A substation underground construction timing automatic reasoning and visualized disclosure method and device

Abstract

The application discloses a substation underground construction timing automatic reasoning and visualization disclosure method and device, including: extracting underground pipeline components and attribute information from a substation three-dimensional BIM model to construct a multi-level space semantic graph; applying five types of construction dependency rules to a pipeline construction dependency rule engine to construct a construction dependency directed acyclic graph; inputting the directed acyclic graph into a multi-constraint timing reasoning engine, sequentially executing constraint modeling, topological sorting with priority weight and constraint propagation based on an arc consistency algorithm to generate a construction timing scheme; dynamically binding the construction timing scheme with the BIM model to generate a phased construction timing animation; automatically outputting process disclosure texts for each construction process based on a domain knowledge graph enhanced sequence-to-sequence model, and integrating three-dimensional local views, process animations and disclosure texts into a two-dimensional code. The application realizes automatic analysis of construction dependency relationships and automatic reasoning of timing, and supports incremental reasoning to quickly respond to dynamic changes.

Description

Technical Field

[0001] This application relates to power engineering construction management, BIM and engineering intelligence, specifically to a method and device for automatic reasoning and visual disclosure of underground construction sequence in substations. Background Technology

[0002] Substation projects are characterized by compact sites, dense underground pipelines, and frequent overlapping construction processes. Underground construction involves the laying and pre-embedding of pipelines from multiple disciplines, including water supply and drainage, fire protection, power cables, grounding grids, communication optical cables, and HVAC systems, as well as the coordinated advancement of civil engineering processes such as road base construction, foundation pit support, and drainage ditches. In recent years, Building Information Modeling (BIM) technology has been widely applied in substation projects. Project managers typically create comprehensive 3D BIM models of the spatial location, attribute parameters, and design specifications of underground pipelines, foundations, and structures for each discipline, forming refined digital assets. However, the timing of substation underground construction still heavily relies on the personal experience of the project's technical manager. Pipeline construction dependencies must be manually assessed line by line. When dealing with hundreds of underground pipelines, it is easy to overlook implicit spatial layer dependencies and professional process dependencies, leading to systemic risks in the construction schedule.

[0003] Regarding the analysis of construction dependencies, while existing BIM software platforms possess pipeline collision detection capabilities, collision detection can only identify spatial interference between pipelines and cannot further infer the logical relationship between their construction sequence. For example, collision detection can report insufficient spatial spacing between two pipelines, but it cannot determine which pipeline should be constructed before the other, or whether the determination is based on spatial layer relationships, horizontal spacing specifications, professional priority requirements, or differences in laying methods. Existing design structure matrix methods can express dependencies between tasks, but require manually pre-defining a complete relationship matrix. They cannot automatically parse various types of construction dependencies from the pipeline spatial topology relationships in the BIM model. As the number of pipelines increases or design changes occur, the workload and error risk of manually maintaining the relationship matrix increase accordingly. Therefore, how to automatically extract the spatial topology information of pipelines from the BIM model and systematically parse various construction dependencies remains an unsolved problem in current technology.

[0004] In terms of construction sequence optimization, traditional critical path methods and program review and approval techniques require manual pre-definition of complete activity lists and logical relationship networks. They optimize time parameters under a known network structure and lack the ability to automatically extract and infer implicit dependencies between processes from the BIM model. Although existing construction scheduling methods have introduced constraint satisfaction techniques or metaheuristic optimization algorithms to handle schedule optimization under multiple constraints, these methods typically treat constraints as static inputs. They fail to form an organic inference pipeline of constraint modeling, topology sorting, and constraint propagation, and also lack the ability for incremental inference when dynamic changes occur at the construction site. In the substation underground construction scenario, process constraints include not only the sequential constraints between processes, but also time-sensitive constraints such as the requirement to complete support and grounding grid inspection within a specified time limit after excavation before subsequent civil construction can proceed, as well as constraints brought about by dynamic changes such as pipeline additions, deletions, or elevation adjustments. Current technologies require recalculation of the entire construction sequence after dynamic changes occur, resulting in a long response cycle and making it difficult to meet the rapid response requirements of substation projects.

[0005] Regarding construction visualization and briefing, existing BIM software's 4D construction simulation function requires engineers to manually bind each process node to a 3D model component. This manual associativity is labor-intensive and error-prone, failing to achieve automatic conversion from construction sequence plans to 4D construction animations. In the construction briefing stage, current practices mostly involve manually writing process briefing documents based on experience or using fixed templates for filling in blanks, lacking automated methods for generating briefing texts. The professionalism and completeness of the briefing documents depend on the individual abilities of the compilers. Some existing technologies use BIM models combined with QR codes to push information to the construction site, but the briefing content still requires manual associativity and is limited to static panoramic views, failing to integrate 3D partial views, process animations, and briefing texts into a multimodal visualized briefing solution.

[0006] The unique characteristics of underground substation construction further complicate the aforementioned technical challenges. As power infrastructure, substations involve specialized underground pipeline systems such as grounding grids, cable ducts, and cable trenches, which differ from conventional municipal pipelines in their laying rules and process requirements. The construction sequence is strictly constrained by the substation construction process; for example, roadbeds must be completed before laying pipelines crossing roads, and grounding grid welding must be completed and tested before subsequent civil construction can proceed. Simultaneously, substation construction must meet the requirements of civilized construction and standardized management, and construction handover documents must include both process parameters and safe and civilized construction specifications. The diverse types of pipelines, varying pipe diameters, diverse laying methods, strict safety distance requirements, and numerous industry-specific mandatory construction rules make it difficult to directly apply general construction progress optimization methods and handover schemes. Summary of the Invention

[0007] To address the technical challenges in substation underground construction scenarios, such as the reliance on manual, line-by-line judgment of pipeline construction dependencies leading to omissions, long response cycles for construction sequence arrangement under multiple constraints, the need for manual association in 4D construction simulation with difficulty in quickly responding to constraint changes, and the lack of automated means to generate construction handover documents, this invention provides an automatic reasoning and visualization handover method and device for substation underground construction sequence. Using the spatial relationships and attribute information of pipelines in the substation's 3D BIM model as input, a pipeline construction dependency rule engine automatically parses the construction dependencies between pipelines and constructs a directed acyclic graph. A multi-constraint sequence reasoning engine automatically infers and generates the optimal construction sequence scheme under multiple constraints. The reasoning results are then dynamically bound to the BIM model to automatically generate phased construction sequence animations and multimodal visualization handover content for the construction site.

[0008] One aspect of the present invention provides an automatic reasoning and visual disclosure method for underground construction sequence in substations, comprising the following steps.

[0009] Step S1 involves extracting underground pipeline components and their attribute information from the substation's 3D BIM model. A pipeline spatial adjacency map is constructed using pipeline components as nodes and spatial proximity relationships between pipelines as edges. Vertical layers are defined based on design elevations, and within each vertical layer, horizontal regions are further divided according to horizontal coordinates, forming a multi-level spatial semantic map. Specifically, the attribute information includes pipeline type, diameter, material, laying method, design elevation, horizontal coordinates, relevant discipline, and design specifications. When performing layered processing on the pipeline spatial adjacency map, pipeline components are divided into different vertical layers based on their design elevations. Within each vertical layer, horizontal regions are divided according to the proximity of horizontal coordinates, forming a multi-level spatial semantic map with cross-indexing of vertical layers and horizontal regions. Furthermore, the method for determining spatial proximity is as follows: A three-dimensional bounding box is extracted for each pipeline component; candidate pipeline pairs are screened through bounding box intersection tests; precise geometric calculations are performed on the candidate pipeline pairs to determine the minimum clearance; when the minimum clearance between two pipelines in three-dimensional space is less than a preset safety distance threshold or when their projections in the vertical direction overlap, they are determined to have a spatial proximity relationship. Preferably, the safety distance threshold is determined by industry standards based on the pipeline type and laying method; for example, the safety distance threshold between cable ducts and water supply pipelines is 0.5 meters, and the safety distance threshold between grounding grids and drainage pipelines is 0.3 meters.

[0010] Step S2 involves applying spatial hierarchy dependency rules, horizontal spacing dependency rules, professional priority dependency rules, laying method dependency rules, and infrastructure dependency rules to each pipeline pair with spatial proximity relationships in the multi-level spatial semantic graph using the pipeline construction dependency rule engine. The matched rules are then converted into directed edges, constructing a directed acyclic graph of construction dependencies. The pipeline construction dependency rule engine embeds five types of construction dependency rules, as detailed below.

[0011] The logic of the spatial layer dependency rule is as follows: when two pipelines have overlapping projections in the vertical direction and are located at different vertical layers, the pipeline with the lower elevation is constructed before the pipeline with the higher elevation. The directed edge points from the pipeline node with the lower elevation to the pipeline node with the higher elevation. This rule reflects the construction process requirement of laying underground pipelines layer by layer from bottom to top.

[0012] The logic of the horizontal spacing dependency rule is as follows: when the horizontal clearance between two pipelines is less than the minimum safe distance required by the standard, the pipeline with a larger diameter or deeper burial depth is constructed before the pipeline with a smaller diameter or shallower burial depth. The directed edge points from the node of the pipeline constructed first to the node of the pipeline constructed later.

