Road network adjusting method and device
By constructing a road network map using a graph neural network (GNN), extracting features, and generating action probability distributions, the inefficiency of traditional road network update methods is solved, enabling intelligent and precise road network optimization and generating scientifically feasible planning schemes.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- GUANGDONG URBAN & RURAL PLANNING & DESIGN INST
- Filing Date
- 2026-01-13
- Publication Date
- 2026-04-24
AI Technical Summary
Traditional road network update methods rely on manual experience, which is inefficient and lacks intelligent data support, making it difficult to achieve scientific and accurate road network optimization. In particular, they cannot meet the needs of modern urban planning in complex update scenarios with multiple constraints.
A graph neural network (GNN) is used for road network adjustment. By acquiring road network data, a road network map is constructed, node and edge features are extracted, and node embeddings and edge embeddings are obtained using graph representation learning layers. Action probability distributions are generated to perform intelligent adjustment of the road network map. The road network scheme is optimized by combining loss functions and generalized advantage estimation.
It achieves a deep understanding and preservation of the spatial topology of road networks, provides a precise and structured data foundation, supports intelligent decision-making, solves the bottleneck problem of traditional methods in processing non-Euclidean spatial data, and generates scientific and feasible road network planning schemes.
Smart Images
Figure CN121920764A_ABST
Abstract
Description
Technical Field
[0001] This application relates to the field of urban planning technology, and in particular to a road network adjustment method and apparatus. Background Technology
[0002] As the "skeleton" of a city, the urban road network is a core element determining the city's spatial structure, functional layout, and traffic efficiency. Its planning rationality directly affects the overall efficiency of urban operation and the quality of life for residents. Road network renewal and optimization is a crucial aspect of urban renewal. It not only needs to meet the basic requirement of improving regional traffic efficiency but also adapt to diverse goals such as optimizing industrial layout, improving public service facilities, and protecting ecological spaces. It is a key support for promoting the upgrading of urban spatial structure. Traditional road network renewal methods have long relied on manual experience and expert knowledge, resulting in strong subjectivity, low efficiency, and a lack of intelligent data support and in-depth analysis capabilities. Summary of the Invention
[0003] Therefore, it is necessary to provide a road network adjustment method and apparatus to address the technical problem of low efficiency in manually delineating road networks in existing technologies.
[0004] To solve the above-mentioned technical problems, the present invention provides the following technical solution: A first aspect of the present invention provides a road network adjustment method, comprising: Obtain road network data for the target area; Based on the road network data, construct the road network map corresponding to the road network data; Based on the road network diagram, obtain the node embeddings of all nodes and the edge embeddings of all edges in the road network diagram; Based on the node embeddings of all nodes and the edge embeddings of all edges, the action probability distribution is obtained; Based on the action probability distribution, a corresponding action is selected, wherein the action is an atomic operation on the road network map; The road network map is adjusted according to the selected action.
[0005] Furthermore, the road network data consists of road centerlines and includes road type, road name, and road width; After acquiring the road network data of the target area, the road network data is further preprocessed, including: The centerlines of roads classified as urban arterial roads, secondary arterial roads, and local roads are retained to obtain a preliminary road network; The preliminary road network is topologically broken to generate road intersections. Basic information for each road segment is filled in by attribute attachment. Then, a topology check is performed to remove hanging nodes and pseudo nodes, resulting in preprocessed road network data.
[0006] Furthermore, constructing the road network map corresponding to the road network data includes: In the preprocessed road network data, each road centerline is used as an edge of the graph, and road intersections are used as nodes of the graph to establish the connection relationship between roads and nodes; each edge is assigned corresponding attribute information, including road type, name and width, to form a road network graph with topological structure and attribute labels.
[0007] Furthermore, feature extraction is performed on the road network map, and the extracted features are fused into the road network map. The feature extraction includes extracting node features and edge features, wherein: The node features are represented as follows:
[0008] In the formula, Represent node characteristics; Spatial information of a point is represented by the coordinates of a road intersection; The topology information is represented by the degree of the road intersection, which is obtained by calculating the number of road segments directly connected to the road intersection. Represents the constraint information of points, and defines the labels of road intersections as reserved or modifiable; Indicates will , and splicing; The edge features are represented as follows:
[0009] In the formula, Represent edge features; Spatial information about an edge is represented by the direction vector of the road segment and the coordinates of its midpoint. It indicates the level information and classifies them according to road type; Represents the constraint information of the edges, and defines the label of each road segment as reserved or modifiable; Indicates will , and splicing.
