A knowledge graph-based three-dimensional line path intelligent planning method for distribution network

Abstract

This invention discloses an intelligent planning method for three-dimensional power distribution network routes based on a knowledge graph, comprising the following steps: acquiring planning data and unifying coordinate and elevation benchmarks, assigning unique identifiers to spatial objects and equipment objects to form a three-dimensional planning base; generating three-dimensional passable corridors, constructing a three-dimensional passability cost field and loading differentiated cost mappings for overhead, cable, and mixed laying methods to form a laying method-related cost field; constructing a power distribution network planning knowledge graph containing rule and exception layers; generating candidate three-dimensional routes and writing them into the knowledge graph to form scheme nodes; extracting scheme subgraphs and inputting them into an improved Graph-Mamba model for inference output of risk, cost, and schedule predictions and evidence importance; hard constraint verification to obtain a feasible candidate set; multi-objective screening to output a Pareto minimum feasible scheme set and a planning result package. This invention improves the feasibility and interpretability of three-dimensional route planning.

Description

Technical Field

[0001] This invention relates to the field of distribution network planning and intelligent decision-making technology, and in particular to a knowledge graph-based intelligent planning method for three-dimensional line paths in distribution networks. Background Technology

[0002] With the increasing demand for refined construction and digital operation and maintenance of power distribution networks, intelligent planning technology for three-dimensional power distribution network routes in complex urban environments has attracted widespread attention. Existing route planning methods largely rely on two-dimensional geographic information overlaid with rule constraints and employ shortest path or single-objective weighted cost search to generate solutions. However, in practical applications, the following problems commonly exist: The diverse sources of planning input data and the difficulty in unifying the expression of elements such as three-dimensional geometry, ledgers, ownership, and risk zoning lead to deviations in corridor accessibility and cost assessment. Rules, construction permits, risk and cost information are scattered across multiple systems, lacking structured connections and traceable evidence chains, resulting in insufficient interpretability of the schemes and high review and verification costs. The cost differences between overhead, cable, and mixed laying methods are often handled with fixed empirical weights, making it difficult to apply them consistently in candidate generation and screening. Hard constraint verification and multi-objective comparison of candidate paths are usually postponed and separated from model reasoning, making it difficult to form a stable Pareto screening and result package output on the feasible candidate set.

[0003] Therefore, how to provide a knowledge graph-based intelligent planning method for three-dimensional power distribution network routes is a problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0004] One objective of this invention is to propose an intelligent planning method for three-dimensional power distribution network routes based on knowledge graphs. This invention constructs a three-dimensional planning base and a passable corridor, establishes differentiated cost mappings for overhead, cable, and mixed laying methods to form a cost field, generates candidate three-dimensional routes and writes them into a knowledge graph containing rule and exception layers to form a scheme subgraph, and uses an improved Graph-Mamba model combined with hard constraint verification and Pareto screening to output a minimum feasible solution set. It has the advantages of strong three-dimensional adaptation, high screening consistency, interpretable evidence chain, and multi-objective optimization.

[0005] According to an embodiment of the present invention, a knowledge graph-based intelligent planning method for three-dimensional power distribution network routes includes the following steps: Step 1: Obtain relevant data for the planning area and unify coordinates and elevation benchmarks; assign unique identifiers to spatial objects and equipment objects to form a three-dimensional planning base. Step 2: Generate a 3D passable corridor based on the 3D planning base, construct a 3D passable cost field, and establish differentiated cost mappings for overhead, cable and mixed laying methods. Load the differentiated cost mappings into the 3D passable cost field to form a laying method-related cost field. Step 3: Construct a knowledge graph for power distribution network planning, including a rule layer and an exception layer; Step 4: Generate no less than a preset number of candidate 3D paths under the cost field constraints related to the 3D passable corridor and the laying method, and determine the corresponding laying method for each candidate 3D path. Write each candidate 3D path into the power distribution network planning knowledge graph to form a scheme node and establish a relationship. Step 5: Extract a subgraph of the scheme with a preset number of hops centered on each scheme node, and input the scheme subgraph into the improved Graph-Mamba model for inference. The improved Graph-Mamba model includes a three-dimensional geometry-topology joint location encoding module, a rule-exception evidence gating module, a scheme-oriented adaptive serialization module, a Graph-Mamba backbone inference module, and a multi-task output head module, which outputs scheme embeddings and risk indicator predictions, cost indicator predictions, schedule indicator predictions, and evidence importance results. Step 6: Perform hard constraint verification on the candidate 3D paths based on the rule layer and 3D geometric constraints to obtain a set of feasible candidates; Step 7: Within the feasible candidate set, perform multi-objective screening based on the cumulative cost corresponding to the cost field related to the laying method, as well as the prediction of risk indicators, cost indicators, schedule indicators, and the importance of evidence, and output the Pareto minimum feasible solution set and the three-dimensional route planning result package.

[0006] Optionally, step one specifically includes: Acquire relevant data for the planning area, including three-dimensional spatial data, channel and ownership data, power distribution equipment object data, and risk zoning data; The coordinate system and elevation datum of the three-dimensional spatial data are unified. The channel and ownership data, distribution network equipment object data and risk zoning data are projected onto the unified coordinate system and a unique identifier is established for each. Based on the unified 3D spatial data, a 3D raster index for the planning area is constructed. The density, occlusion ratio and noise level of each raster are statistically analyzed to form a spatial quality field. The spatial quality field, along with the geometric information, attribute information and unique identifier of the corresponding object, are written into the same data structure to obtain the 3D planning base.

[0007] Optionally, step two specifically includes: A three-dimensional passable corridor is generated based on a three-dimensional planning base. The three-dimensional passable corridor is composed of grid units in the three-dimensional grid index of the planning area that meet the passability conditions. The passability conditions include the spatial range corresponding to the avoidance of restricted land use signs, the volume range of the avoidance of building entities, the volume range of the avoidance of water bodies, and meeting the preset clearance conditions and preset setback conditions. For different laying methods, separate access cost items are constructed. The access cost items for overhead laying include crossing cost, corner cost, pole access cost, and clearance margin cost. The access cost items for cable laying include road excavation cost, trench or pipe jacking cost, crossing cost, and access cost to existing cable channels. The access cost items for mixed laying include overhead section cost, cable section cost, and overhead section to cable section conversion cost. Each passage cost item is normalized according to a unified scale to form a basic three-dimensional passage cost field. Establish a differentiated cost mapping for overhead, cable and mixed laying methods. The differentiated cost mapping includes setting laying method-related weights for each access cost item and setting laying method conversion costs. Load the laying method-related weights and laying method conversion costs into the basic three-dimensional access cost field to obtain the laying method-related cost field.