[0013] The logic of the professional priority dependency rule is as follows: Construction is carried out according to the priority order of grounding grid, drainage pipelines, water supply pipelines, fire protection pipelines, cable ducts, cable trenches, communication optical cables, and HVAC pipelines. Based on the construction priority order of each professional pipeline as specified in the substation construction specifications, higher-priority professional pipelines must be constructed before lower-priority professional pipelines. This priority order can be customized and adjusted according to the specific design requirements of the project.

[0014] The logic of the laying method dependency rule is as follows: when there are pipelines with different laying methods on the same path, the pipelines constructed by trenching must be laid before the pipelines constructed by non-trenching, the direct buried pipelines must be laid before the pipelines laid by pipe laying or cable tray, and the directed edge points from the pipeline node of the earlier construction method to the pipeline node of the later construction method.

[0015] The logic of the infrastructure dependency rule is as follows: roadbed construction must precede pipeline construction crossing the road; foundation pit support must be constructed immediately after foundation pit excavation; and subgrade concrete must be poured after the corresponding pipelines are laid.

[0016] Furthermore, after constructing the construction-dependent directed acyclic graph (DAG), a loop detection algorithm is executed. If a directed loop is detected, a conflict resolution weight is calculated for each directed edge in the loop. This conflict resolution weight comprehensively considers the construction importance coefficient corresponding to the rule type that generated the directed edge, the significance of the attribute differences of the pipelines involved in the directed edge, and the structural importance of the directed edge in the DAG. The directed edge with the lowest conflict resolution weight is deleted to resolve the directed loop. Specifically, the construction importance coefficient is assigned according to the rule type, with rules related to safety distance having a higher weight than rules related to laying method; the significance of attribute differences is measured by quantitative indicators such as pipeline elevation difference and pipe diameter difference, with a higher weight for larger differences; and the structural importance is measured by the difference in the number of subsequent paths covered by the nodes connected by the directed edge. After the conflict resolution is completed, a resolution rationality verification is performed: for the construction dependency represented by the deleted directed edge, check whether the dependency is indirectly satisfied in the final construction sequence plan through the transitive relationship of other directed edges. If the indirect dependency is not valid, the implicit constraint is converted into a time interval constraint and attached to the corresponding process object.

[0017] Step S3 involves inputting the construction-dependent directed acyclic graph into a multi-constraint temporal reasoning engine, sequentially executing constraint modeling, priority-weighted topology sorting, and constraint propagation based on the arc consensus algorithm to generate a construction sequence scheme that satisfies physical space constraints, construction process constraints, and industry standard constraints. The multi-constraint temporal reasoning engine includes a constraint modeling module, a topology sorting module, and a constraint propagation module.

[0018] Specifically, the constraint modeling maps each node in the construction-dependent directed acyclic graph to a construction process object, and transforms industry standard constraints and construction process constraints into time constraint expressions attached to the corresponding construction process objects. These time constraint expressions include earliest start time constraints, latest finish time constraints, minimum time interval constraints between processes, maximum time interval constraints between processes, and resource mutual exclusion constraints. Each construction process object includes a process name, associated pipeline component identifiers, estimated working hours, process preconditions, and process postconditions. For example, the maximum time interval constraint between the foundation pit excavation process and the foundation pit support process indicates that the foundation pit support must be completed within a specified time limit after excavation; the minimum time interval constraint between the grounding grid welding inspection process and subsequent civil construction processes indicates that subsequent civil construction can only begin after the grounding grid inspection is passed.

[0019] The priority-weighted topology sorting selects a node based on its priority weight when multiple candidate nodes with zero in-degree exist. This priority weight is calculated by weighting and summing three factors: the construction priority coefficient of the pipeline's specialty, the range of influence of the pipeline's constraints on subsequent procedures, and the availability weight of construction resources. The range of influence of constraint propagation is measured by the number of subsequent nodes covered by the directed path originating from that node. By introducing priority weights, the topology sorting is no longer arbitrary but produces a reasonable arrangement of procedures that conforms to industry construction practices.

[0020] The constraint propagation based on the arc consensus algorithm propagates the earliest start time constraint forward along the directed edge direction and the latest completion time constraint backward along the directed edge direction. When the time window of a certain process node is empty, the conflict information is fed back to the topology sorting module to adjust the weights and re-sort. The constraint propagation is based on the arc consensus algorithm, starting from the process nodes directly restricted by industry standard constraints and construction process constraints, propagating the earliest start time constraint forward along the directed edge direction of the directed acyclic graph, and propagating the latest completion time constraint backward along the directed edge direction, until the time windows of all process nodes reach a stable state.

[0021] Furthermore, upon receiving a dynamic change event, the multi-constraint temporal reasoning engine performs incremental reasoning, re-performing constraint propagation only on the process nodes directly affected by the change event and their successor and predecessor nodes on the directed path. The dynamic change events include pipeline addition, pipeline deletion, pipeline elevation adjustment, and schedule compression. Key steps of incremental reasoning include: locating the affected subset of process nodes, updating the constraints of the affected nodes, performing local constraint propagation on the affected subset of nodes, and merging the results of the local constraint propagation into the global construction sequence plan.

[0022] Step S4 involves matching the pipeline components associated with each construction process in the construction sequence plan with the components in the BIM model, adding a time dimension attribute, dividing the time into slices according to the construction stage, and generating a phased construction sequence animation. Specifically, the dynamic binding process is as follows: matching the pipeline component identifiers associated with each construction process object in the construction sequence plan with the components in the BIM model, adding a time dimension attribute to each construction process, including the planned start time, planned completion time, and construction stage identifier, thus upgrading the 3D BIM model to a 4D construction model. Based on the 4D construction model, time slices are divided according to the construction stage identifiers. For each time slice, the set of pipeline components involved in that stage is extracted, generating a 3D construction scene snapshot for that stage. All time slices are arranged according to the construction sequence and automatically played in animation form, forming a phased construction sequence animation. Preferably, during the generation of the construction sequence animation, standardized civilized construction content is automatically integrated, specifically including: automatically adding 3D annotation models of barriers, warning signs, and safety passages to the animation scene of the road construction stage; automatically adding earthwork stockpiling area markers and dust suppression measure markers to the animation scene of the pipeline excavation stage; and automatically adding process annotations for fireproofing and grounding connections to the animation scene of the cable laying stage. The standardized civilized construction content is sourced from a pre-set standardized civilized construction template library, and the corresponding annotation models are automatically matched and inserted according to the pipeline type and construction stage.

[0023] Step S5 involves automatically generating process disclosure text for each construction process. The partial view of the 3D model, process animation clips, and disclosure text are encapsulated into a structured data package and encoded as a QR code. Specifically, the process disclosure text is generated using a sequence-to-sequence generation model enhanced with a domain knowledge graph. This domain knowledge graph contains entities related to pipeline types, construction methods, quality standards, and safety specifications in the substation underground construction field, along with their relationships. During generation, process knowledge entities related to the current process are retrieved from the knowledge graph and used as enhanced context input to the sequence-to-sequence generation model. For each construction process, attribute information of the associated pipeline components is extracted from the BIM model. Combined with the construction sequence parameters of the process, the sequence-to-sequence generation model is invoked to generate the process disclosure text. The process disclosure text includes a process overview section, a construction preparation section, a construction process section, a quality standard section, a safety precautions section, a civilized construction measures section, and an emergency response plan section.

[0024] Furthermore, the integrated encoding process of the QR code is as follows: For each construction procedure, a 3D partial view of the pipeline component set involved in that procedure is extracted from the 4D construction model and rendered as a static image; an animation segment corresponding to that procedure is extracted from the construction sequence animation and encoded as a short video; the process briefing text, the 3D partial view image, and the process animation short video are packaged into a structured data package; the structured data package is uploaded to the cloud server and the corresponding Uniform Resource Locator (URL) is obtained; the URL is then encoded into a QR code. After on-site personnel scan the QR code with their mobile terminals, the mobile terminals download and display the structured data package from the cloud server, simultaneously displaying the 3D partial view, playing the process animation short video, and presenting the process briefing text on the mobile terminal screen.

[0025] Another aspect of the present invention provides an automatic reasoning and visualization disclosure device for underground construction sequence in substations, comprising: a spatial semantic graph construction module, used to extract underground pipeline components and their attribute information from the three-dimensional BIM model of the substation, construct a pipeline spatial adjacency graph with pipeline components as nodes and spatial proximity relationships between pipelines as edges, divide vertical layers according to design elevations, and divide horizontal regions within each vertical layer according to horizontal coordinates to form a multi-level spatial semantic graph; and a construction dependency parsing module, used to apply spatial layer dependency rules, horizontal spacing dependency rules, professional priority dependency rules, laying method dependency rules, and infrastructure dependency rules to pipeline pairs with spatial proximity relationships in the multi-level spatial semantic graph through a pipeline construction dependency rule engine, and to determine the rules that are matched. The system transforms the data into directed edges, constructing a construction-dependent directed acyclic graph (DAG). A multi-constraint temporal reasoning module sequentially performs constraint modeling, priority-weighted topological sorting, and arc-consistency-based constraint propagation on the DAG to generate a construction sequence plan that satisfies physical space constraints, construction process constraints, and industry standard constraints. A 4D visualization binding module matches pipeline components associated with each construction process in the construction sequence plan with BIM model components, adds time dimension attributes, divides time slices according to construction stages, and generates phased construction sequence animations. A visualization handover generation module automatically generates process handover text for each construction process, encapsulating the 3D model partial view, process animation clips, and handover text into a structured data package and encoding it as a QR code.

