with spatial reference system identifiers and coordinate deviations are corrected. The BIM 3D model is decomposed into component-level geometric units and structural attributes are extracted. Perform unified coordinate registration on CAD drawings, GIS geographic data and BIM 3D models, and use feature point matching algorithms to locate corresponding components in CAD drawings and BIM models, and establish spatial location association between GIS geographic data and BIM models. The coordinate-registered CAD drawings, GIS geographic data, and BIM 3D models are input into the graph neural network to supplement the layer affiliation, spatial zoning, and structural level attributes of components, optimize the construction accuracy of cross-format attribute mapping relationships, and simultaneously write the format conversion parameters and coordinate registration results into the blockchain node.

5. The underground space planning and management method based on multi-source data fusion according to claim 4, characterized in that, The method also includes: Based on the coordinate registration results, the geometric parameter consistency of the vector data of the CAD drawings and the geometric units of the BIM model is checked. The check includes component dimensions, burial depth error, and elevation deviation, and a unified check tolerance range is set. If there is a deviation in the verification, the geometric correction algorithm is called to synchronously adjust the vector data of the CAD drawing and the geometric units of the BIM model. Among them, the spatial location information of the corresponding components is updated synchronously in the GIS geographic data. After the correction is completed, the geometric verification results and correction parameters are added to the standardized dataset, the attribute mapping weights of the graph neural network are updated synchronously, and the correction process log is written to the blockchain node.

6. The underground space planning and management method based on multi-source data fusion according to claim 1, characterized in that, The method also includes: Construct an attribute source dictionary for three types of data: computer-aided design drawings, geographic data from geographic information systems, and building information models. Clarify the correspondence between synonymous attributes in different data sources. The correspondence between synonymous attributes includes the mapping rules between layer names of computer-aided design drawings and component categories of building information models, and between land use codes of geographic information systems and geological category attributes. Based on the attribute homology dictionary, the component attributes in the standardized dataset are deduplicated and merged, core attribute fields are retained and homology attribute notes are added to form a unified attribute system. Add timestamps to the attributes of each type of component to record the time nodes of data generation, modification and synchronization, build the data evolution trajectory based on the time series, and associate it with the operation trajectory of blockchain nodes; The attribute homology dictionary, deduplication and fusion results, and time series trajectories are synchronously added to the standardized dataset; when performing incremental update operations, the three types of data that have changed within the specified time interval are filtered based on the timestamp identifier of the attribute data, and the changed components are associated according to the attribute homology relationship to form a changed data set; The topology relationship reconstruction algorithm is called to update the component topology association of the 3D visualized digital base plate, and to verify the spatial accessibility and geometric constraint compatibility between the changed component and the surrounding components. The topology reconstruction results are fed back to the AI ​​engine to optimize the structured review rule judgment logic, and the attribute mapping rules of the graph neural network are updated synchronously. The topology reconstruction parameters and the optimized review rules are written into the blockchain node to supplement the data evolution trajectory.

7. The underground space planning and management method based on multi-source data fusion according to claim 6, characterized in that, The method also includes: After the unified semantic representation is generated, a multi-dimensional dynamic prediction model for geological parameters is constructed based on the time series features of historical geological datasets and real-time geological sensing data through a long short-term memory network, and the prediction results of confidence intervals for short-term key geological parameters are output. The confidence interval prediction results are cross-dimensionally correlated and mapped with the burial depth, elevation and geological category attributes of underground space components in the standardized dataset. The dynamic risk score of each underground space component is calculated, and risk warning indicators are generated. The risk warning indicators include geological stability risk level, potential conflict probability and predicted impact range. The geological stability risk level includes low risk, medium risk and high risk. Before dynamically generating the adaptation strategy, the risk warning indicators are used as the core input parameters. The conflict judgment threshold is dynamically corrected by the adaptive threshold adjustment algorithm based on the threshold function optimized by the genetic algorithm. The conflict judgment threshold of high-risk components is appropriately lowered, and the priority queue scheduling mechanism of the edge nodes is triggered to prioritize the execution of the fast response correction process for high-risk components. At the same time, the execution status of the correction process is fed back to the central database in real time. The generation process of risk warning indicators, dynamic threshold adjustment parameters, and priority queue scheduling results are synchronously recorded to blockchain nodes and stored in association with data operation trajectories.

8. An underground space planning and management system based on multi-source data fusion, characterized in that, include: The multi-source data semantic fusion module synchronously parses data from CAD drawings, GIS geographic data, BIM 3D models, and real-time geological sensing data at the data access end. It extracts the core attributes corresponding to underground space components and generates a unified semantic representation. The core attributes include burial depth, elevation, geological category, and component type. The semantic representation is based on a graph neural network to construct cross-format attribute mapping relationships and is transmitted to the central database through a distributed consistency synchronization mechanism. The synchronization process embeds blockchain nodes to record data operation trajectories, forming a traceable standardized dataset. The dynamic conflict adaptation execution module, based on the standardized dataset and underground space geometric constraint rules, dynamically generates corresponding adaptation strategies when multi-source data conflict. The adaptation strategies are executed by the edge node local preprocessing module. The edge node local preprocessing module performs a fast response correction process for sudden geological data and a batch fusion verification process for periodic construction data. After conflict resolution, the 3D digital base plate reconstruction module dynamically configures the cached dataset of edge nodes in the edge-center collaborative architecture, taking into account geological conditions and construction progress. It then calls reinforcement learning algorithms to reconstruct the corresponding 3D visualized digital base plate based on the cached data and standardized dataset, and synchronously performs incremental update operations, updating only the data of changed components. The intelligent compliance review optimization module, based on the aforementioned 3D visualized digital base, uses an AI engine to analyze planning specifications and generate structured review rules to complete compliance checks for pipeline conflicts and safety clearances; it also collects user feedback data during the detection and operation process and updates the attribute mapping rules and incremental update strategies corresponding to the graph neural network based on the user feedback data.

9. An electronic device, characterized in that, Includes a processor, wherein the processor runs a program for the underground space planning and management method based on multi-source data fusion as described in any one of claims 1-7.

10. A storage medium, characterized in that, The program stores the underground space planning and management method based on multi-source data fusion as described in any one of claims 1-7.

Images