[0008] Optionally, step three specifically includes: A distribution network planning knowledge graph is constructed, which is represented by a directed attribute graph and consists of a set of nodes and a set of edges. Set attribute fields for each node, construct an edge set and set the edge type and edge attribute fields; The distribution network planning knowledge graph is divided into a rule layer and an exception layer. The rule layer is formed by connecting the procedure clause nodes with the equipment entity nodes and spatial object nodes through the procedure constraint association edges, and the procedure clause nodes are marked with hard constraint clause labels or soft constraint clause labels. The exception layer is formed by connecting the exception condition nodes with the procedure clause nodes through the exception association edges, and the exception condition nodes are connected to the construction method nodes, permit item nodes and cost element nodes through the construction constraint association edges, permit item association edges and cost impact association edges, respectively.

[0009] Optionally, step four specifically includes: Within the three-dimensional passable corridor, the starting point object and the ending point object are determined. The starting point object is the equipment entity node corresponding to the power supply point, and the ending point object is the equipment entity node corresponding to the load access point. Candidate generation channels for overhead laying method, cable laying method and mixed laying method are determined in the cost field related to the laying method. Candidate 3D paths are generated under the constraints of the cost field related to the 3D passable corridor and the laying method. The generation methods include priority search along the road corridor, priority search to avoid high-risk zones, priority search with the fewest crossings, and priority search with the fewest turns. A preset number of candidate 3D paths are output for each generation method. For each candidate 3D path, the corresponding laying method is determined by dividing the candidate 3D path into overhead section and cable section according to the path segment. Based on the cumulative cost of the path segment in the laying method related cost field, the overhead laying method, cable laying method or mixed laying method is selected as the laying method identifier of the candidate 3D path. Each candidate 3D path is written into the distribution network planning knowledge graph to form a scheme node. The attribute fields of the scheme node include path geometry description, laying method identifier and path segment sequence description, and an association relationship is established for the scheme node.

[0010] Optionally, the improved Graph-Mamba model includes a three-dimensional geometry-topology joint location encoding module, a rule-exception evidence gating module, a scheme-oriented adaptive serialization module, a Graph-Mamba backbone inference module, and a multi-task output header module; The three-dimensional geometry-topology joint location encoding module generates node three-dimensional location features based on the spatial geometric description of device entity nodes and spatial object nodes in the scheme subgraph, and generates edge geometric features based on the edge attribute fields of spatially associated edges in the scheme subgraph. It performs two-layer linear mapping on the node three-dimensional location features and edge geometric features respectively to obtain node location embedding and edge location embedding. Finally, it concatenates the node location embedding with the corresponding node attribute field encoding vector to obtain the node input representation. The rule-exceptional evidence gating module constructs a gating input vector based on the hard constraint clause identifier and soft constraint clause identifier of the procedure clause node in the scheme subgraph, the trigger condition description and additional cost field of the exception condition node, and the laying method identifier of the scheme node. It performs a linear mapping on the gating input vector and obtains the node gating weight and edge gating weight through the Sigmoid function. It multiplies the node gating weight and the node input representation element by element to obtain the gated node representation, and writes the edge gating weight into the edge position embedding to form the gated edge representation. The adaptive serialization module for the scheme generates an input sequence centered on the scheme node. The input sequence is structured in layers: the first layer consists of spatial object nodes and procedure clause nodes that are associated with the scheme node; the second layer consists of construction method nodes and permit items nodes that are associated with the scheme node; and the third layer consists of risk event nodes and cost element nodes that are associated with the scheme node. Within each layer, nodes are sorted from smallest to largest based on the clearance margin in the edge geometry features. Nodes with the same clearance margin are sorted from smallest to largest based on the distance between them, thus obtaining the input sequence. The Graph-Mamba backbone inference module is composed of stacked Mamba inference layers of a preset number. Each Mamba inference layer sequentially performs input projection, selective state update, and output projection on the input sequence. The selective state update uses the edge gating weight in the gated edge representation as the multiplicative modulation coefficient for the state update. At the end of each Mamba inference layer, the output and input are summed and layer normalized to obtain the updated sequence representation. The updated sequence representation is then aggregated with the sequence position corresponding to the scheme node as the index to obtain the scheme embedding. The multi-task output head module includes a risk indicator output head, a cost indicator output head, a schedule indicator output head, and an evidence importance output head. The risk indicator output head, cost indicator output head, and schedule indicator output head perform two-layer fully connected mapping on the scheme embedding and obtain risk indicator prediction, cost indicator prediction, and schedule indicator prediction through the ReLU function, respectively. The evidence importance output head calculates the similarity between the updated sequence representation and the scheme embedding and normalizes it through the Softmax function to obtain the importance weight. The importance weight is then backfilled into the corresponding nodes and edges in the scheme subgraph to obtain the evidence importance result.

[0011] Optionally, step six specifically includes: Hard constraint verification is performed on candidate 3D paths based on the rule layer; For each hard constraint clause, a set of verification elements is generated for the verification object. The set of verification elements consists of spatial object nodes that establish spatial association edges with the candidate 3D path, device entity nodes that establish spatial association edges with the candidate 3D path, and path geometric descriptions and path segment sequence descriptions of the candidate 3D path. Based on the path geometry description of the candidate 3D path and the spatial geometry description of the spatial object node, spatial intersection and spatial inclusion are performed. If the judgment result satisfies the hard constraint clause description corresponding to the prohibited and restricted land use in the rule layer, the candidate 3D path is marked as unsuccessful. The minimum distance is calculated based on the path geometry description of the candidate 3D path and the spatial geometry description of the spatial object node. The minimum distance is then compared with the preset backoff threshold of the hard constraint clause node. If the minimum distance is less than the preset backoff threshold, the candidate 3D path is marked as unacceptable. The path geometry description of the candidate 3D path is sampled at equal intervals along the path segment sequence description to obtain a set of sampling points. Based on the set of sampling points, the corresponding ground elevation and obstacle height are queried in the 3D planning base. The clearance margin at the sampling point is calculated and compared with the preset clearance threshold of the hard constraint clause node. If there is a sampling point where the clearance margin is less than the preset clearance threshold, the candidate 3D path is marked as unpassable. Output hard constraint verification marks for each candidate 3D path and summarize them to form a feasible candidate set.