[0026] This invention automatically converts the spatial topology and attribute information of pipelines in a BIM model into a directed acyclic graph (DAG) of construction dependencies through a pipeline construction dependency rule engine. This replaces the tedious process of manually determining the construction dependencies between pipelines one by one, solving the technical problem of easily overlooking implicit dependencies when dealing with hundreds of underground pipelines. Through a three-stage joint reasoning mechanism of constraint modeling, topology sorting, and constraint propagation in a multi-constraint temporal reasoning engine, this invention automatically generates a construction sequence plan that satisfies all constraints, including physical space constraints, construction process constraints, and industry standard constraints. It also supports incremental reasoning to quickly update the construction sequence after dynamic changes occur. Furthermore, by dynamically binding the construction sequence plan to the BIM model and automatically generating phased construction sequence animations, this invention achieves automated construction of 4D construction simulation, eliminating the workload of manually associating components one by one. This invention automatically generates complete process disclosure texts containing construction process parameters, safety precautions, and civilized construction measures through a sequence-to-sequence generation model enhanced by domain knowledge graph. It also integrates partial views of the 3D model, process animations, and disclosure texts into visualized disclosure content that can be scanned and viewed on the construction site through QR code integration encoding technology. Attached Figure Description

[0027] The embodiments of the present invention will now be described in detail with reference to the accompanying drawings.

[0028] Figure 1 This is a schematic diagram of the overall process of the automatic reasoning and visual disclosure method for underground construction sequence of substations provided in this embodiment of the invention.

[0029] Figure 2 This is a schematic diagram of the structure of the pipeline construction dependency rule engine for constructing a directed acyclic graph of construction dependencies, provided in an embodiment of the present invention.

[0030] Figure 3 This is a schematic diagram of the three-stage joint reasoning process of the multi-constraint temporal reasoning engine provided in the embodiments of the present invention.

[0031] Figure 4 This is a structural block diagram of the automatic reasoning and visualization disclosure device for underground construction sequence of substations provided in an embodiment of the present invention.

[0032] Figure 5 This is a schematic diagram illustrating the pipeline spatial adjacency relationship and the construction of a multi-level spatial semantic graph according to the present invention.

[0033] Figure 6 This is a schematic diagram of the three-factor dynamic weighting mechanism for conflict resolution in this invention.

[0034] Figure 7 This is a schematic diagram illustrating the constraint propagation forward and backward reasoning and time window convergence of the present invention.

[0035] Figure 8This is a schematic diagram of the vertical strata of the underground pipelines in the substation of this invention.

[0036] Figure 9 This is a comparison chart of the performance of construction dependency parsing and temporal reasoning in this invention.

[0037] Figure 10 The heatmap shows the contribution of the ablation experimental technique of this invention.

[0038] Figure 11 This is a graph comparing the response times of incremental reasoning and global reasoning in this invention. Detailed Implementation

[0039] To make the objectives and technical solutions of this invention clearer, the embodiments of this invention will be further described in detail below with reference to the accompanying drawings. The following embodiments are only used to more clearly illustrate the technical solutions of this invention and should not be construed as limiting the scope of protection of this invention.

[0040] Example 1

[0041] See Figure 1 This embodiment provides an automatic reasoning and visual disclosure method for underground construction sequence in substations, using a newly built 220kV substation as an application scenario. The substation's underground pipeline system includes grounding grids, water supply and drainage pipelines, fire protection pipelines, cable ducts, cable trenches, communication optical cables, and HVAC pipelines, totaling over 300 pipelines distributed at different elevations from 0.5 meters to 4.2 meters below the ground surface, involving various laying methods such as direct burial, ductwork, cable trenches, and cable trays. The method includes the following steps.

[0042] Step S1: Extract underground pipeline components and their attribute information from the 3D BIM model of the substation, and construct a multi-level spatial semantic map.

[0043] First, the application programming interface (API) of the BIM model is used to traverse all components marked as underground pipelines in the 3D BIM model of the substation, extracting the attribute information of each pipeline component. The attribute information includes pipeline type (e.g., water supply pipe, drainage pipe, cable duct, grounding flat steel, etc.), pipe diameter (in millimeters), material (e.g., high-density polyethylene, ductile iron, galvanized steel pipe, etc.), laying method (direct burial, ductwork, cable trench, cable tray), design elevation (in meters, with reference to the engineering datum), horizontal coordinates (x and y coordinate values ​​in the engineering coordinate system), affiliated discipline (e.g., water supply, drainage, electrical, low-voltage, HVAC, etc.), and design specifications (e.g., pipe wall thickness, cable cross-sectional area, net dimensions inside the trench, etc.). In this embodiment, complete attribute information for 328 pipeline components was extracted from the BIM model.

[0044] Then, a pipeline spatial adjacency graph is constructed using pipeline components as nodes and spatial proximity relationships between pipelines as edges. The determination of spatial proximity relationships employs a two-level screening strategy: the first level is coarse screening, where the 3D bounding box (Axis-Aligned Bounding Box, or AABB) of each pipeline component is extracted, and the bounding box is expanded (the expansion amount equals the safety distance threshold corresponding to the pipeline type). Candidate pipeline pairs that may have spatial proximity relationships are quickly screened through bounding box intersection tests. The second level is fine screening, where precise geometric calculations are performed on the candidate pipeline pairs to determine the minimum clearance. The minimum clearance is obtained by calculating the shortest Euclidean distance between the centerlines of the two pipelines and subtracting the outer radius of each pipeline. When the minimum clearance between two pipelines in 3D space is less than the preset safety distance threshold, they are determined to have a spatial proximity relationship. Furthermore, when the projections of two pipelines in the vertical direction overlap, even if their horizontal clearance is greater than the safety distance threshold, they are also determined to have a spatial proximity relationship. The safety distance threshold is determined by industry standards based on the type of pipeline and the laying method. For example, the safety distance threshold between cable ducts and water supply pipelines is 0.5 meters, the safety distance threshold between grounding grids and drainage pipelines is 0.3 meters, and the safety distance threshold between high-voltage cables and low-voltage communication optical cables is 0.5 meters.

[0045] Furthermore, the pipeline spatial adjacency map is layered. Vertical layers are defined based on design elevations, and within each vertical layer, horizontal regions are further divided based on horizontal coordinates, forming a multi-level spatial semantic map. Specifically, the design elevation range of the pipelines is divided into several vertical layers at equal intervals. The thickness of each layer is preferably 0.5 to 1.0 meters, and in this embodiment, it is set to 0.8 meters. For each vertical layer, the horizontal coordinates of the pipelines within that layer are projected onto a plane. A density-based clustering algorithm (such as the DBSCAN algorithm, where the neighborhood radius is set to 2.0 meters and the minimum sample size is set to 2) is used to cluster horizontally adjacent pipelines into the same horizontal region. Finally, a multi-level spatial semantic map with cross-indexing of vertical layers and horizontal regions is obtained. Each node in this map contains not only the original attribute information of the pipeline component but also its vertical layer identifier and horizontal region identifier. In the application scenario of this embodiment, 328 pipeline components are divided into 5 vertical layers and 23 horizontal regions.

[0046] like Figure 8As shown, this invention illustrates the multi-level spatial distribution of underground pipelines in a substation from a vertical cross-sectional perspective. The top of the figure shows the ground line and the substation building outline, while the sides are the substation perimeter walls. The underground portion is divided into four vertical layers according to elevation: Layer 1 (-0.6m to -1.4m) contains DN100 water supply pipes, DN150 fire-fighting pipes, communication optical cables, and DN100 HVAC pipes; Layer 2 (-1.4m to -2.2m) contains DN200 drainage pipes, four-hole cable conduits, and 40x4 grounding flat steel; Layer 3 (-2.2m to -3.0m) contains DN300 drainage pipes, a 600x400 cable trench, and DN200 water supply pipes; Layer 4 (-3.0m to -3.8m) contains the main grounding grid and DN400 drainage pipes. Each pipeline cross-section is drawn as a circle or rectangle according to its actual shape. The pipeline name and diameter are marked by the side leader line. The layers are separated by horizontal dashed lines and the elevation value is marked. The safety distance between adjacent pipelines is marked. A depth scale is provided on the right side.

[0047] like Figure 5 As shown, the multi-level spatial semantic graph constructed by this invention divides pipeline components into several vertical layers according to design elevation, and forms horizontal regions within each vertical layer by clustering using horizontal coordinates. The pipeline nodes in the graph are arranged in layers according to vertical layers. Layer 1 (-0.6m to -1.4m) includes DN100 water supply pipes, DN150 fire-fighting pipes, communication optical cables, and DN100 HVAC pipes; Layer 2 (-1.4m to -2.2m) includes DN200 drainage pipes, 4-hole cable conduits, and 40x4 grounding flat steel; Layer 3 (-2.2m to -3.0m) includes DN300 drainage pipes, cable trenches, and DN200 water supply pipes; and Layer 4 (-3.0m to -3.8m) includes the main grounding grid line and DN400 drainage pipes. Within the same layer, nodes are clustered into region A and region B based on their horizontal proximity. Solid and dashed lines between nodes represent spatial proximity within the same region and across regions, respectively. The values ​​marked on the connecting lines represent the minimum net distance between two pipelines.

[0048] Step S2: Apply five types of construction dependency rules to each pipeline pair with spatial proximity in the multi-level spatial semantic graph through the pipeline construction dependency rule engine to construct a directed acyclic graph of construction dependencies.