[0010] Furthermore, based on the road network diagram, the node embeddings of all nodes and the edge embeddings of all edges in the road network diagram are obtained, including: Using the road network map as input, the node embedding of each node and the edge embedding of each edge are obtained through the graph representation learning layer of the graph neural network.
[0011] Furthermore, based on the node embeddings of all nodes and the edge embeddings of all edges, the action probability distribution is obtained, including: The nodes of all nodes are embedded into the node deletion operation head to obtain the node deletion probability distribution, which represents the probability of deleting each node; The nodes of all nodes are embedded into the segment addition operation head to obtain the segment addition probability distribution, which represents the probability of connecting each pair of nodes to form a new segment; The edges of all edges are embedded into the road segment deletion operation header to obtain the road segment deletion probability distribution, which represents the probability of deleting each edge; Embed all edges into the node addition operation head to obtain the node addition probability distribution, which represents the probability of breaking each edge to form a new node; The probability distributions of node deletion, road segment addition, road segment deletion, and node addition constitute the probability distribution of the action.
[0012] Furthermore, it also includes evaluating the adjusted road network map, constructing a loss function based on the evaluation and action probability distribution, and updating the graph representation learning layer, node deletion operation head, road segment addition operation head, road segment deletion operation head, and node addition operation head of the graph neural network according to the loss function.
[0013] Furthermore, the adjusted road network map will be evaluated, including:
[0014] In the formula, This represents the evaluation function of the adjusted road network map. The adjusted road network map includes road network performance indicators, including road network connectivity. Road network density and the rationality of road network structure , These are the weight values for various road network performance indicators; This represents the road network constraint penalty in the adjusted road network map. It applies when edges or nodes labeled "to be retained" in the original road network map are deleted in the adjusted map. Take a negative value if the adjusted road network map does not delete edges or nodes labeled as "to be retained" from the original road network map. Set to 0; Among them, road network connectivity The calculation method is as follows:
[0015] In the formula, The set of nodes in the road network diagram; Represents a node v i With nodes v j The shortest path distance between them; This represents the total number of nodes in the road network. Road network density The calculation method is as follows:
[0016] In the formula, Let be the set of edges in the road network graph. Representing an edge Length, This indicates the total area of the road network planning area; Rationality of road network structure The calculation method is as follows:
[0017] In the formula, This refers to the total number of road classifications, including urban arterial roads, secondary arterial roads, and local roads. Indicates the first The percentage of the length of Class II roads in the road network map Indicates the first The proportion of the length of Class I roads in an ideal road network.
[0018] Furthermore, a loss function is constructed based on the evaluation and action probability distribution, including: Continuous adjustments are made based on the initial road network map, and each adjustment yields a corresponding evaluation. A long-term cumulative evaluation signal is formed based on all the evaluations corresponding to the adjustments. Based on the long-term cumulative evaluation signal, the long-term advantage value of each action is calculated using the generalized advantage estimation method; The loss function is obtained by weighting the action probability distribution with the corresponding long-term advantage value.
[0019] A second aspect of the present invention provides a road network adjustment device, the device comprising: The acquisition module acquires road network data of the target area; A construction module that constructs a road network map corresponding to the road network data based on the road network data; The first calculation module obtains the node embeddings of all nodes and the edge embeddings of all edges in the road network map based on the road network map. The second calculation module obtains the action probability distribution based on the node embeddings of all nodes and the edge embeddings of all edges; The selection module selects a corresponding action based on the action probability distribution, wherein the action is an atomic operation on the road network map; An adjustment module adjusts the road network map according to the selected action.