[0012] Optionally, step seven specifically includes: For each candidate three-dimensional path within the feasible candidate set, a multi-objective evaluation record is established. The multi-objective evaluation record includes the cumulative cost value, risk index prediction value, cost index prediction value, construction period index prediction value, and evidence importance result corresponding to the cost field related to the laying method. The cumulative cost value is obtained by summing the path segments of the candidate three-dimensional path segment by segment in the cost field related to the laying method. A summary value of evidence importance is generated for each candidate 3D path based on the evidence importance results; A multi-objective screening vector is constructed using the cumulative cost, risk index prediction, cost index prediction, schedule index prediction, and evidence importance summary. The Pareto dominance relationship of candidate three-dimensional paths within the feasible candidate set is determined, and candidate three-dimensional paths that are not Pareto-dominated by other candidate three-dimensional paths are retained to form the Pareto minimum feasible solution set. Output a 3D route planning result package based on the Pareto minimum feasible solution set.

[0013] The beneficial effects of this invention are: This invention integrates multi-source spatial, equipment, ownership, and risk information under a unified coordinate and elevation datum by using a 3D planning base and a 3D accessible corridor. It also maps the differentiated costs of overhead, cable, and mixed laying methods into a laying method-related cost field, ensuring consistent introduction of laying differences and engineering constraints during the candidate path generation stage. Furthermore, it constructs a distribution network planning knowledge graph containing rule and exception layers, establishing traceable associations between candidate paths as scheme nodes and regulatory clauses, construction methods, permits, and risk and cost elements. Finally, it uses scheme subgraphs as input to introduce 3D geometric-topological joint location coding, rule-exception evidence gating, and scheme-oriented adaptive features. The serialized improved Graph-Mamba model outputs risk, cost, schedule predictions, and evidence importance, enabling stable reasoning and evidence localization for multi-hop constraint chains. Based on this, a feasible candidate set is obtained through hard constraint verification of rule layers and three-dimensional geometric constraints. Pareto multi-objective screening is then performed within the feasible set to output the minimum feasible solution set and result package. Compared with existing two-dimensional overlay and empirical weighting methods, this effectively reduces corridor misjudgments caused by data inconsistency, improves the consistency of the cost of different laying methods in generation and screening, and enhances the interpretability and verifiability of scheme review. It is of great importance in supporting the three-dimensional refined planning, rapid review, and engineering implementation of power distribution networks. Attached Figure Description

[0014] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings: Figure 1 This is a flowchart of a knowledge graph-based intelligent planning method for three-dimensional power distribution network routes proposed in this invention. Figure 2 This is a schematic diagram of a knowledge graph-based intelligent planning method for three-dimensional power distribution network routes proposed in this invention. Figure 3 This is a framework diagram of the improved Graph-Mamba model in the intelligent planning method for three-dimensional power distribution network routes based on knowledge graphs proposed in this invention. Detailed Implementation

[0015] The present invention will now be described in further detail with reference to the accompanying drawings. These drawings are simplified schematic diagrams, illustrating only the basic structure of the invention, and therefore only show the components relevant to the invention.

[0016] refer to Figure 1-3 A knowledge graph-based intelligent planning method for three-dimensional power distribution network routes includes the following steps: Step 1: Obtain relevant data for the planning area and unify coordinates and elevation benchmarks; assign unique identifiers to spatial objects and equipment objects to form a three-dimensional planning base. Step 2: Generate a 3D passable corridor based on the 3D planning base, construct a 3D passable cost field, and establish differentiated cost mappings for overhead, cable and mixed laying methods. Load the differentiated cost mappings into the 3D passable cost field to form a laying method-related cost field. Step 3: Construct a knowledge graph for power distribution network planning, including a rule layer and an exception layer; Step 4: Generate no less than a preset number of candidate 3D paths under the cost field constraints related to the 3D passable corridor and the laying method, and determine the corresponding laying method for each candidate 3D path. Write each candidate 3D path into the power distribution network planning knowledge graph to form a scheme node and establish a relationship. Step 5: Extract a subgraph of the scheme with a preset number of hops centered on each scheme node, and input the scheme subgraph into the improved Graph-Mamba model for inference. The improved Graph-Mamba model includes a three-dimensional geometry-topology joint location encoding module, a rule-exception evidence gating module, a scheme-oriented adaptive serialization module, a Graph-Mamba backbone inference module, and a multi-task output head module, which outputs scheme embeddings and risk indicator predictions, cost indicator predictions, schedule indicator predictions, and evidence importance results. Step 6: Perform hard constraint verification on the candidate 3D paths based on the rule layer and 3D geometric constraints to obtain a set of feasible candidates; Step 7: Within the feasible candidate set, perform multi-objective screening based on the cumulative cost corresponding to the cost field related to the laying method, as well as the prediction of risk indicators, cost indicators, schedule indicators, and the importance of evidence, and output the Pareto minimum feasible solution set and the three-dimensional route planning result package.

[0017] In this embodiment, step one specifically includes: Acquire relevant data for the planning area, including three-dimensional spatial data, channel and ownership data, power distribution network equipment object data, and risk zoning data. The three-dimensional spatial data includes one or both of laser point cloud data and oblique photogrammetry three-dimensional mesh data. The laser point cloud data is a set of points containing three-dimensional coordinates and echo intensity with acquisition time stamps. The oblique photogrammetry three-dimensional mesh data is a mesh model data containing vertex coordinates and triangular facet topology, reconstructed from multi-view images. The three-dimensional spatial data further includes digital elevation data or digital surface data and corresponding geographic coordinates. Reference information and elevation benchmark information; the channel and ownership data include channel centerline or channel boundary, channel width, channel type, land use or easement scope, ownership unit identifier, ownership boundary, and restricted land use identifier; the distribution network equipment object data includes the spatial location or spatial range of poles, conductors, transformers, switching stations, ring main units, cable wells, and cable channels, equipment type identifier, voltage level identifier, commissioning or ledger date identifier, and operation and maintenance status identifier; the risk zoning data includes the spatial range and risk level identifiers of geological disaster risk zones, flood risk zones, icing risk zones, and lightning strike risk zones. The coordinate system and elevation datum of the three-dimensional spatial data are unified. The spatial registration method with control points or common feature points as constraints is used to align the three-dimensional spatial data from different sources to the same coordinate system. For data with time differences, the nearest neighbor matching is performed according to the acquisition time mark. The channel and ownership data, distribution network equipment object data and risk zoning data are projected to the unified coordinate system and a unique identifier is established for each. The unique identifier is generated by combining the object category identifier, data source identifier and regional grid index. Based on the unified 3D spatial data, a 3D raster index for the planning area is constructed. The density, occlusion ratio and noise level of each raster are statistically analyzed to form a spatial quality field. The spatial quality field, along with the geometric information, attribute information and unique identifier of the corresponding object, are written into the same data structure to obtain the 3D planning base.