[0049] See Figure 2 The pipeline construction dependency rule engine embeds five types of construction dependency rules, ranked from highest to lowest priority: infrastructure dependency rules, spatial hierarchy dependency rules, horizontal spacing dependency rules, professional priority dependency rules, and laying method dependency rules. For each pair of pipelines with spatial proximity in the multi-level spatial semantic graph, the engine sequentially checks whether the above five types of rules are matched.

[0050] The spatial layer dependency rule applies to pipeline pairs located at different vertical layers: when two pipelines have overlapping projections in the vertical direction and are located at different vertical layers, the pipeline with the lower elevation is constructed before the pipeline with the higher elevation, and the directed edge points from the node of the pipeline with the lower elevation to the node of the pipeline with the higher elevation. The technical basis for this rule is the construction process requirement of laying underground pipelines layer by layer from bottom to top—if the high-elevation pipeline is laid first and then the trench for the low-elevation pipeline is excavated, it may cause disturbance and displacement of the foundation of the already laid pipeline. In this embodiment, the spatial layer dependency rule applied to 147 pipeline pairs.

[0051] The horizontal spacing dependency rule applies to pipeline pairs with insufficient horizontal spacing within the same vertical layer: when the horizontal clearance between two pipelines is less than the minimum safe distance required by the standard, the pipeline with the larger diameter or deeper burial depth should be constructed before the pipeline with the smaller diameter or shallower burial depth. The technical basis for this rule is that the trench excavation and backfilling operations for large-diameter pipelines have a larger impact area. If the trench for the small-diameter pipeline is constructed first and then the trench for the large-diameter pipeline is excavated, the small-diameter pipeline may be affected by the deformation of the trench slope.

[0052] The professional priority rule is based on the professional order in the substation construction specifications: the construction priority is ordered as follows: grounding grid, drainage pipelines, water supply pipelines, fire protection pipelines, cable ducts, cable trenches, communication optical cables, and HVAC pipelines. This professional priority order reflects industry practice in substation construction—the grounding grid, as the foundation of electrical safety, must be constructed first and pass inspection before other processes can proceed; the drainage system must be completed as early as possible during the site excavation stage to prevent water accumulation in the foundation pit; and the preparation period for the construction conditions of high-voltage pipelines is longer than that for low-voltage pipelines. This priority order can be customized according to the specific design requirements of the project.

[0053] The laying method dependency rule applies to pipelines with different laying methods along the same path: pipelines laid using trenching must be laid before those laid using non-trenching methods, and direct-buried pipelines must be laid before pipelines laid in cable trays or conduits. The technical basis for this rule is that trenching causes greater disturbance to the ground than non-trenching methods. If trenching work is carried out on a path where non-trenched pipelines have already been laid, it may damage the existing pipelines.

[0054] Infrastructure dependency rules address dependencies between pipelines and civil infrastructure: roadbed construction must precede pipeline construction crossing the road, foundation pit support must be constructed immediately after foundation pit excavation, and subgrade concrete must be poured after the corresponding pipelines are laid.

[0055] After converting the dependencies matched by the above five rules into directed edges, all directed edges and pipeline nodes together form a directed acyclic graph (DAG) of construction dependencies. After construction, a depth-first search-based cycle detection algorithm is executed to verify that no directed cycles exist in the graph. If a directed cycle is detected, a dynamic weight resolution mechanism based on pipeline attribute features is used to resolve the conflict. Specifically, a conflict resolution weight is calculated for each directed edge in the directed cycle. The conflict resolution weight comprehensively considers three factors: The first factor is the construction importance coefficient corresponding to the rule type that generated the directed edge. The coefficient for infrastructure dependency rules is 1.0, the coefficient for spatial layer dependency rules is 0.9, the coefficient for horizontal spacing dependency rules is 0.7, the coefficient for professional priority dependency rules is 0.6, and the coefficient for laying method dependency rules is 0.5. The second factor is the significance of the attribute differences of the pipelines involved in the directed edge. For example, the larger the elevation difference, the higher the weight of spatial layer dependency; the larger the pipe diameter difference, the higher the weight of horizontal spacing dependency. The significance of attribute differences is measured by the normalized difference ratio. The third factor is the structural importance of the directed edge in the directed acyclic graph, which is measured by the number of nodes covered by the source node connected by the directed edge in the subsequent path. The more covered nodes, the higher the structural importance. The three factors are weighted and summed according to preset weights (preferred values ​​are 0.4, 0.3, and 0.3) to obtain the conflict resolution weight. The directed edge with the lowest conflict resolution weight is deleted to resolve the directed cycle.

[0056] After conflict resolution is completed, a resolution rationality verification is performed: For the construction dependencies represented by the deleted directed edges, it is checked whether the dependency is indirectly satisfied in the final construction sequence plan through the transitive relationship of other directed edges. That is, after deleting the directed edge, does a directed path still exist from the original earlier construction node to the later construction node? If the indirect dependency is valid, the resolution result is reasonable; if the indirect dependency is invalid, the implicit constraint is converted into a time interval constraint and attached to the corresponding process object to ensure construction safety. The entire conflict resolution process is recorded in a resolution log for review by engineering personnel.

[0057] like Figure 6 As shown, the conflict resolution mechanism of this invention is completed in three stages. The left panel displays the directed loop detection results. A directed loop is formed between nodes C, D, and E in the figure. Each directed edge in the loop is labeled with its corresponding conflict resolution weight value. The middle panel displays the three-factor weight decomposition of each directed edge in the loop. The stacked horizontal bar chart decomposes the conflict resolution weight of each edge into the weighted contribution of three factors: construction importance coefficient, attribute difference significance, and structural importance. Among them, the conflict resolution weight of directed edge C to D is 0.72, the conflict resolution weight of directed edge D to E is the highest at 0.85, and the conflict resolution weight of directed edge E to C is the lowest at 0.43. The right panel displays the resolution results. The system deletes the directed edge E to C with the lowest conflict resolution weight to resolve the directed loop. The deleted edges are marked with gray dashed lines.

[0058] Step S3: Input the construction-dependent directed acyclic graph into the multi-constraint temporal reasoning engine, and generate a construction time sequence scheme through a three-stage joint reasoning process of constraint modeling, topological sorting with priority weights, and constraint propagation based on arc consensus algorithm.

[0059] See Figure 3 The first stage of the multi-constraint temporal reasoning engine is constraint modeling. The constraint modeling module maps each node in the construction dependency directed acyclic graph to a construction process object. Each construction process object includes a process name, associated pipeline component identifiers, estimated working hours (calculated based on engineering quotas), process preconditions, and process postconditions. The constraint modeling module transforms industry standard constraints and construction process constraints into time constraint expressions and attaches them to the corresponding construction process objects. These time constraint expressions include five types: earliest start time constraint, used to limit the process from starting earlier than a specified time (e.g., subsequent civil construction can only begin after the grounding grid inspection is qualified); latest finish time constraint, used to limit the process from being completed before a specified time (e.g., schedule milestone requirements); minimum time interval constraint between processes, used to ensure that there is at least a specified interval between two processes (e.g., after concrete pouring, it must be cured to the design strength before proceeding to the next process, with a curing period of no less than 7 days); maximum time interval constraint between processes, used to limit the interval between two processes from exceeding a specified time (e.g., after foundation pit excavation, support must be completed within 48 hours); and resource mutual exclusion constraint, used to ensure that two mutually interfering construction operations cannot be carried out simultaneously in the same area.

[0060] The second stage is topology sorting with priority weights. The topology sorting module executes an improved Kahn algorithm on the construction-dependent directed acyclic graph. In each step of the topology sorting, when multiple candidate nodes with an in-degree of zero exist, a node is selected based on its priority weight. This priority weight is calculated by weighting and summing three factors: the construction priority coefficient of the pipeline's specialty, the range of influence of the pipeline's constraint propagation on subsequent processes, and the availability weight of construction resources. The construction priority coefficient is assigned according to the order in the specialty priority dependency rules (e.g., 1.0 for grounding grids, 0.9 for drainage pipelines, 0.8 for water supply pipelines, etc.). The range of influence of constraint propagation is measured by the number of subsequent nodes covered by the directed path originating from that node; the number of subsequent nodes is normalized before being used as the value of this factor. The availability weight of construction resources reflects the idle level of the construction team and machinery at the current time point, preferably ranging from 0 to 1. The preferred weighting coefficients for the three factors are 0.35, 0.40, and 0.25. By introducing priority weights, topology sorting can produce a reasonable arrangement of procedures that meets industry construction practices, rather than producing arbitrary legal topology sequences.

[0061] The third stage is constraint propagation based on the ArcConsistency (AC-3) algorithm. The constraint propagation module uses the ArcConsistency (AC-3) algorithm to perform constraint propagation reasoning on the initial construction time sequence generated by topological sorting. Starting from the process nodes directly constrained by industry standards and construction process constraints, the module propagates the earliest start time constraint forward along the directed edges of the directed acyclic graph—for a directed edge (u, v), the earliest start time of node v is updated to max(v's current earliest start time, u's earliest start time + u's estimated working hours + the minimum time interval constraint between u and v). It then propagates the latest finish time constraint backward along the directed edges—for a directed edge (u, v), the latest finish time of node u is updated to min(u's current latest finish time, v's latest finish time - v's estimated working hours - the minimum time interval constraint between u and v). Forward and backward propagation are performed alternately until the time windows (i.e., the interval between the earliest start time and the latest finish time) of all process nodes reach a stable state. When the constraint propagation process finds that the time window of a certain process node is empty (i.e., the earliest start time is later than the latest finish time), it indicates that there is an unsatisfactorable constraint conflict in the current construction sequence. The constraint propagation module feeds back the conflict information to the topology sorting module, which adjusts the priority weights and re-sorts the constraints. Then, the constraint propagation is executed again, iterating up to 5 times, until all constraints can be satisfied or the maximum number of iterations is reached.