[0020] Compared with the prior art, the beneficial effects of the present invention are as follows: This invention discloses a road network adjustment method. First, road network data for a target area is acquired. Then, based on the road network data, a road network graph corresponding to the data is constructed, transforming the data into structured, numerical graph data for subsequent training calculations. Next, based on the road network graph, node embeddings of all nodes and edge embeddings of all edges are obtained. This process transforms the original road network data into node and edge embeddings rich in semantic information, providing a foundation for subsequent decision-making. Then, based on the node and edge embeddings, an action probability distribution is obtained. According to the action probability distribution, a corresponding action is selected, which is an atomic operation on the road network graph, to adjust the graph. Finally, the road network graph is adjusted based on the selected action. This invention abstracts urban road networks as graph structures for modeling, achieving a deep understanding and preservation of the spatial topological relationships of the road network. Furthermore, it understands and analyzes the road network from a holistic perspective, solving the bottleneck problem of traditional models' inability to effectively handle non-Euclidean spatial data like road networks, providing a precise and structured data foundation for subsequent intelligent decision-making. Attached Figure Description
[0021] Figure 1 A schematic flowchart of a road network adjustment method provided in an embodiment of the present invention; Figure 2 This is a schematic diagram of road network map structure modeling and feature extraction provided in an embodiment of the present invention; Figure 3 A schematic diagram of a GNN graph representation learning intelligent decision-making model provided in an embodiment of the present invention; Figure 4 This is a schematic diagram of model iterative training provided in an embodiment of the present invention; Figure 5 This is a schematic diagram of a road network adjustment device provided in an embodiment of the present invention. Detailed Implementation
[0022] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.
[0023] As the "skeleton" of a city, the urban road network is a core element determining the city's spatial structure, functional layout, and traffic efficiency. Its planning rationality directly affects the overall efficiency of urban operations and the quality of life for residents. At this critical stage of the transformation from new urbanization to high-quality development, urban renewal has become a core path to address issues such as urban space resource scarcity, functional imbalance, and aging infrastructure. The renewal and optimization of the road network is a crucial aspect of urban renewal, not only meeting the basic need for improved regional traffic efficiency but also adapting to diverse goals such as optimized industrial layout, improved public service facilities, and ecological space protection. It is a key support for promoting the upgrading of urban spatial structure. In the process of urban renewal, how to scientifically and efficiently generate or optimize regional road networks so that they can adapt to future traffic demands and land use changes while organically integrating with the existing urban fabric has long been a major challenge for the fields of urban planning and transportation engineering.
[0024] Traditional road network renewal methods have long relied on manual experience and expert knowledge, resulting in high subjectivity, low efficiency, and a lack of intelligent data support and in-depth analysis capabilities. Furthermore, existing methods struggle to quickly quantify and evaluate the comprehensive performance of renewal schemes in terms of macro-level traffic flow and road network accessibility. When faced with large-scale, multi-constraint, and complex renewal scenarios, they fail to meet the demands of modern urban planning for scientific rigor, precision, and efficiency.
[0025] While emerging data-driven approaches offer new ideas for road network updates, they also have limitations in terms of logical rigor and constraint handling. Taking mainstream generative artificial intelligence as an example, these image-based methods are essentially pixel-level calculations, which can easily lead to logical errors in road network connections. Furthermore, they struggle to effectively integrate existing road network structures and complex planning constraints during the generation process, thus failing to achieve constrained, localized, fine-grained adjustments. They are more suitable for scenarios like game map generation that involve creating something from scratch, but their applicability to urban road planning tasks requiring scientific optimization and updates based on the existing infrastructure is clearly limited.
[0026] Based on this, the purpose of this invention is to propose a road network adjustment method based on graph neural networks (GNNs). This method uses spatial topology rules as hard constraints and the learned deep features of the road network as the core basis to automatically and intelligently generate scientifically feasible road network planning schemes, ultimately providing efficient auxiliary decision-making tools and solid technical methodological support for urban planning and renewal activities.