[0018] In this embodiment, step two specifically includes: A three-dimensional passable corridor is generated based on a three-dimensional planning base. The three-dimensional passable corridor is composed of grid units in the three-dimensional grid index of the planning area that meet the passability conditions. The passability conditions include the spatial range corresponding to the avoidance of restricted land use signs, the volume range of the avoidance of building entities, the volume range of the avoidance of water bodies, and meeting the preset clearance conditions and preset setback conditions. For different laying methods, separate access cost items are constructed. The access cost items for overhead laying include crossing cost, corner cost, pole access cost, and clearance margin cost. The access cost items for cable laying include road excavation cost, trench or pipe jacking cost, crossing cost, and access cost to existing cable channels. The access cost items for mixed laying include overhead section cost, cable section cost, and overhead section to cable section conversion cost. Each passage cost item is normalized according to a unified scale to form a basic three-dimensional passage cost field. Establish a differentiated cost mapping for overhead, cable and mixed laying methods. The differentiated cost mapping includes setting laying method-related weights for each access cost item and setting laying method conversion costs. Load the laying method-related weights and laying method conversion costs into the basic three-dimensional access cost field to obtain the laying method-related cost field.

[0019] In this embodiment, step three specifically includes: A distribution network planning knowledge graph is constructed. The distribution network planning knowledge graph is represented by a directed attribute graph and consists of a set of nodes and a set of edges. The set of nodes includes equipment entity nodes, spatial object nodes, code clause nodes, exception condition nodes, construction method nodes, permit items nodes, risk event nodes, and cost element nodes. Attribute fields are set for each node. The attribute fields for equipment entity nodes and spatial object nodes include unique identifier, spatial geometry description, object category identifier, and status identifier. The attribute fields for procedure clause nodes include clause number identifier, scope of application description, and constraint type identifier. The attribute fields for exception condition nodes include trigger condition description, additional measure description, additional cost field, and scope of application description. The attribute fields for construction method nodes include construction method category identifier and process constraint description. The attribute fields for permit item nodes include permit category identifier and processing constraint description. The attribute fields for risk event nodes include risk category identifier, risk level identifier, and spatial scope description. The attribute fields for cost element nodes include cost category identifier and measurement field. Construct an edge set and set edge type and edge attribute fields. The edge type includes spatial association edge, electrical association edge, code constraint association edge, exception association edge, construction constraint association edge, risk impact association edge and cost impact association edge. The edge attribute fields include relationship direction identifier, applicable condition identifier and weight field. The distribution network planning knowledge graph is divided into a rule layer and an exception layer. The rule layer is formed by connecting the procedure clause nodes with the equipment entity nodes and spatial object nodes through the procedure constraint association edges, and the procedure clause nodes are marked with hard constraint clause labels or soft constraint clause labels. The exception layer is formed by connecting the exception condition nodes with the procedure clause nodes through the exception association edges, and the exception condition nodes are connected to the construction method nodes, permit item nodes and cost element nodes through the construction constraint association edges, permit item association edges and cost impact association edges, respectively.

[0020] In this embodiment, step four specifically includes: Within the three-dimensional passable corridor, the starting point object and the ending point object are determined. The starting point object is the equipment entity node corresponding to the power supply point, and the ending point object is the equipment entity node corresponding to the load access point. Candidate generation channels for overhead laying method, cable laying method and mixed laying method are determined in the cost field related to the laying method. Candidate 3D paths are generated under the constraints of the cost field related to the 3D passable corridor and the laying method. The generation methods include priority search along the road corridor, priority search to avoid high-risk zones, priority search with the fewest crossings, and priority search with the fewest turns. A preset number of candidate 3D paths are output for each generation method. For each candidate 3D path, the corresponding laying method is determined by dividing the candidate 3D path into overhead section and cable section according to the path segment. Based on the cumulative cost of the path segment in the laying method related cost field, the overhead laying method, cable laying method or mixed laying method is selected as the laying method identifier of the candidate 3D path. Each candidate 3D path is written into the power distribution network planning knowledge graph to form a scheme node. The attribute fields of the scheme node include path geometry description, laying method identifier and path segment sequence description. Relationships are established for the scheme nodes. The relationships include: the scheme node is connected to intersecting or crossing spatial object nodes via spatial association edges; connected to applicable regulation clause nodes via regulation constraint association edges; connected to corresponding construction method nodes via construction constraint association edges; connected to corresponding permission item nodes via permission item association edges; connected to covered risk event nodes via risk impact association edges; and connected to corresponding cost element nodes via cost impact association edges.

[0021] In this embodiment, the improved Graph-Mamba model includes a three-dimensional geometry-topology joint location encoding module, a rule-exception evidence gating module, a scheme-oriented adaptive serialization module, a Graph-Mamba backbone inference module, and a multi-task output header module; The three-dimensional geometry-topology joint location encoding module generates node three-dimensional location features based on the spatial geometric description of equipment entity nodes and spatial object nodes in the scheme subgraph, and generates edge geometric features based on the edge attribute fields of spatially associated edges in the scheme subgraph. The edge geometric features are composed of distance between nodes, azimuth angle, relative height difference, crossing type, clearance margin and slope information. Two-layer linear mapping is performed on the node three-dimensional location features and edge geometric features respectively to obtain node location embedding and edge location embedding. The node location embedding and the corresponding node attribute field encoding vector are concatenated to obtain the node input representation. The rule-exceptional evidence gating module constructs a gating input vector based on the hard constraint clause identifier and soft constraint clause identifier of the procedure clause node in the scheme subgraph, the trigger condition description and additional cost field of the exception condition node, and the laying method identifier of the scheme node. It performs a linear mapping on the gating input vector and obtains the node gating weight and edge gating weight through the Sigmoid function. It multiplies the node gating weight and the node input representation element by element to obtain the gated node representation, and writes the edge gating weight into the edge position embedding to form the gated edge representation. The adaptive serialization module for the scheme generates an input sequence centered on the scheme node. The input sequence is structured in layers: the first layer consists of spatial object nodes and procedure clause nodes that are associated with the scheme node; the second layer consists of construction method nodes and permit items nodes that are associated with the scheme node; and the third layer consists of risk event nodes and cost element nodes that are associated with the scheme node. Within each layer, nodes are sorted from smallest to largest based on the clearance margin in the edge geometry features. Nodes with the same clearance margin are sorted from smallest to largest based on the distance between them, thus obtaining the input sequence. The Graph-Mamba backbone inference module is composed of stacked Mamba inference layers of a preset number. Each Mamba inference layer sequentially performs input projection, selective state update, and output projection on the input sequence. The selective state update uses the edge gating weight in the gated edge representation as the multiplicative modulation coefficient for the state update. At the end of each Mamba inference layer, the output and input are summed and layer normalized to obtain the updated sequence representation. The updated sequence representation is then aggregated with the sequence position corresponding to the scheme node as the index to obtain the scheme embedding. The multi-task output head module includes a risk indicator output head, a cost indicator output head, a schedule indicator output head, and an evidence importance output head. The risk indicator output head, cost indicator output head, and schedule indicator output head perform two-layer fully connected mapping on the scheme embedding and obtain risk indicator prediction, cost indicator prediction, and schedule indicator prediction through the ReLU function, respectively. The evidence importance output head calculates the similarity between the updated sequence representation and the scheme embedding and normalizes it through the Softmax function to obtain the importance weight. The importance weight is then backfilled into the corresponding nodes and edges in the scheme subgraph to obtain the evidence importance result.

