A method and system for evaluating and selecting a newly built road based on a graph neural network

Abstract

The present invention discloses a method and system for evaluating and selecting newly built roads based on graph neural networks, which relates to the field of road design and includes: obtaining multiple newly built road construction plans, each plan including the layout, connection method, and design parameters of the newly built road; modeling each newly built road together with the existing roads as a graph structure; assigning features to the nodes of each graph structure; using a trained graph neural network evaluation model to predict the accident risk of each road section, calculating a global risk index based on the predicted accident risk, and selecting the plan with the lowest risk. The present invention solves the problems of low evaluation efficiency and insufficient accuracy of traditional newly built road plan evaluation.

Description

Technical Field

[0001] The present invention relates to the field of road design, and more specifically, to a method and system for evaluating and selecting a newly built road based on a graph neural network. Background Art

[0002] Road construction occupies an important position in urban planning and infrastructure development. Scientific and reasonable road design and layout have a direct impact on traffic safety, operation efficiency, and the overall development of the city. In the actual process of designing a newly built road, not only the starting point, ending point, and environmental constraints need to be considered, but also the connection relationship with the existing road network needs to be comprehensively evaluated to ensure that the newly built road can be efficiently integrated with the existing road system. For example, design parameters such as the width, curvature radius, and slope of the newly built road often need to be coordinated with the connected existing roads to avoid traffic accidents and improve the operation efficiency of the overall road network. However, the traditional road design evaluation process mostly relies on empirical judgment and manual analysis, lacking systematic and automated methods, and it is difficult to quickly, accurately, and comprehensively evaluate multiple road schemes. Summary of the Invention

[0003] The technical problem to be solved by the present invention is to provide a method and system for evaluating and selecting a newly built road based on a graph neural network to solve the problems mentioned in the background art.

[0004] To achieve the above object, the present invention adopts the following technical solutions:

[0005] A method for evaluating a newly built road construction plan based on a graph neural network, comprising the following steps:

[0006] Obtain multiple newly built road construction plans, each plan including the layout, connection method, and design parameters of the newly built road;

[0007] Model each newly built road construction plan as a graph structure ;

[0008] Wherein represents a set of road segments, including newly built road segments and existing road segments; represents the connection relationship between road segments;

[0009] Assign features to the nodes of each graph structure. The node features of the existing road include the existing features of the road, and the node features of the newly built road include the design features of the newly built road;

[0010] Use the trained graph neural network evaluation model to process each graph structure, predict the accident risk of each road segment, and calculate the global risk index of each newly built road construction plan according to the predicted accident risk;

[0011] Based on the global risk metrics, compare multiple new road construction plans and select the plan with the lowest risk as the optimal new road construction plan.

[0012] Optionally, the method for generating a new road construction plan includes:

[0013] Take the starting point constraint conditions, ending point constraint conditions, environmental constraint conditions, and the structure of the existing road network diagram as input information;

[0014] Use a graph neural network encoder to jointly encode the input information to obtain a latent space representation vector;

[0015] Input the latent space representation vector into a graph structure decoder to generate multiple candidate new road plans that meet the constraint conditions and can be connected to the existing road network, with each plan represented by a graph structure.

[0016] Optionally, the graph structure decoder is a graph variational autoencoder, specifically including:

[0017] The encoding module is used to jointly encode the starting point constraint conditions, ending point constraint conditions, environmental constraint conditions, and the existing road structure into Gaussian distribution parameters in the latent space, and sample the latent space representation vector using the Gaussian distribution parameters;

[0018] The decoding module gradually generates the nodes and connection relationships of the new road through the latent space representation vector, forming a new road graph structure that can be topologically connected to the existing road network.

[0019] Optionally, the method further includes fusing the generated new road graph structure with the existing road graph structure into a combined graph;

[0020] Use the graph neural network evaluation model to perform risk prediction and evaluation on the combined graph;

[0021] Take the prediction result as feedback information to optimize the latent space sampling strategy or the graph structure decoder parameters in the structure generation stage.

[0022] Optionally, the starting point constraint conditions, ending point constraint conditions, and environmental constraint conditions include at least one or more of the following:

[0023] The geographical coordinates of the starting point and the ending point;

[0024] The maximum allowable slope;

[0025] Avoidance areas;

[0026] The minimum turning radius;

[0027] Regional planning layer information.