[0027] The first embodiment of the present invention provides a road network adjustment method, such as... Figure 1 As shown, it includes: Obtain road network data for the target area; In this embodiment, road network data can be obtained from the Open Street Map (https: / / www.openstreetmap.org) website, by defining the target area and extracting the road network data within the range; Based on the road network data, construct the road network map corresponding to the road network data; Based on the road network diagram, obtain the node embeddings of all nodes and the edge embeddings of all edges in the road network diagram; Based on the node embeddings of all nodes and the edge embeddings of all edges, the action probability distribution is obtained; Based on the action probability distribution, a corresponding action is selected, wherein the action is an atomic operation on the road network map; The road network map is adjusted according to the selected action.
[0028] In a further embodiment, the road network data consists of road centerlines and includes road type, road name, and road width; After acquiring the road network data of the target area, the road network data is further preprocessed, including: The centerlines of roads classified as urban arterial roads, secondary arterial roads, and local roads are retained to obtain a preliminary road network; The preliminary road network is topologically broken to generate road intersections. Basic information for each road segment is filled in by attribute attachment. Then, a topology check is performed to remove hanging nodes and pseudo nodes, resulting in preprocessed road network data.
[0029] In a further embodiment, the construction of the road network map corresponding to the road network data is as follows: Figure 2 As shown, it includes: In the preprocessed road network data, each road centerline is used as an edge of the graph, and road intersections are used as nodes of the graph to establish the connection relationship between roads and nodes; each edge is assigned corresponding attribute information, including road type, name and width, to form a road network graph with topological structure and attribute labels.
[0030] In this embodiment, the road network map Expressed as a formula:
[0031] In the formula, V For a set of nodes, each node v i ∈ V This indicates the intersection between road segments. E For an edge set, each edge e ij ∈ V This represents an independent road segment, serving as an intersection node.v i Intersection node v j Direct spatial connection between them; Furthermore, feature extraction is performed on the road network map, and the extracted features are fused into the road network map. The feature extraction includes extracting node features and edge features, wherein: The node features are used to describe the spatial, topological, and constraint information of road intersections, and are represented as follows:
[0032] In the formula, Represent node characteristics; It represents the spatial information of a point, using the coordinates of a road intersection, to determine the absolute location of the intersection; The topology information is represented by the degree of the road intersection. The degree of the road intersection is obtained by calculating the number of road segments directly connected to the road intersection, which is used to reflect the local connectivity importance of the intersection in the road network. The constraint information of the points is represented. Combined with the current status of the road network and planning requirements, constraint labels are assigned to the intersections. The labels of the road intersections are defined as reserved or modifiable, which serve as key hard constraints to prohibit modifications to the protected intersections in subsequent decisions. Indicates will , and After concatenating and extracting the original attribute values, the data undergoes standardization, normalization, and encoding processes. Finally, all features are concatenated to obtain the complete feature vector of the node. The edge features describe information such as the spatial structure, hierarchy, and constraints of the road segment, and are represented as follows:
[0033] In the formula, Represent edge features; The spatial information of the edge is represented by the direction vector and midpoint coordinates of the road segment, and is used to record and calculate the length, direction and spatial location of the road segment. The classification information is based on the highway label in the OSM data, including urban arterial roads, secondary arterial roads, and local roads, which are used to reflect the functional level and basic traffic conditions of the road segment. The constraint information representing the edge is based on the current status of the road network and planning requirements. Constraint labels are assigned to road segments, and the label of each road segment is defined as either reserved or modifiable. These labels serve as key hard constraints to prohibit modifications to protected road segments in subsequent decision-making. Indicates will , and After splicing and extracting the original attribute values, the data undergoes standardization, normalization, and encoding processes. Finally, all features are combined to obtain the complete feature vector of the edges.
[0034] Through the above process, the original road network map G is transformed into a feature-rich graph, forming structured and numerical graph data, which can then be input into the graph neural network for training and computation.
[0035] In a further embodiment, such as Figure 3 As shown, based on the road network diagram, the node embeddings of all nodes and the edge embeddings of all edges in the road network diagram are obtained, including: Using the road network map as input, the node embedding of each node and the edge embedding of each edge are obtained through the graph representation learning layer of the graph neural network.