[0022] This implementation introduces an improved Graph-Mamba model to replace traditional graph convolutional networks and global attention-based graph Transformers in the subgraph reasoning of the scheme, achieving efficient modeling of long-chain, multi-hop constraint relationships in the knowledge graph of power distribution planning. The reason for choosing this model is that the computational cost of its state space backbone increases linearly with the node size in long sequence reasoning, making it suitable for power distribution planning scenarios with a large number of candidate schemes and requiring rapid iterative review. Compared with the full attention mechanism, it is less prone to significant increases in latency and memory usage due to increased scale. Unlike the original model, which mainly relies on topological adjacency diffusion and is prone to attenuation of far-distance constraint information and oversmoothing, this invention directly injects three-dimensional constraints such as clearance margin, crossing type, and relative elevation difference into the representation through three-dimensional geometric-topological joint position encoding. It modulates the weights of hard constraint clauses and exception triggering conditions through rule-exception evidence gating, and improves the stability and consistency of risk, cost, and schedule predictions by using adaptive serialization oriented towards the scheme. The importance of evidence is more focused on key crossing segments and clause nodes, reducing review rework and improving the feasibility of Pareto screening output schemes.

[0023] In this embodiment, step six specifically includes: Hard constraint verification is performed on candidate 3D paths based on the rule layer. The hard constraint verification is performed one by one according to the procedure clause nodes marked as hard constraint clauses in the procedure clause nodes, and the procedure constraint association edge between the scheme node corresponding to the candidate 3D path and the procedure clause node is used as the verification entry point. For each hard constraint clause, a set of verification elements is generated for the verification object. The set of verification elements consists of spatial object nodes that establish spatial association edges with the candidate 3D path, device entity nodes that establish spatial association edges with the candidate 3D path, and path geometric descriptions and path segment sequence descriptions of the candidate 3D path. Based on the path geometry description of the candidate 3D path and the spatial geometry description of the spatial object node, spatial intersection and spatial inclusion are performed. If the judgment result satisfies the hard constraint clause description corresponding to the prohibited and restricted land use in the rule layer, the candidate 3D path is marked as unsuccessful. The minimum distance is calculated based on the path geometry description of the candidate 3D path and the spatial geometry description of the spatial object node. The minimum distance is then compared with the preset backoff threshold of the hard constraint clause node. If the minimum distance is less than the preset backoff threshold, the candidate 3D path is marked as unacceptable. The path geometry description of the candidate 3D path is sampled at equal intervals along the path segment sequence description to obtain a set of sampling points. Based on the set of sampling points, the corresponding ground elevation and obstacle height are queried in the 3D planning base. The clearance margin at the sampling point is calculated and compared with the preset clearance threshold of the hard constraint clause node. If there is a sampling point where the clearance margin is less than the preset clearance threshold, the candidate 3D path is marked as unpassable. Hard constraint verification marks are output for each candidate 3D path and summarized to form a feasible candidate set. The feasible candidate set consists of the scheme nodes corresponding to the candidate 3D paths whose hard constraint verification marks are passed.

[0024] In this embodiment, step seven specifically includes: For each candidate three-dimensional path within the feasible candidate set, a multi-objective evaluation record is established. The multi-objective evaluation record includes the cumulative cost value, risk index prediction value, cost index prediction value, construction period index prediction value, and evidence importance result corresponding to the cost field related to the laying method. The cumulative cost value is obtained by summing the path segment sequence of the candidate three-dimensional path in the cost field related to the laying method segment by segment. The segment-by-segment summation method is to sample each path segment at equal intervals along the path segment direction to obtain a set of sampling points, read the cost value of the corresponding cost field related to the laying method for each sampling point and sum them to obtain the path segment cost value, and then sum the cost values ​​of all path segments to obtain the cumulative cost value. Based on the evidence importance results, a summary value of evidence importance is generated for each candidate three-dimensional path. The summary value of evidence importance is obtained by summing the importance weights of the procedure clause nodes, exception condition nodes, risk event nodes, and cost element nodes that are associated with the candidate three-dimensional path in the corresponding scheme subgraph. A multi-objective screening vector is constructed using the cumulative cost, risk index prediction, cost index prediction, schedule index prediction, and evidence importance summary. The Pareto dominance relationship of candidate three-dimensional paths within the feasible candidate set is determined, and candidate three-dimensional paths that are not Pareto-dominated by other candidate three-dimensional paths are retained to form the Pareto minimum feasible solution set. The three-dimensional route planning result package is output based on the Pareto minimum feasible solution set. The three-dimensional route planning result package includes the route geometry description, laying method identifier, route segment sequence description, cumulative cost value, risk index prediction value, cost index prediction value, construction period index prediction value, evidence importance result, and a set of identifiers for the procedure clause nodes, exception condition nodes, construction method nodes and permission item nodes that are associated with the candidate three-dimensional route.

[0025] Example 1: To verify the feasibility of this invention in practice, it was applied to the planning scenario of a new 10kV distribution network interconnection line in a district of a coastal city. This area has a high density of old residential communities and narrow roads. A river approximately 38 meters wide flows along its north side, flanked by green belts and walkways. The area contains existing underground utility tunnels and various pipelines such as gas and water supply / drainage lines, and construction access on the main roads is restricted by traffic regulations. The project objective is to generate a feasible three-dimensional line path scheme between the power supply point and the corresponding ring main unit at the load access point, outputting a result package suitable for review and construction handover, and maintaining a stable multi-objective balance across risk, cost, and schedule indicators.