[0062] like Figure 7 As shown, the constraint propagation mechanism of this invention is presented in two parts. The upper part is a directed acyclic graph view, showing the construction dependencies of eight process nodes, including grounding grid welding, grounding grid inspection, drainage pipe laying, water supply pipe laying, cable duct installation, cable trench construction, subgrade concrete pouring, and roadbed construction. Key constraints, such as a minimum interval of 7 days and a maximum interval of 48 hours, are marked on the directed edges. The black solid arrows indicate the forward propagation direction used to propagate the earliest start time constraint, and the gray dashed arrows indicate the reverse propagation direction used to propagate the latest completion time constraint. The lower part is the corresponding Gantt chart view, with the horizontal axis representing the construction date. Each process corresponds to a horizontal bar. The light gray area represents the initial time window before constraint propagation, and the dark gray area represents the final time window after convergence of forward and reverse constraint propagation. The converged time window is significantly narrower than the initial time window, demonstrating the precise constraint effect of constraint propagation on the construction sequence.

[0063] Furthermore, upon receiving a dynamic change event, the multi-constraint temporal reasoning engine performs incremental reasoning, re-performing constraint propagation only on the process nodes directly affected by the change event and their successor and predecessor nodes on the directed path. The dynamic change events include pipeline addition, pipeline deletion, pipeline elevation adjustment, and schedule compression. Key steps of incremental reasoning include: traversing forward and backward along the directed path from the change node using breadth-first search to locate the affected subset of process nodes; updating the constraints of the affected nodes (e.g., updating corresponding directed edges and temporal constraints due to changes in spatial layer dependencies caused by pipeline elevation adjustments); performing local constraint propagation on the affected subset of nodes (forward and backward propagation only within the subset); and merging the results of local constraint propagation into the global construction sequence scheme to verify whether the time windows at the boundary nodes are consistent with the global scheme. Through the incremental reasoning mechanism, the response time for dynamic changes is reduced from minutes or even hours of recalculating the entire construction sequence to seconds.

[0064] Step S4: Dynamically bind the construction sequence plan with the BIM model to generate phased construction sequence animations.

[0065] The dynamic binding process involves matching the pipeline component identifiers associated with each construction procedure object in the construction sequence plan with the components in the BIM model. Precise correspondence is achieved using globally unique identifiers (GUIDs) for each component. A time dimension attribute is added to each construction procedure, including planned start time, planned completion time, and construction stage identifier, upgrading the 3D BIM model to a 4D construction model. Based on the 4D construction model, time slices are divided according to the construction stage identifiers. Each time slice corresponds to an independent construction stage. The set of pipeline components involved in each stage is extracted, and a 3D construction scene snapshot for that stage is generated using a 3D rendering engine. All time slices are arranged according to the construction sequence and automatically played in animation format, forming a phased construction sequence animation. The animation playback supports stage-by-stage jumps and variable speed playback.

[0066] During the generation of construction sequence animations, standardized civilized construction content is automatically integrated. Specifically, based on the type of construction stage, matching 3D annotation models are retrieved from a pre-set standardized civilized construction template library and automatically inserted into the animation scene: 3D annotation models of barriers, warning signs, and safety passages are automatically added to the animation scene for the road construction stage; earthwork stockpiling area markers and dust suppression measure markers are automatically added to the animation scene for the pipeline excavation stage; and process annotations for fireproofing and grounding connections are automatically added to the animation scene for the cable laying stage. The standardized civilized construction template library stores pre-modeled 3D models of various standardized civilized construction components and their insertion rules. The insertion rules define the construction stage and spatial location where each standardized civilized construction component should appear.

[0067] Step S5: Automatically generate process disclosure text for each construction process, and encapsulate the partial view of the 3D model, process animation clips, and disclosure text into a structured data package and encode it into a QR code.

[0068] The automatic generation process of the process disclosure text is as follows: For each construction procedure, the attribute information of the associated pipeline components (including pipeline type, diameter, material, design elevation, laying method, interface type, and design specifications) is extracted from the BIM model. Combined with the construction sequence parameters of the procedure (including planned construction date, name of the preceding procedure, name of the following procedure, and construction duration), the process disclosure text is generated by calling a sequence-to-sequence generation model enhanced by a domain knowledge graph. The domain knowledge graph pre-constructs process knowledge entities in the field of substation underground construction—including pipeline type, construction method, quality standard, and safety specification—and their relationships. When generating the process disclosure text, the knowledge graph is used to retrieve process knowledge entities and associated entities related to the current procedure. The retrieval results are used as enhanced context input to the sequence-to-sequence generation model to improve the professionalism and completeness of the disclosure text.

[0069] The sequence-to-sequence generation model employs an encoder-decoder structure based on the Transformer architecture. The encoder receives structured representations of pipeline attribute information and construction timing parameters, as well as textual representations of knowledge graph-enhanced context, as input sequences. The decoder autoregressively generates process disclosure text. The model's training dataset consists of approximately 5,000 to 10,000 process disclosure documents written by senior engineers from historical substation construction projects. These documents were manually reviewed and annotated before being used as training samples. The training process uses the AdamW optimizer with an initial learning rate of 2.3 × 10^-4, a batch size of 16, and 30 to 50 training epochs. A cosine annealing learning rate decay strategy is employed, and cross-entropy loss is used as the loss function. Model training is performed on a server equipped with four GPUs, using gradient accumulation to simulate a larger effective batch size. Evaluation metrics include BLEU score and ROUGE-L score to assess the similarity between the generated text and the manually written text. Domain experts are also invited to manually review the professionalism and completeness of the generated text. When there is a significant deviation between the distribution of pipeline types at the construction site and the distribution in the training data (for example, the proportion of briefing documents for a certain type of pipeline in the training data is less than 5%), the quality of the generated text may decrease. In this case, it is recommended to manually review the briefing documents for that type of pipeline.

[0070] The process handover text includes seven paragraphs: the process overview paragraph describes the construction content, pipelines involved, and spatial location of the process; the construction preparation paragraph lists the required materials, equipment, and personnel configuration for the process; the construction process paragraph details the construction steps and process parameters of the process; the quality standard paragraph lists the quality inspection standards and acceptance points for the process; the safety precautions paragraph lists the safety risk points and protective measures for the process; the civilized construction measures paragraph lists the standardized civilized construction requirements that must be implemented for the process; and the emergency response plan paragraph lists the possible abnormal situations and emergency response plans for the process.

[0071] The integrated encoding process of the QR code is as follows: For each construction procedure, a 3D partial view of the pipeline components involved in that procedure is extracted from the 4D construction model and rendered as a static image (preferably with a resolution of 1920×1080 pixels). An animation segment corresponding to that procedure is extracted from the construction sequence animation and encoded as a short video in H.264 format (generally between 30 and 120 seconds in length). The process disclosure text, 3D partial view image, and process animation short video for that procedure are encapsulated into a structured data package in JSON format. This structured data package is uploaded to the cloud server and the corresponding Uniform Resource Locator (URL) is obtained. The URL is then encoded into a QR code (using the QR Code standard, with an error correction level of L). After on-site personnel scan the QR code with their mobile terminals, the mobile terminals download and display the structured data package from the cloud server. Simultaneously, the 3D partial view, the process animation short video, and the process disclosure text are displayed on the mobile terminal screen, enabling visualized and readily accessible access to construction disclosure information.

[0072] Example 2

[0073] In a preferred embodiment of the present invention, the systematic application of five types of rules in the pipeline construction dependency rule engine and the conflict resolution mechanism are explained in detail.

[0074] The pipeline construction dependency rule engine receives a multi-level spatial semantic graph as input. For each pair of pipelines with spatial proximity in the graph, it checks whether five types of construction dependency rules are matched according to a preset rule detection order. The rule detection order directly affects the final constructed directed acyclic graph of construction dependencies—when the same pipeline pair is matched by multiple rules simultaneously, the rule detection order determines the order in which multiple directed edges are generated, thus affecting the number of directed cycles to be processed in the subsequent cycle detection stage. In this embodiment, the rule detection order is arranged from high to low according to the degree of influence of the rules on construction safety: infrastructure dependency rules take precedence over spatial layer dependency rules, spatial layer dependency rules take precedence over horizontal spacing dependency rules, horizontal spacing dependency rules take precedence over professional priority dependency rules, and professional priority dependency rules take precedence over laying method dependency rules.

[0075] For the spatial layer dependency rule, its judgment condition includes a logical AND operation of two sub-conditions: the first sub-condition is that the vertical projection areas of the two pipelines have an intersection with a greater than zero overlapping area; the second sub-condition is that the absolute value of the difference in design elevation between the two pipelines is greater than a preset layer distinction threshold (preferred value is 0.15 meters, which is less than the thickness of a vertical layer). When both sub-conditions are satisfied simultaneously, the rule is hit, and a directed edge is generated from the pipeline node with the lower elevation to the pipeline node with the higher elevation. The calculation method for vertical projection overlap is as follows: extract the projection contours of the two pipelines on the horizontal plane (take the projection circle for circular pipelines and the projection rectangle for rectangular cross-section pipelines), and calculate the intersection area of ​​the two projection contours using the Sutherland-Hodgman polygon clipping algorithm.