[0036] In this embodiment, the road network map fully encodes the topological relationships and attribute characteristics of the current road network. By aggregating information about nodes and their neighbors through a message passing mechanism, it provides a complete representation of each node in the map. v i ∈ V and every edge e ij ∈ E A high-dimensional feature representation vector is calculated. This process transforms the original road network data into node embeddings (H) rich in semantic information. V and edge embedding H E This provides a basis for subsequent decision-making.
[0037] This invention, for the first time, abstracts urban road networks as a graph structure for modeling and employs a Graph Neural Network (GNN) as the core computational framework, achieving a deep understanding and preservation of the spatial topological relationships of the road network. Unlike traditional methods that treat road networks as isolated point sets or raster data, this invention maps intersections to graph nodes and road segments to graph edges. Utilizing the message passing mechanism of GNNs, it explicitly models and learns the connections between nodes and the global topological structure. This method enables computers to understand and analyze road networks from a holistic, interconnected perspective, much like human experts. It solves the bottleneck problem of traditional models' inability to effectively handle non-Euclidean spatial data like road networks, providing a precise and structured data foundation for subsequent intelligent decision-making.
[0038] Based on the node embeddings of all nodes and the edge embeddings of all edges, the action probability distribution is obtained, including: The node embeddings of all nodes are input into the node deletion operation head to obtain the node deletion probability distribution, representing the probability of deleting each node. This node deletion operation head is for each existing node. v iCalculate a scalar score, normalize it using the Softmax function, and obtain the probability distribution Premove_node on the node set; The node embeddings of all nodes are input into the segment addition operation head to obtain the segment addition probability distribution, which represents the probability of connecting each pair of nodes to form a new segment. This operation head evaluates all possible node pairs ( v p , v q (Especially for node pairs that are not currently directly connected), calculate a connection score for each pair of nodes, and after normalization, obtain a probability distribution Padd_edge on potential connections, recommending the intersection pairs that should be connected; The edges of all edges are embedded into the segment deletion operation header to obtain the segment deletion probability distribution, representing the probability of deleting each edge. This operation header contains the probability of deleting each existing edge. e ij Calculate a scalar fraction, and then normalize it using Softmax to obtain the probability distribution Premove_edge on the edge set; Embedding all edges into the node addition operation head yields the node addition probability distribution, representing the probability of breaking each edge to form a new node. This operation head is for each existing edge. e ij Calculate a score, normalize it to obtain the probability distribution Padd_node, which represents the possibility of choosing to split on a certain road segment to increase the probability of a new intersection; The probability distributions of node deletion, road segment addition, road segment deletion, and node addition constitute the probability distribution of the action.
[0039] In a further embodiment, the method further includes evaluating the adjusted road network map, constructing a loss function based on the evaluation and action probability distribution, and updating the graph representation learning layer, node deletion operation head, road segment addition operation head, road segment deletion operation head, and node addition operation head of the graph neural network according to the loss function. Figure 4 As shown.
[0040] In this embodiment, the entire update process is as follows: 1) State: At each optimization step t, the state s_t is the current road network graph structure G_t, which fully encodes the topology and attributes of the road network at this time; 2) Decision-making: The GNN intelligent decision-making model (i.e., the model constructed by the graph representation learning layer of the graph neural network, the node deletion operation head, the road segment addition operation head, the road segment deletion operation head, and the node addition operation head in this embodiment of the invention) acts as the policy function π of the framework. It receives the state s_t and outputs a structured adjustment action probability distribution π(|s_t). The system samples according to this distribution to obtain a specific graph operation action a_t (such as AddEdge(v_i, v_j)); 3) Value Judgment: A comprehensive evaluation function R is established, as shown in the formula. This function is used to quantitatively evaluate the new state s_t+1 (i.e., the adjusted new road network) resulting from the model's decisions. This function quantifies the planning objectives (e.g., improving connectivity, ensuring structural rationality) and constraints (e.g., protecting specific elements) into a scalarized evaluation signal r_t+1. This signal directly measures the merits of action a_t and constitutes the core feedback driving model parameter updates. Specifically, it can be:
[0041] in, Represents road network performance indicators, calculation The performance characteristics of the road network under its current state, including road network connectivity. Road network density Rationality of road network structure . These are the calculated weight values for each feature. This indicates penalties for violations of road network constraints, detecting actions that violate these constraints. If the action deletes a road segment or intersection that must be retained, then... It is assigned a very large negative value; otherwise, if the action satisfies the constraints... It is 0.