[0026] During implementation, the three-dimensional spatial data of the planning area, channel and ownership data, power distribution equipment ledgers, and risk zoning data are first collected and unified to the same coordinate and elevation datum to form a three-dimensional planning base. The base is organized using a 1m×1m×1m three-dimensional raster index. Rasteres with a point cloud density of less than 15 points / cubic meter are marked as low-quality rasters, and rasters with an occlusion ratio higher than 0.35 are marked as high-occlusion rasters. The noise level is mapped from the echo intensity dispersion to a spatial quality field of 0 to 1. This field is used as an uncertainty weighting factor in the subsequent cost field, so that uncertain areas in the mapping are automatically tended to be avoided or downweighted during path search. A three-dimensional passable corridor is generated on this base. In this embodiment, the passability threshold is fixed as follows: no passability is allowed in restricted land areas; the minimum setback distance from the elevated section to the boundary of the building facade is not less than 1.5m; the clearance for elevated sections crossing roads is not less than 5.0m; the clearance for elevated sections crossing rivers is not less than 7.0m; the burial depth of cable sections is controlled between 0.8m and 1.2m, and the minimum horizontal distance from gas pipelines is not less than 1.0m and the minimum horizontal distance from main water supply and drainage lines is not less than 0.6m; when the spatial quality field is lower than 0.4, the passability cost for the corresponding area is increased by 20%.

[0027] To ensure consistency in the candidate generation and screening stages for different laying methods, this embodiment maps and loads differentiated costs onto a three-dimensional access cost field to form a laying method-related cost field. For overhead laying, the crossing cost is set at 1.0 for roads, 2.5 for rivers, and 1.8 for high-density residential areas; the corner cost is accumulated based on the absolute angle of each corner / 45°; the pole reachability cost is 0.6 within 3m of the road centerline on both sides, and 1.0 beyond that range; the clearance margin cost is accumulated based on the non-negative difference between the threshold clearance and the actual clearance. For cable laying, the road excavation cost is calculated by length: 1.4 for road widths less than 6m, and 1.0 for widths between 6m and 12m; the cost of pipe jacking across main roads is 3.0, and 4.0 for crossing rivers; the cost of accessing existing cable channels is mapped by the remaining rate: 2.0 if the remaining rate is below 0.2, otherwise 1.0. For mixed laying, the costs for overhead and cable sections are accumulated separately, and a conversion cost of 2.2 is added for each "overhead-to-cable" conversion.

[0028] Subsequently, a distribution network planning knowledge graph containing rule and exception layers is constructed, structurally associating regulation clauses, exception conditions, construction methods, permits, risk events, cost elements, and spatial objects and equipment entities. The exception layer focuses on two types of conditions in this area: one is the triggering condition for road occupancy and nighttime construction permits, bound to permit nodes and additional construction period fields; the other is the triggering condition for "pipe jacking instead of excavation," bound to construction method nodes and additional cost fields. This ensures that candidate schemes can simultaneously generate an executable constraint list and cost changes when exceptions are triggered. Candidate 3D paths are generated under the constraints of the corridor body and the cost field related to the laying method. This embodiment presets 30 candidate paths, covering four strategies: priority along road corridors, priority to avoid high-risk zones, priority to minimum crossings, and priority to minimum turns. Each candidate path determines its laying method and writes it into the knowledge graph to form a scheme node, automatically establishing associations with crossing objects, clauses, exceptions, permits, construction methods, risks, and cost elements.

[0029] In the reasoning and evaluation phase, a subgraph of schemes with a preset hop count of 2 is extracted centered on the scheme nodes and input into an improved Graph-Mamba model for reasoning. The model parameters were set experimentally: embedding dimension 128, Mamba reasoning layers 6; the input sequence was organized into three layers: the first layer contained spatial object nodes and regulation clause nodes associated with the scheme nodes; the second layer contained construction method nodes and permit nodes; and the third layer contained risk event nodes and cost element nodes. Within each layer, the order was based on net clearance from smallest to largest, and for nodes with the same net clearance, the order was based on distance between nodes from smallest to largest. The model outputs risk indicator predictions, cost indicator predictions, and schedule indicator predictions, as well as evidence importance results to locate the clause nodes and crossing nodes most sensitive to the review. Subsequently, hard constraint checks were performed independently. Net clearance checks involved sampling along the path at 5m intervals and querying ground elevation and obstacle heights to calculate net clearance. Setback checks calculated the minimum distance between the path and the boundaries of building and pipeline voxels. Restricted land use checks performed spatial intersection and inclusion determination. Candidates that failed the hard constraints were eliminated, resulting in a feasible candidate set. Within the feasible candidate set, a multi-objective screening vector is constructed using the cumulative cost, risk prediction, cost prediction, schedule prediction, and evidence importance summary value. Pareto dominance is then determined, and the Pareto minimum feasible solution set and result package are output.

[0030] To demonstrate the comparative effect, two existing solutions were selected as baselines. Method A used a two-dimensional overlay rule filter for shortest path search and estimated cost and duration with fixed empirical weights. Method B used a three-dimensional cost field + A* to generate candidates and performed hard constraint checks, but did not construct a knowledge graph or subgraph reasoning for the solutions. Risk and duration were estimated using linear regression and no evidence chain was output. All three methods completed the planning of six start-endpoint tasks within the same project area, and indicators such as candidate selection, approval, rework, review time, and error were statistically analyzed.

[0031] Table 1. Comparison of the average values ​​of the six tasks

[0032] As can be seen from the comparison of the mean values ​​in Table 1, the present invention significantly improves the number of hard constraints passed, averaging 9.8 passes, an increase of 7.1 passes compared to 2.7 passes in Comparison A, and an increase of 4.5 passes compared to 5.3 passes in Comparison B. This indicates that the candidate solutions generated under the joint constraints of the cost field related to the corridor structure and the laying method are closer to the feasible space, reducing a large number of invalid solutions that are "removed by hard constraints immediately after generation". In terms of review time, the present invention averaged 3.53 hours, which is about 3.40 hours shorter than Comparison A and about 1.52 hours shorter than Comparison B. Combined with the implementation process records, the main difference comes from the fact that the evidence importance results can focus the review attention on the crossing section with the smallest clearance margin, the building boundary section with the tightest setback, and the section with the highest risk level. The review communication has shifted from checking each section of the drawings to checking around the key evidence chain, and the number of rework times has also decreased from 2.3 times to 0.7 times. Regarding cost and schedule errors, the average values ​​for this invention were 1.32% and 1 day, respectively, significantly lower than those of Comparison A and Comparison B. This is because the differentiated cost mapping is loaded into the cost field and runs through candidate generation and Pareto screening, ensuring consistency between the selected solution and the cost structure. Furthermore, the exception layer of the knowledge graph explicitly binds permitted items with additional costs, making the schedule prediction closer to the approval and organizational rhythm. The average risk score decreased from 59.83 to 45.00, demonstrating that the model's stable inference over multi-hop constraint chains can retain lower-risk non-dominated solutions within the feasible set.