[0076] For the horizontal spacing dependency rule, the determination condition is that the horizontal clearance between two pipelines is less than the minimum safety clearance required by the standard. The minimum safety clearance is retrieved from the safety clearance lookup table based on the combination of the two pipeline types. This lookup table stores the minimum safety clearance values ​​for each pair of pipeline types in matrix form. When there is no corresponding entry for the pipeline type combination in the lookup table, the default value of 0.5 meters is used. After the rule is matched, directed edges are generated according to the pipe diameter priority principle: the pipeline with the larger diameter is the first node to be constructed; if the pipe diameters are the same, the pipeline with the deeper burial depth is the first node to be constructed; if both the pipe diameter and burial depth are the same, the order is determined by the component numbering order of the pipelines in the BIM model.

[0077] The application condition for the professional priority dependency rule is that two pipelines belong to different professions and there is no directed edge connecting these two nodes in the directed graph. Since the same pipeline pair may have already been hit by the spatial layer dependency rule or the horizontal spacing dependency rule and generated a directed edge, the professional priority dependency rule only generates a directed edge when the preceding rule is not hit, in order to avoid generating redundant parallel directed edges. The priority relationship between each level of professional priority ranking is encoded as a directed edge direction determination rule—when the professional priority of the pipeline constructed earlier is higher than that of the pipeline constructed later, the directed edge points from the higher priority pipeline node to the lower priority pipeline node.

[0078] The calculation of conflict resolution weights is a crucial step in the conflict resolution mechanism. Upon detecting a directed cycle, the system extracts all directed edges within the cycle and calculates their conflict resolution weights one by one. The first factor (construction importance coefficient) ranges from 0.5 to 1.0, and its specific value is given in Example 1. The second factor (attribute difference significance) is calculated as follows: for directed edges generated by spatial layer dependency rules, the attribute difference significance equals the absolute value of the elevation difference between the two pipelines divided by the maximum elevation difference of the pipeline system; for directed edges generated by horizontal spacing dependency rules, the attribute difference significance equals the absolute value of the pipe diameter difference between the two pipelines divided by the maximum pipe diameter difference of the pipeline system; for directed edges generated by other rules, the attribute difference significance is set to 0.5. The third factor (structural importance) is calculated as follows: a depth-first traversal is performed on the source node of the directed edge, and the number of all successor nodes reachable from the source node is counted. This number is then divided by the total number of nodes in the directed acyclic graph to obtain the normalized structural importance value. The conflict resolution weight is obtained by summing the three factors with weights of 0.4, 0.3, and 0.3, with values ​​ranging from 0 to 1.

[0079] The resolution validity verification process performs a reachability check on each deleted directed edge. Let the deleted directed edge be (A, B), indicating that the original rule requires pipeline A to be constructed before pipeline B. The verification process starts from node A and performs a breadth-first search on the directed graph after deleting the edge to determine whether node B is still reachable. If it is reachable, it means that there is an indirect path that allows pipeline A to still be constructed before pipeline B in the final construction sequence, and the resolution result is valid. If it is not reachable, the deleted construction dependency is transformed into a soft constraint—specifically, a minimum time interval constraint is added between the two corresponding construction process objects, with the constraint value set to 0 (i.e., it only requires that the construction of pipeline A is not later than the start time of the construction of pipeline B). This constraint is incorporated into the reasoning during the constraint propagation phase, thereby maintaining the semantics of the original construction dependency as much as possible without introducing directed cycles.

[0080] The systematic application of the above five types of rules and the technical effects of the conflict resolution mechanism are reflected in the following aspects. In the underground construction scenario of substations with dense intersections of multiple professional pipelines, the orderly detection and hierarchical application of the five types of rules can extract the construction dependencies implicit in the BIM model in a structured way and express them explicitly as directed acyclic graphs, transforming the construction dependencies from implicit engineering experience knowledge into computable and traceable graph structure data. Compared with static rule priority resolution, the dynamic weight resolution mechanism based on pipeline attribute characteristics can adaptively adjust the resolution strategy according to the attribute differences of specific pipeline pairs. When the pipeline elevation difference is large, it tends to retain spatial layer dependency edges, and when the pipe diameter difference is large, it tends to retain horizontal spacing dependency edges, improving the rationality of resolution decisions under complex working conditions. The closed-loop verification of resolution rationality ensures that even if directed edges are deleted, the semantics of the original construction dependencies are still indirectly or softly constrained in the final time sequence scheme.

[0081] From a principle perspective, the five rule categories cover five dimensions of substation underground pipeline construction dependencies: vertical spatial dimension (spatial layer dependency), horizontal spatial dimension (horizontal spacing dependency), professional process dimension (professional priority dependency), construction technology dimension (laying method dependency), and infrastructure dimension (infrastructure dependency). This multi-dimensional rule coverage enables the rule engine to capture various types of dependencies between pipelines, reducing the risk of overlooking implicit dependencies. The three-factor design of conflict resolution weights (rule importance, attribute difference, and structural importance) evaluates the retention value of each directed edge from three perspectives: rule level, data level, and graph structure level. The weighted fusion of these three factors ensures that the resolution decision considers not only the rigid requirements of construction safety (through the rule importance factor), but also the physical characteristics of specific pipeline pairs (through the attribute difference factor), and the impact of the dependency relationship on the overall construction sequence (through the structural importance factor).

[0082] Example 3

[0083] In another embodiment of the invention, an alternative to the constraint propagation network is provided for reasoning construction timing, replacing the three-stage joint reasoning mechanism in Example 1.

[0084] In this alternative, constraint modeling, topology sorting, and constraint propagation are no longer executed sequentially as three separate stages. Instead, the construction-dependent directed acyclic graph (DAG) is directly transformed into a constraint propagation network, on which the construction sequence reasoning is synchronously completed using a message-passing algorithm. Specifically, the constraint propagation network is constructed as follows: each pipeline node in the construction-dependent DAG is mapped to a variable node in the constraint propagation network, and the value domain of this variable node is the feasible construction time interval of that pipeline operation (initialized to the entire project duration); each directed edge in the DAG is mapped to a constraint node in the constraint propagation network, and this constraint node encodes the corresponding construction sequence constraint and time interval constraint.

[0085] The message-passing algorithm executes iteratively on the constraint propagation network. In each iteration, each constraint node collects the current value range of its two connected variable nodes, calculates the range of values ​​that should be pruned for each variable node based on the constraints, and sends the pruning information as a message to the corresponding variable node. After receiving messages from all adjacent constraint nodes, each variable node takes the intersection of all pruning information and updates its own value range. The message-passing algorithm iterates until the value ranges of all variable nodes no longer change (reaching a fixed point) or it is detected that the value range of a certain variable node has been pruned to an empty set (indicating that the constraint is unsatisfiable).

[0086] After message passing converges, the value range of each variable node represents the feasible construction time window for that pipeline operation under all constraints. The node with the narrowest value range is selected from all variable nodes, and its construction time is fixed as the lower bound of its value range (earliest feasible time). This fixed decision is then propagated as a new constraint to adjacent nodes, triggering a new round of message passing. This "selection-fixing-propagation" process is repeated until the construction times of all variable nodes are determined, thus obtaining a complete construction sequence plan. The strategy of selecting the node with the narrowest value range for fixing is similar to the "most constrained variable" heuristic in constraint satisfaction problems, enabling early detection and backtracking of constraint conflicts and reducing the search space.

[0087] The difference between this alternative and the three-stage joint reasoning scheme lies in the following: the three-stage joint reasoning scheme determines the basic order of operations through topological sorting and then refines the time parameters through constraint propagation, its core being a sequential approach of "sorting first, constraint adjustment"; while the constraint propagation network scheme integrates sorting and constraint reasoning synchronously during message passing, without needing to predetermine the order of operations, the timing scheme is entirely determined by the result of constraint propagation. In small substation projects with a small number of pipelines (less than 100) and relatively simple constraints, the constraint propagation network scheme has a simpler reasoning process because it eliminates the topological sorting stage. However, when the number of pipelines exceeds 300 and there are a large number of resource mutual exclusion constraints, the convergence speed of message passing may decrease significantly. In this case, the three-stage joint reasoning scheme has better computational efficiency due to the pre-pruning effect of the search space during the topological sorting stage.

[0088] Regarding dynamic change handling, the constraint propagation network scheme also supports incremental message passing—upon receiving a dynamic change event, it is only necessary to reinitialize the value domains of the affected variable nodes and trigger local message passing from these nodes, without having to rebuild the entire constraint propagation network. In the constraint propagation network, because the constraint nodes explicitly encode construction dependencies, the scope of the impact of dynamic changes can be accurately located through the adjacency relationships of the constraint nodes, avoiding unnecessary global recalculation.

[0089] Example 4

[0090] See Figure 4 This embodiment provides an automatic reasoning and visualization disclosure device for underground construction sequence of substations. The device is deployed on an engineering management server equipped with a multi-core processor, at least 64GB of memory, and at least 2TB of storage space, and is equipped with BIM model parsing middleware and a 3D rendering engine. The device includes the following five functional modules.