[0042]
[0043] To assess road network connectivity, the degree of connectivity and traffic efficiency of the road network are quantified. Among these, The set of intersection nodes in the road network; Calculate the shortest path distance between node vi and node vj. This indicates the connection efficiency between nodes vi and vj, when the two nodes are unreachable. , =0; This represents the total number of nodes in the road network.
[0044]
[0045] Road network density quantifies the spatial coverage of the road network. Among other things, A set of road segments in a road network; Calculate road segment Length; This refers to the total area of the road network planning area.
[0046]
[0047] To ensure the rationality of the road network structure, the rationality of the road network hierarchy at the functional level is quantified. Among these, This represents the total number of road classifications (arterial roads, secondary arterial roads, and local roads). For the first The proportion of the length of Class II roads in the actual road network. For the first The proportion of the length of a Class II road in an ideal structure.
[0048] This invention breaks through the traditional optimization model that relies on static rules or a single objective function, proposing a new paradigm for road network optimization through dynamic decision-making and trial-and-error in a simulated environment. This mechanism guides the optimization process by establishing a dual-drive signal that combines hard constraints (unchangeable retained elements) with soft objectives (evaluation indicators such as connectivity and road network density). Specifically, hard constraints fundamentally guarantee the feasibility of the solution; while soft objectives are quantified as dynamic evaluation signals to continuously assess the effectiveness of decisions and drive the updating of model parameters. This mechanism enables the model to automatically explore and converge to the road network adjustment scheme with optimal overall performance while strictly satisfying actual planning constraints, achieving integrated and coordinated optimization of constraints and objectives.
[0049] Iterative training process: 1) Interaction and Data Collection: The current GNN intelligent decision-making model is placed in a simulation environment, starting from its initial state s_0. Based on the probability distribution output by the model, it executes decision actions, thus generating a state evolution trajectory (s_0, a_0, s_1, a_1, ..., s_T). During this process, the comprehensive evaluation function R generates a judgment signal r_{t+1} at each step based on the new state s_{t+1}.
[0050] 2) Policy Evaluation and Gradient Calculation: Based on long-term cumulative evaluation signals, the generalized advantage estimation method is used to calculate the long-term advantage value of each action, and a policy loss function is constructed accordingly. The loss function is a weighted combination of the action probability distribution and the corresponding long-term advantage value. Subsequently, the gradient of this loss function with respect to the GNN model parameters θ is calculated through differentiation, yielding the policy gradient. This gradient indicates how the GNN model parameters θ should be adjusted to increase the probability of generating high-value decision sequences.
[0051] 3) Parameter Update and Model Iteration: Using the gradients described above, all parameters of the GNN model are updated via backpropagation. The updated model will have better decision-making capabilities and is subsequently used for a new round of interaction and data collection, thus forming an automatic closed loop of "decision-evaluation-optimization". This loop repeats continuously until the model's decision-making performance tends to stabilize and reach its optimal level.
[0052] In a further embodiment, after the model training is completed, the road network to be updated is processed according to the same specifications, a graph structure is constructed and features are extracted, and then input into the final trained GNN intelligent decision model. After calculation, a preliminary road network adjustment plan is generated, and a preliminary road network planning adjustment plan after processing road segments and intersections is output. The initial road network scheme output by the model is quantitatively evaluated. The evaluation metrics are highly correlated with the reward metrics during model training, covering road network connectivity efficiency, density, and hierarchical structure matching, to measure the overall performance of the output scheme. After evaluation, local manual optimization and correction of the scheme details are allowed, provided that the original hard constraints are strictly adhered to.
[0053] A second embodiment of the present invention provides a road network adjustment device, such as... Figure 5 As shown, the device includes: The acquisition module acquires road network data of the target area; A construction module that constructs a road network map corresponding to the road network data based on the road network data; The first calculation module obtains the node embeddings of all nodes and the edge embeddings of all edges in the road network map based on the road network map. The second calculation module obtains the action probability distribution based on the node embeddings of all nodes and the edge embeddings of all edges; The selection module selects a corresponding action based on the action probability distribution, wherein the action is an atomic operation on the road network map; An adjustment module adjusts the road network map according to the selected action.