[0033] Table 2 Comparison of Example Tasks in River Crossing Scenarios

[0034] In river-crossing tasks, this invention employs 11 hard constraints, enabling reviewers to perform Pareto comparisons on various feasible solutions, such as "overhead crossing of the river," "cable jacking across the river," and "hybrid laying with transition points avoiding sensitive areas of the walkway," within the feasible set. In contrast, comparing A and B results in a narrower feasible solution space, increasing the likelihood of "rejection of the only solution leading to rework." Cost error decreased from 11.5% to 1.8%, and time error decreased from 6 days to 1 day, demonstrating the significant role of differentiated cost mapping and exception-level permission binding in complex river-crossing scenarios. The reduced risk score is also related to the prioritization of evidence; during review, the focus can be directly placed on sensitive areas on both sides of the river and the clearance margin of the crossing section, reducing overlooked potential risks.

[0035] Based on the above embodiments, the beneficial effects of the present invention are mainly reflected in: using the cost field related to the three-dimensional corridor volume and the laying method to uniformly constrain candidate generation and screening, significantly improving the pass rate of hard constraints and reducing rework; using rule-layer hard constraint verification to ensure compliance bottom line; using knowledge graph to structure and associate clauses, exceptions, permissions, construction methods, risks and cost elements, so that the solution has a traceable evidence chain; using improved Graph-Mamba to perform long-range dependency reasoning on the solution subgraph and output the importance of evidence, so as to achieve stable Pareto screening of multiple objectives of risk, cost and schedule within the feasible set, and finally outputting a result package that can be directly used for review, approval and construction handover, thereby improving the implementation efficiency and engineering consistency of the three-dimensional distribution network line planning.

[0036] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A knowledge graph-based three-dimensional line path intelligent planning method for distribution network, characterized in that, Includes the following steps: Step 1: Obtain relevant data for the planning area and unify coordinates and elevation benchmarks; assign unique identifiers to spatial objects and equipment objects to form a three-dimensional planning base. Step 2: Generate a 3D passable corridor based on the 3D planning base, construct a 3D passable cost field, and establish differentiated cost mappings for overhead, cable and mixed laying methods. Load the differentiated cost mappings into the 3D passable cost field to form a laying method-related cost field. Step 3: Construct a knowledge graph for power distribution network planning, including a rule layer and an exception layer; Step 4: Generate no less than a preset number of candidate 3D paths under the cost field constraints related to the 3D passable corridor and the laying method, and determine the corresponding laying method for each candidate 3D path. Write each candidate 3D path into the power distribution network planning knowledge graph to form a scheme node and establish a relationship. Step 5: Extract a subgraph of the scheme with a preset number of hops centered on each scheme node, and input the scheme subgraph into the improved Graph-Mamba model for inference. The improved Graph-Mamba model includes a three-dimensional geometry-topology joint location encoding module, a rule-exception evidence gating module, a scheme-oriented adaptive serialization module, a Graph-Mamba backbone inference module, and a multi-task output head module, which outputs scheme embeddings and risk indicator predictions, cost indicator predictions, schedule indicator predictions, and evidence importance results. Step 6: Perform hard constraint verification on the candidate 3D paths based on the rule layer and 3D geometric constraints to obtain a set of feasible candidates; Step 7: Within the feasible candidate set, perform multi-objective screening based on the cumulative cost corresponding to the cost field related to the laying method, as well as the prediction of risk indicators, cost indicators, schedule indicators, and the importance of evidence, and output the Pareto minimum feasible solution set and the three-dimensional route planning result package. 2.The knowledge graph-based three-dimensional line path intelligent planning method for distribution network according to claim 1, characterized in that, Step one specifically includes: Acquire relevant data for the planning area, including three-dimensional spatial data, channel and ownership data, power distribution equipment object data, and risk zoning data; The coordinate system and elevation datum of the three-dimensional spatial data are unified. The channel and ownership data, distribution network equipment object data and risk zoning data are projected onto the unified coordinate system and a unique identifier is established for each. Based on the unified 3D spatial data, a 3D raster index for the planning area is constructed. The density, occlusion ratio and noise level of each raster are statistically analyzed to form a spatial quality field. The spatial quality field, along with the geometric information, attribute information and unique identifier of the corresponding object, are written into the same data structure to obtain the 3D planning base.

3. The intelligent planning method for three-dimensional power distribution network routes based on knowledge graphs according to claim 1, characterized in that, Step two specifically includes: A three-dimensional passable corridor is generated based on a three-dimensional planning base. The three-dimensional passable corridor is composed of grid units in the three-dimensional grid index of the planning area that meet the passability conditions. The passability conditions include the spatial range corresponding to the avoidance of restricted land use signs, the volume range of the avoidance of building entities, the volume range of the avoidance of water bodies, and meeting the preset clearance conditions and preset setback conditions. For different laying methods, separate access cost items are constructed. The access cost items for overhead laying include crossing cost, corner cost, pole access cost, and clearance margin cost. The access cost items for cable laying include road excavation cost, trench or pipe jacking cost, crossing cost, and access cost to existing cable channels. The access cost items for mixed laying include overhead section cost, cable section cost, and overhead section to cable section conversion cost. Each passage cost item is normalized according to a unified scale to form a basic three-dimensional passage cost field. Establish a differentiated cost mapping for overhead, cable and mixed laying methods. The differentiated cost mapping includes setting laying method-related weights for each access cost item and setting laying method conversion costs. Load the laying method-related weights and laying method conversion costs into the basic three-dimensional access cost field to obtain the laying method-related cost field.

4. The intelligent planning method for three-dimensional power distribution network routes based on knowledge graphs according to claim 1, characterized in that, Step three specifically includes: A distribution network planning knowledge graph is constructed, which is represented by a directed attribute graph and consists of a set of nodes and a set of edges. Set attribute fields for each node, construct an edge set and set the edge type and edge attribute fields; The distribution network planning knowledge graph is divided into a rule layer and an exception layer. The rule layer is formed by connecting the procedure clause nodes with the equipment entity nodes and spatial object nodes through the procedure constraint association edges, and the procedure clause nodes are marked with hard constraint clause labels or soft constraint clause labels. The exception layer is formed by connecting the exception condition nodes with the procedure clause nodes through the exception association edges, and the exception condition nodes are connected to the construction method nodes, permit item nodes and cost element nodes through the construction constraint association edges, permit item association edges and cost impact association edges, respectively.