[0091] The spatial semantic graph construction module extracts underground pipeline components and their attribute information from the substation's 3D BIM model. It constructs a pipeline spatial adjacency graph using pipeline components as nodes and spatial proximity relationships between pipelines as edges. Vertical layers are defined based on the pipeline's design elevation, and within each vertical layer, horizontal regions are further divided according to horizontal coordinates, forming a multi-level spatial semantic graph. This module reads model data through the BIM model's application programming interface (API), performs intersection tests and precise geometric calculations on the 3D bounding boxes of pipeline components to determine spatial proximity relationships, and employs a density-based clustering algorithm to divide the pipelines within each vertical layer into horizontal regions. The output of the spatial semantic graph construction module is a multi-level spatial semantic graph data structure, which is passed to the construction dependency parsing module via a memory-sharing mechanism.

[0092] The construction dependency parsing module applies spatial hierarchy dependency rules, horizontal spacing dependency rules, professional priority dependency rules, laying method dependency rules, and infrastructure dependency rules to each pipeline pair with spatial proximity in the multi-level spatial semantic graph using the pipeline construction dependency rule engine. It converts the matched rules into directed edges, constructing a directed acyclic graph (DAG) of construction dependencies. This module embeds a rule configuration file to store the parameters and priorities of various rules, allowing engineers to adjust the professional priority order and safety spacing thresholds. The construction dependency parsing module also performs cycle detection and conflict resolution, outputting resolution logs to a log storage module for engineer review. The output of the construction dependency parsing module, the DAG of construction dependencies, is then passed to the multi-constraint temporal reasoning module.

[0093] The multi-constraint temporal reasoning module sequentially performs constraint modeling, priority-weighted topology sorting, and arc-consistency-based constraint propagation on the construction-dependent directed acyclic graph (DAG), generating a construction sequence plan that satisfies physical space constraints, construction process constraints, and industry standard constraints. This module receives the DAG output from the construction dependency parsing module and externally input schedule constraint parameters. In the constraint modeling submodule, graph nodes are mapped to construction process objects and time constraint expressions are appended. In the topology sorting submodule, priority-weighted topology sorting is performed. In the constraint propagation submodule, arc-consistency constraint propagation is performed. When a dynamic change event is received, this module performs incremental reasoning, recalculating only the affected subset of process nodes. The output of the multi-constraint temporal reasoning module is the construction sequence plan, which is simultaneously passed to the 4D visualization binding module and the visualization briefing generation module.

[0094] The 4D visualization binding module matches pipeline components associated with each construction process in the construction sequence plan with BIM model components, adds a time dimension attribute to each construction process, divides the construction into time slices according to construction stages, and generates phased construction sequence animations. This module calls a 3D rendering engine to generate a construction scene snapshot for each time slice and automatically inserts matching annotated models from the standardized template library into the animation scene. The output of the 4D visualization binding module includes a 4D construction model and phased construction sequence animations, which are then passed to the visualization briefing generation module.

[0095] The visualization-based briefing generation module automatically generates process briefing text for each construction procedure. It encapsulates partial views of the 3D model, process animation clips, and briefing text into a structured data package and encodes it as a QR code. This module integrates a domain knowledge graph and a sequence-to-sequence generation model based on the Transformer architecture. It receives 3D partial views and animation clips from the 4D visualization binding module, combines pipeline attribute information extracted from the BIM model, and construction sequence parameters to generate process briefing text. After encapsulating and uploading these three types of information, it encodes them into a QR code for on-site personnel to scan and view.

[0096] Example 5

[0097] To verify the feasibility and technical effectiveness of the present invention, a new 110kV substation construction project is used as an application case. This substation covers an area of ​​approximately 8,000 square meters. The underground pipeline system includes eight categories of specialized pipelines: grounding flat steel, water supply pipes, drainage pipes, fire-fighting pipes, high-voltage cable ducts, low-voltage cable trenches, communication optical cables, and HVAC pipelines, totaling 256 pipelines. These pipelines are distributed at elevations ranging from 0.6 meters to 3.8 meters below the ground surface, involving four laying methods: direct burial, ductwork, cable trenches, and cable trays. The construction period for this project is 180 days, involving five specialized construction teams. Coordination of the cross-construction of various specialized pipelines within a limited site space is required.

[0098] The construction sequence reasoning of this project was performed using the method of this invention, and the specific process is as follows. In step S1, 256 pipeline components were extracted from the 3D BIM model of the substation. 1847 pairs of pipelines with spatial proximity were identified through bounding box intersection testing and precise geometric calculation. These were divided into four vertical layers based on a 0.8-meter layer thickness, and further divided into 18 horizontal regions using DBSCAN clustering. The construction of the multi-level spatial semantic map took approximately 12 seconds. In step S2, the pipeline construction dependency rule engine checked each of the 1847 pipeline pairs for five types of rules. Among these, 98 pairs of spatial layer dependency rules were hit, 63 pairs of horizontal spacing dependency rules were hit, 141 pairs of professional priority dependency rules were hit, 37 pairs of laying method dependency rules were hit, and 24 pairs of infrastructure dependency rules were hit, generating a total of 363 directed edges. In the cycle detection phase, seven directed cycles were identified. After deleting these seven directed edges using a dynamic weight resolution mechanism, a construction-dependent directed acyclic graph was constructed. The resolution validity verification results showed that five of the deleted directed edges had indirectly satisfied paths, while two did not and were converted into time interval constraints. Step S2 took approximately 8 seconds. In step S3, constraint modeling generated 256 construction process objects and attached 46 time constraint expressions. Topological sorting produced a valid process arrangement in the first round. After three rounds of forward and backward constraint propagation, the time windows of all process nodes reached a stable state, and no unsatisfiable constraint conflicts were found. Step S3 took approximately 15 seconds. The entire construction timing reasoning process (steps S1 to S3) took a total of 35 seconds.

[0099] To evaluate the effectiveness of incremental reasoning, a typical design change event was simulated—the design elevation of three drainage pipelines was lowered by 0.4 meters. Incremental reasoning located a subset of 17 affected process nodes (including 3 directly affected nodes and 14 successor / predecessor propagation nodes). Local constraint propagation converged within 2 rounds. Incremental reasoning took approximately 1.8 seconds, while performing global re-reasoning for the same change event took 31 seconds.

[0100] To evaluate the performance advantages of the incremental reasoning mechanism of this invention, design change events of different scales were simulated on the same 110kV substation project, and the response time of incremental reasoning and global re-reasoning was recorded respectively. The test conditions were: a total of 256 pipelines, and change scenarios with 5, 10, 17, 30, 50 and 80 affected process nodes were simulated in sequence. Each group of experiments was repeated 10 times and the average value was taken.

[0101] like Figure 11 As shown, the horizontal axis represents the number of affected process nodes, and the vertical axis represents the inference time (logarithmic scale). The solid dotted curve represents the global re-inference time, which increased from 15 seconds with 5 affected nodes to 105 seconds with 80 affected nodes; the dashed square dotted curve represents the incremental inference time, which increased from 0.5 seconds with 5 affected nodes to 10.5 seconds with 80 affected nodes. The gray filled area between the two curves represents the time saving of incremental inference compared to global inference, with a response time reduction of approximately 94%. The vertical dashed line in the figure marks the experimental point in Example 5 (17 affected nodes), corresponding to a comparison of incremental inference time of 1.8 seconds and global re-inference time of 31 seconds.

[0102] To compare the technical effectiveness, three senior technicians from the project were invited to arrange the construction sequence using manual experience methods, with the evaluation results of an industry expert review panel serving as the benchmark. Evaluation dimensions included dependency completeness (the coverage rate of construction dependency identification), constraint satisfaction rate (the proportion of the final sequence scheme that satisfies all constraints), and arrangement time. The dependency completeness of the method presented in this invention was 96.8%, while the average of the three technicians was 81.3%. The constraint satisfaction rate of the method presented in this invention was 100% (constraint propagation ensures all constraints are satisfied), while the average of the three technicians was 89.7% (there were instances where some inter-process time interval constraints were not satisfied). The arrangement time of the method presented in this invention was 35 seconds, while the average arrangement time of the three technicians was 3.2 days.

[0103] To verify the performance of the construction dependency resolution and timing inference of this invention, a new 110kV substation construction project (256 pipelines) was used as the test object. The method of this invention was compared with the manual experience method of three senior technicians. The experimental environment was a project management server equipped with a multi-core processor and 64GB of memory. Evaluation metrics included dependency integrity, constraint satisfaction rate, and orchestration time.

[0104] like Figure 9 As shown, the left vertical axis corresponds to two sets of bar charts displaying dependency integrity and constraint satisfaction rates, respectively. Regarding dependency integrity, the method of this invention achieves 96.8%, while the three technical personnel achieved 78.5%, 83.1%, and 82.3%, respectively, with an average of 81.3%. Regarding constraint satisfaction rate, the method of this invention achieves 100%, while the three technical personnel achieved 87.2%, 91.5%, and 90.4%, respectively, with an average of 89.7%. The right vertical axis corresponds to a line chart displaying the arrangement time for each group (logarithmic scale). The arrangement time of the method of this invention is only 35 seconds, while the average arrangement time of the three technical personnel is 3.2 days, a difference of more than three orders of magnitude.