[0054] A third embodiment of the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the steps of the method described in the first embodiment of the present invention.
[0055] The processor mentioned above is the processor in the computer device described in the above embodiments. The computer-readable storage medium may be a computer read-only memory (ROM), random access memory (RAM), magnetic disk, or optical disk.
[0056] In this embodiment, the module can be implemented in software so that it can be executed by various types of processors. For example, an identified executable code module may include one or more physical or logical blocks of computer instructions, which may be constructed as objects, procedures, or functions. Nevertheless, the executable code of the identified module does not need to be physically located together, but may include different instructions stored in different bits, which, when logically combined, constitute the module and achieve the module's intended purpose.
[0057] Through the above description of the implementation methods, those skilled in the art can clearly understand that the methods of the above embodiments can be implemented by means of software plus necessary general-purpose hardware platforms. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product is stored in a storage medium (such as ROM / RAM, disk, optical disk) and includes several instructions for executing the methods described in the various embodiments of this application.
[0058] The exemplary embodiments described above are with reference to the accompanying drawings. Many different forms and embodiments are feasible without departing from the spirit and teachings of this application. Therefore, this application should not be construed as limiting the exemplary embodiments set forth herein. Rather, these exemplary embodiments are provided to make this application complete and convey the scope of this application to those skilled in the art. In these drawings, component dimensions and relative dimensions may be exaggerated for clarity. The terminology used herein is for the purpose of describing particular exemplary embodiments only and is not intended to be limiting. As used herein, unless clearly indicated otherwise, the singular forms “a,” “an,” and “the” are intended to include all such forms. It will be further understood that the terms “comprising” and / or “including”, when used in this specification, indicate the presence of the stated features, integers, steps, operations, components, and / or elements, but do not exclude the presence or addition of one or more other features, integers, steps, operations, components, and / or groups thereof. Unless otherwise indicated, when stated, a range of values includes the upper and lower limits of the range and any subranges in between.
[0059] The above description is the preferred embodiment of this application. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the principles described in this application, and these improvements and modifications should also be considered within the scope of protection of this application.
Claims
1. A road network adjustment method, characterized in that, include: Obtain road network data for the target area; Based on the road network data, construct the road network map corresponding to the road network data; Based on the road network diagram, obtain the node embeddings of all nodes and the edge embeddings of all edges in the road network diagram; Based on the node embeddings of all nodes and the edge embeddings of all edges, the action probability distribution is obtained; Based on the action probability distribution, a corresponding action is selected, wherein the action is an atomic operation on the road network map; The road network map is adjusted according to the selected action.
2. The road network adjustment method according to claim 1, characterized in that, The road network data consists of road centerlines, including road type, road name, and road width; After acquiring the road network data of the target area, the road network data is further preprocessed, including: The centerlines of roads classified as urban arterial roads, secondary arterial roads, and local roads are retained to obtain a preliminary road network; The preliminary road network is topologically broken to generate road intersections. Basic information for each road segment is filled in by attribute attachment. Then, a topology check is performed to remove hanging nodes and pseudo nodes, resulting in preprocessed road network data.
3. The road network adjustment method according to claim 2, characterized in that, The construction of the road network map corresponding to the road network data includes: In the preprocessed road network data, each road centerline is used as an edge of the graph, and road intersections are used as nodes of the graph to establish the connection relationship between roads and nodes; each edge is assigned corresponding attribute information, including road type, name and width, to form a road network graph with topological structure and attribute labels.