5. The intelligent planning method for three-dimensional power distribution network routes based on knowledge graphs according to claim 1, characterized in that, Step four specifically includes: Within the three-dimensional passable corridor, the starting point object and the ending point object are determined. The starting point object is the equipment entity node corresponding to the power supply point, and the ending point object is the equipment entity node corresponding to the load access point. Candidate generation channels for overhead laying method, cable laying method and mixed laying method are determined in the cost field related to the laying method. Candidate 3D paths are generated under the constraints of the cost field related to the 3D passable corridor and the laying method. The generation methods include priority search along the road corridor, priority search to avoid high-risk zones, priority search with the fewest crossings, and priority search with the fewest turns. A preset number of candidate 3D paths are output for each generation method. For each candidate 3D path, the corresponding laying method is determined by dividing the candidate 3D path into overhead section and cable section according to the path segment. Based on the cumulative cost of the path segment in the laying method related cost field, the overhead laying method, cable laying method or mixed laying method is selected as the laying method identifier of the candidate 3D path. Each candidate 3D path is written into the distribution network planning knowledge graph to form a scheme node. The attribute fields of the scheme node include path geometry description, laying method identifier and path segment sequence description, and an association relationship is established for the scheme node.

6. The intelligent planning method for three-dimensional power distribution network routes based on knowledge graphs according to claim 1, characterized in that, The improved Graph-Mamba model includes a 3D geometry-topology joint location encoding module, a rule-exception evidence gating module, a scheme-oriented adaptive serialization module, a Graph-Mamba backbone inference module, and a multi-task output header module; The three-dimensional geometry-topology joint location encoding module generates node three-dimensional location features based on the spatial geometric description of device entity nodes and spatial object nodes in the scheme subgraph, and generates edge geometric features based on the edge attribute fields of spatially associated edges in the scheme subgraph. It performs two-layer linear mapping on the node three-dimensional location features and edge geometric features respectively to obtain node location embedding and edge location embedding. Finally, it concatenates the node location embedding with the corresponding node attribute field encoding vector to obtain the node input representation. The rule-exceptional evidence gating module constructs a gating input vector based on the hard constraint clause identifier and soft constraint clause identifier of the procedure clause node in the scheme subgraph, the trigger condition description and additional cost field of the exception condition node, and the laying method identifier of the scheme node. It performs a linear mapping on the gating input vector and obtains the node gating weight and edge gating weight through the Sigmoid function. It multiplies the node gating weight and the node input representation element by element to obtain the gated node representation, and writes the edge gating weight into the edge position embedding to form the gated edge representation. The adaptive serialization module for the scheme generates an input sequence centered on the scheme node. The input sequence is structured in layers: the first layer consists of spatial object nodes and procedure clause nodes that are associated with the scheme node; the second layer consists of construction method nodes and permit items nodes that are associated with the scheme node; and the third layer consists of risk event nodes and cost element nodes that are associated with the scheme node. Within each layer, nodes are sorted from smallest to largest based on the clearance margin in the edge geometry features. Nodes with the same clearance margin are sorted from smallest to largest based on the distance between them, thus obtaining the input sequence. The Graph-Mamba backbone inference module is composed of stacked Mamba inference layers of a preset number. Each Mamba inference layer sequentially performs input projection, selective state update, and output projection on the input sequence. The selective state update uses the edge gating weight in the gated edge representation as the multiplicative modulation coefficient for the state update. At the end of each Mamba inference layer, the output and input are summed and layer normalized to obtain the updated sequence representation. The updated sequence representation is then aggregated with the sequence position corresponding to the scheme node as the index to obtain the scheme embedding. The multi-task output head module includes a risk indicator output head, a cost indicator output head, a schedule indicator output head, and an evidence importance output head. The risk indicator output head, cost indicator output head, and schedule indicator output head perform two-layer fully connected mapping on the scheme embedding and obtain risk indicator prediction, cost indicator prediction, and schedule indicator prediction through the ReLU function, respectively. The evidence importance output head calculates the similarity between the updated sequence representation and the scheme embedding and normalizes it through the Softmax function to obtain the importance weight. The importance weight is then backfilled into the corresponding nodes and edges in the scheme subgraph to obtain the evidence importance result.

7. The intelligent planning method for three-dimensional power distribution network routes based on knowledge graphs according to claim 1, characterized in that, Step six specifically includes: Hard constraint verification is performed on candidate 3D paths based on the rule layer; For each hard constraint clause, a set of verification elements is generated for the verification object. The set of verification elements consists of spatial object nodes that establish spatial association edges with the candidate 3D path, device entity nodes that establish spatial association edges with the candidate 3D path, and path geometric descriptions and path segment sequence descriptions of the candidate 3D path. Based on the path geometry description of the candidate 3D path and the spatial geometry description of the spatial object node, spatial intersection and spatial inclusion are performed. If the judgment result satisfies the hard constraint clause description corresponding to the prohibited and restricted land use in the rule layer, the candidate 3D path is marked as unsuccessful. The minimum distance is calculated based on the path geometry description of the candidate 3D path and the spatial geometry description of the spatial object node. The minimum distance is then compared with the preset backoff threshold of the hard constraint clause node. If the minimum distance is less than the preset backoff threshold, the candidate 3D path is marked as unacceptable. The path geometry description of the candidate 3D path is sampled at equal intervals along the path segment sequence description to obtain a set of sampling points. Based on the set of sampling points, the corresponding ground elevation and obstacle height are queried in the 3D planning base. The clearance margin at the sampling point is calculated and compared with the preset clearance threshold of the hard constraint clause node. If there is a sampling point where the clearance margin is less than the preset clearance threshold, the candidate 3D path is marked as unpassable. Output hard constraint verification marks for each candidate 3D path and summarize them to form a feasible candidate set.

8. The intelligent planning method for three-dimensional power distribution network routes based on knowledge graphs according to claim 1, characterized in that, Step seven specifically includes: For each candidate three-dimensional path within the feasible candidate set, a multi-objective evaluation record is established. The multi-objective evaluation record includes the cumulative cost value, risk index prediction value, cost index prediction value, construction period index prediction value, and evidence importance result corresponding to the cost field related to the laying method. The cumulative cost value is obtained by summing the path segments of the candidate three-dimensional path segment by segment in the cost field related to the laying method. A summary value of evidence importance is generated for each candidate 3D path based on the evidence importance results; A multi-objective screening vector is constructed using the cumulative cost, risk index prediction, cost index prediction, schedule index prediction, and evidence importance summary. The Pareto dominance relationship of candidate three-dimensional paths within the feasible candidate set is determined, and candidate three-dimensional paths that are not Pareto-dominated by other candidate three-dimensional paths are retained to form the Pareto minimum feasible solution set. Output a 3D route planning result package based on the Pareto minimum feasible solution set.

Images