[0105] To evaluate the technical contributions of each core innovation, ablation comparison experiments were conducted. After removing the conflict resolution mechanism (i.e., replacing dynamic weight resolution with simple static rule priority resolution), dependency integrity decreased from 96.8% to 91.2%, a drop of 5.6 percentage points. This was because static resolution made suboptimal resolution decisions for complex pipeline pairs where multiple rules were simultaneously applied in three locations. After removing incremental reasoning (i.e., performing global re-reasoning for each change), constraint satisfaction remained unchanged at 100%, but change response time increased from 1.8 seconds to 31 seconds. After removing domain knowledge graph enhancement (i.e., directly generating disclosure text from sequence-to-sequence models without using knowledge graph retrieval results as enhancement context), the BLEU score of the generated process disclosure text decreased from 0.687 to 0.523, the ROUGE-L score decreased from 0.714 to 0.581, and the domain expert's rating of the disclosure text's professionalism (out of 5) decreased from 4.1 to 3.3.

[0106] To evaluate the contribution of each core technology component of this invention, an ablation comparison experiment was conducted on the same 110kV substation project (256 pipelines). The evaluation indicators included six dimensions: dependency integrity, constraint satisfaction rate, change response time, BLEU score, ROUGE-L score, and expert score.

[0107] like Figure 10As shown, the heatmap's vertical axis represents four ablation configurations (complete method, conflict resolution removed, incremental reasoning removed, and knowledge graph enhancement removed), and the horizontal axis represents six evaluation metrics. Each cell is labeled with a specific value, and the color intensity indicates the normalized performance level. Values ​​marked with an asterisk represent significant reductions exceeding 5% relative to the complete method. After removing the conflict resolution mechanism, dependency integrity decreased from 96.8% to 91.2%; after removing incremental reasoning, change response time increased from 1.8 seconds to 31 seconds; after removing domain knowledge graph enhancement, the BLEU score decreased from 0.687 to 0.523, the ROUGE-L score decreased from 0.714 to 0.581, and the expert score decreased from 4.1 to 3.3. The results indicate that each core component makes an irreplaceable contribution to the overall system performance.

[0108] The above description is merely a preferred embodiment of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for automatic reasoning and visual disclosure of underground construction sequence in substations, characterized in that, Includes the following steps: Step S1: Extract underground pipeline components and their attribute information from the 3D BIM model of the substation. Construct a pipeline spatial adjacency diagram with pipeline components as nodes and spatial proximity relationships between pipelines as edges. Divide vertical layers according to design elevation and divide horizontal areas within each vertical layer according to horizontal coordinates to form a multi-level spatial semantic map. Step S2: Apply spatial hierarchy dependency rules, horizontal spacing dependency rules, professional priority dependency rules, laying method dependency rules and infrastructure dependency rules to pipeline pairs with spatial proximity relationships in the multi-level spatial semantic graph through the pipeline construction dependency rule engine, and transform the hit rules into directed edges to construct a construction dependency directed acyclic graph. Step S3: Input the construction-dependent directed acyclic graph into the multi-constraint temporal reasoning engine, and sequentially execute constraint modeling, topological sorting with priority weights, and constraint propagation based on arc consistency algorithm to generate a construction sequence scheme that satisfies physical space constraints, construction process constraints, and industry standard constraints. Step S4: Match the pipeline components associated with each construction process in the construction sequence plan with the BIM model components, add time dimension attributes, divide the time slices according to the construction stage, and generate phased construction sequence animations. Step S5: Automatically generate process disclosure text for each construction process, and encapsulate the partial view of the 3D model, process animation clips, and disclosure text into a structured data package and encode it into a QR code.

2. The method for automatic reasoning and visual disclosure of underground construction sequence in substations according to claim 1, characterized in that, In step S1, the method for determining the spatial proximity relationship is as follows: extract the three-dimensional bounding box of each pipeline component, screen candidate pipeline pairs through bounding box intersection test, perform precise geometric calculation on the candidate pipeline pairs to determine the minimum clearance, and determine that the two pipelines have a spatial proximity relationship when the minimum clearance between the two pipelines in the three-dimensional space is less than the preset safety distance threshold or when there is an overlapping area in the projection in the vertical direction.

3. The method for automatic reasoning and visual disclosure of underground construction sequence in substations according to claim 1, characterized in that, In step S2, the logic of the spatial layer dependency rule is as follows: when two pipelines have overlapping projections in the vertical direction and are located in different vertical layers, the pipeline with the lower elevation is constructed before the pipeline with the higher elevation; the logic of the horizontal spacing dependency rule is as follows: when the horizontal net distance between two pipelines is less than the minimum safe distance required by the standard, the pipeline with the larger diameter or deeper burial depth is constructed before the pipeline with the smaller diameter or shallower burial depth; the logic of the professional priority dependency rule is as follows: the construction priority is ordered according to the grounding grid, drainage pipeline, water supply pipeline, fire protection pipeline, cable duct, cable trench, communication optical cable, and HVAC pipeline.

4. The method for automatic reasoning and visual disclosure of underground construction sequence in substations according to claim 1, characterized in that, In step S2, after constructing the construction-dependent directed acyclic graph, a loop detection algorithm is executed. If a directed loop is detected, a conflict resolution weight is calculated for each directed edge in the directed loop. The conflict resolution weight comprehensively considers the construction importance coefficient corresponding to the rule type that generated the directed edge, the significance of the attribute differences of the pipeline involved in the directed edge, and the structural importance of the directed edge in the directed acyclic graph. The directed edge with the lowest conflict resolution weight is deleted to resolve the directed loop.

5. The method for automatic reasoning and visual disclosure of underground construction sequence in substations according to claim 4, characterized in that, After the conflict resolution is completed, a resolution rationality verification is performed: for the construction dependency represented by the deleted directed edge, check whether the dependency is indirectly satisfied in the final construction sequence scheme through the transitive relationship of other directed edges. If the indirect dependency is not valid, the construction dependency is converted into a time interval constraint and attached to the corresponding process object.

6. The method for automatic reasoning and visual disclosure of underground construction sequence in substations according to claim 1, characterized in that, In step S3, the constraint modeling maps each node in the construction dependency directed acyclic graph to a construction process object, and transforms industry standard constraints and construction process constraints into time constraint expressions and attaches them to the corresponding construction process objects. The time constraint expressions include the earliest start time constraint, the latest finish time constraint, the minimum time interval between processes constraint, the maximum time interval between processes constraint, and resource mutual exclusion constraints.

7. The method for automatic reasoning and visual disclosure of underground construction sequence in substations according to claim 1, characterized in that, In step S3, when there are multiple candidate nodes with an in-degree of zero, the topology sorting with priority weights selects the node based on its priority weight value. The priority weight value is calculated by weighting and summing three factors: the construction priority coefficient of the pipeline's specialty, the influence range of the pipeline's constraint propagation on subsequent processes, and the availability weight of construction resources. The constraint propagation based on the arc consensus algorithm propagates the earliest start time constraint forward along the directed edge direction and the latest completion time constraint backward along the directed edge direction. When the time window of a certain process node is empty, the conflict information is fed back to the topology sorting module to adjust the weights and re-sort.

8. The method for automatic reasoning and visual disclosure of underground construction sequence in substations according to claim 1, characterized in that, In step S3, when a dynamic change event is received, the multi-constraint temporal reasoning engine performs incremental reasoning, and only re-executes constraint propagation on the process nodes directly affected by the change event and their successor and predecessor nodes on the directed path. The dynamic change event includes pipeline addition, pipeline deletion, pipeline elevation adjustment and schedule compression.

9. The method for automatic reasoning and visual disclosure of underground construction sequence in substations according to claim 1, characterized in that, In step S5, the process disclosure text is generated using a sequence-to-sequence generation model enhanced by a domain knowledge graph. The domain knowledge graph contains entities related to pipeline types, construction methods, quality standards, and safety specifications in the field of substation underground construction, as well as their relationships. During generation, process knowledge entities related to the current process are retrieved from the knowledge graph and used as enhanced context input to the sequence-to-sequence generation model. The process disclosure text includes a process overview section, a construction preparation section, a construction process section, a quality standard section, a safety precautions section, a civilized construction measures section, and an emergency response plan section.

10. An automatic reasoning and visualization disclosure device for underground construction sequence of substations, characterized in that, include: The spatial semantic graph construction module is used to extract underground pipeline components and their attribute information from the 3D BIM model of the substation. It constructs a pipeline spatial adjacency graph with pipeline components as nodes and spatial proximity relationships between pipelines as edges. It divides vertical layers according to the design elevation and divides horizontal regions within each vertical layer according to the horizontal coordinates to form a multi-level spatial semantic graph. The construction dependency parsing module is used to apply spatial hierarchy dependency rules, horizontal spacing dependency rules, professional priority dependency rules, laying method dependency rules and infrastructure dependency rules to pipeline pairs with spatial proximity relationships in the multi-level spatial semantic graph through the pipeline construction dependency rule engine, and convert the hit rules into directed edges to construct a directed acyclic graph of construction dependencies. The multi-constraint temporal reasoning module is used to sequentially perform constraint modeling, topological sorting with priority weights, and constraint propagation based on arc consensus algorithm on the construction-dependent directed acyclic graph to generate a construction sequence scheme that satisfies physical space constraints, construction process constraints, and industry standard constraints. The 4D visualization binding module is used to match the pipeline components associated with each construction process in the construction sequence plan with the BIM model components, add time dimension attributes, divide the time slices according to the construction stage, and generate phased construction sequence animations. The visualization-based briefing generation module is used to automatically generate process briefing texts for each construction process. It encapsulates the partial views of the 3D model, process animation clips, and briefing texts into a structured data package and encodes them into QR codes.

Images