4. The road network adjustment method according to claim 3, characterized in that, Furthermore, feature extraction is performed on the road network map, and the extracted features are fused into the road network map. The feature extraction includes extracting node features and edge features, wherein: The node features are represented as follows: In the formula, Represent node characteristics; Spatial information of a point is represented by the coordinates of a road intersection; The topology information is represented by the degree of the road intersection, which is obtained by calculating the number of road segments directly connected to the road intersection. Represents the constraint information of points, and defines the labels of road intersections as reserved or modifiable; Indicates will , and splicing; The edge features are represented as follows: In the formula, Represent edge features; Spatial information about an edge is represented by the direction vector of the road segment and the coordinates of its midpoint. It indicates the level information and classifies them according to road type; Represents the constraint information of the edges, and defines the label of each road segment as reserved or modifiable; Indicates will , and splicing.
5. The road network adjustment method according to any one of claims 1 to 4, characterized in that, Based on the road network diagram, the node embeddings of all nodes and the edge embeddings of all edges in the road network diagram are obtained, including: Using the road network map as input, the node embedding of each node and the edge embedding of each edge are obtained through the graph representation learning layer of the graph neural network.
6. The road network adjustment method according to claim 5, characterized in that, Based on the node embeddings of all nodes and the edge embeddings of all edges, the action probability distribution is obtained, including: The nodes of all nodes are embedded into the node deletion operation head to obtain the node deletion probability distribution, which represents the probability of deleting each node; The nodes of all nodes are embedded into the segment addition operation head to obtain the segment addition probability distribution, which represents the probability of connecting each pair of nodes to form a new segment; The edges of all edges are embedded into the road segment deletion operation header to obtain the road segment deletion probability distribution, which represents the probability of deleting each edge; Embed all edges into the node addition operation head to obtain the node addition probability distribution, which represents the probability of breaking each edge to form a new node; The probability distributions of node deletion, road segment addition, road segment deletion, and node addition constitute the probability distribution of the action.
7. The road network adjustment method according to claim 5, characterized in that, It also includes evaluating the adjusted road network map, constructing a loss function based on the evaluation and action probability distribution, and updating the graph representation learning layer, node deletion operation head, road segment addition operation head, road segment deletion operation head, and node addition operation head of the graph neural network according to the loss function.
8. The road network adjustment method according to claim 7, characterized in that, An evaluation of the revised road network map was conducted, including: In the formula, This represents the evaluation function of the adjusted road network map. The adjusted road network map includes road network performance indicators, including road network connectivity. Road network density and the rationality of road network structure , These are the weight values for various road network performance indicators; This represents the road network constraint penalty in the adjusted road network map. It applies when edges or nodes labeled "to be retained" in the original road network map are deleted in the adjusted map. Take a negative value if the adjusted road network map does not delete edges or nodes labeled as "to be retained" from the original road network map. Set to 0; Among them, road network connectivity The calculation method is as follows: In the formula, The set of nodes in the road network diagram; Represents a node v i With nodes v j The shortest path distance between them; This represents the total number of nodes in the road network. Road network density The calculation method is as follows: In the formula, Let be the set of edges in the road network graph. Representing an edge Length, This indicates the total area of the road network planning area; Rationality of road network structure The calculation method is as follows: In the formula, This refers to the total number of road classifications, including urban arterial roads, secondary arterial roads, and local roads. Indicates the first The percentage of length of Class I roads in the road network map. Indicates the first The proportion of the length of Class I roads in an ideal road network.
9. The road network adjustment method according to claim 7 or 8, characterized in that, A loss function is constructed based on the evaluation and action probability distribution, including: Continuous adjustments are made based on the initial road network map, and each adjustment yields a corresponding evaluation. A long-term cumulative evaluation signal is formed based on all the evaluations corresponding to the adjustments. Based on the long-term cumulative evaluation signal, the long-term advantage value of each action is calculated using the generalized advantage estimation method; The loss function is obtained by weighting the action probability distribution with the corresponding long-term advantage value.
10. A road network adjustment device, characterized in that, The device includes: The acquisition module acquires road network data of the target area; A construction module that constructs a road network map corresponding to the road network data based on the road network data; The first calculation module obtains the node embeddings of all nodes and the edge embeddings of all edges in the road network map based on the road network map. The second calculation module obtains the action probability distribution based on the node embeddings of all nodes and the edge embeddings of all edges; The selection module selects a corresponding action based on the action probability distribution, wherein the action is an atomic operation on the road network map; An adjustment module adjusts the road network map according to the selected action.