[0028] Optionally, the node features of the newly built road include any one or more of the following:

[0029] Virtual road image, obtained by: constructing a 3D road model based on road design parameters and terrain elevation data; rendering a virtual image of the road based on the 3D road model;

[0030] Road features, including road width, road length, radius of curvature, and slope.

[0031] Optionally, the node features of the existing road section include image features, road features, historical traffic flow data, and historical accident data.

[0032] Optionally, the calculation formula of the global risk index is:

[0033] ;

[0034] where is the global risk index of the graph structure G, is the number of nodes in the graph structure G, is the accident risk prediction value of the th node, is the weight of the corresponding node, and the weight is determined comprehensively by the road section length, traffic flow, and accident severity.

[0035] The present invention also discloses a newly built road evaluation and selection system based on a graph neural network, including:

[0036] A scheme acquisition module, configured to acquire multiple newly built road construction schemes, and each scheme includes the layout, connection method, and design parameters of the newly built road;

[0037] A graph construction module, configured to model each newly built road construction scheme as a graph structure ;

[0038] where represents the road section set, including newly built road sections and existing road sections; represents the connection relationship between road sections;

[0039] A feature assignment module, configured to assign features to the nodes of each graph structure. The node features of the existing road include the existing features of the road, and the node features of the newly built road include the design features of the newly built road;

[0040] An evaluation module, including a trained graph neural network evaluation model, configured to process each graph structure, predict the accident risk of each road section, and calculate the global risk index of each newly built road construction scheme according to the predicted accident risk;

[0041] A comparison module, configured to compare multiple newly-built road construction plans based on the global risk indicator, and select the plan with the lowest risk as the optimal newly-built road construction plan.

[0042] Optionally, the system further includes:

[0043] A structure generation module, configured to generate multiple candidate newly-built road structure plan diagrams based on the starting point constraint conditions, ending point constraint conditions, environmental constraint conditions, and existing road structure information of road construction;

[0044] A feedback optimization module, configured to use the global risk indicator output by the evaluation module as a feedback signal to optimize and update the potential space sampling strategy, generation model parameters, or path selection strategy of the structure generation module, so as to achieve closed-loop self-optimization of structure generation and evaluation.

[0045] The advantages of the present invention over the prior art are that the present invention effectively solves the problems of low efficiency and strong subjectivity in manual evaluation during the traditional road evaluation process. By combining the newly-built road with the existing road to form a graph structure, and assigning specific design and existing feature information to each node (section), using the trained graph neural network model to automatically predict the accident risk of each section and calculate the global risk indicator, the rapid comparison and evaluation of multiple newly-built road construction plans can be realized. The present invention not only considers the safety of the newly-built road, but also fully considers the impact of the newly-built road on the existing road after being connected to the existing road.

[0046] Furthermore, the present invention also provides a method for generating a newly-built road structure, which uses a graph neural network to encode the starting point, ending point, and environmental constraint conditions, and combines the existing road network structure information, and jointly encodes this information into the potential space, and automatically generates multiple candidate plans for selection through a graph structure decoder, greatly reducing manpower and material resources. In addition, the present invention also realizes a closed-loop feedback mechanism between the structure generation and evaluation processes, feeds the evaluation results back into the generation model to optimize the sampling strategy or model parameters of the potential space, and continuously improves the rationality and practicality of the plan generation. Description of the Drawings

[0047] Figure 1 is a schematic diagram of the method of the present invention;

[0048] Figure 2 is a schematic diagram of the method of the present invention for automatically generating a newly-built road plan;

[0049] Figure 3 is a simplified schematic diagram of the graph structure of the present invention. Detailed Embodiments

[0050] The following describes the detailed embodiments of the present invention with reference to the drawings.

[0051] As shown in Figure 1 the schematic diagram of the method of the present invention, the following steps are included:

[0052] Obtain multiple new road construction plans, each plan including the layout, connection method and design parameters of the new road;

[0053] Model each new road construction plan as a graph structure ;

[0054] wherein represents the set of road segments, including new road segments and existing road segments; represents the connection relationship between road segments;

[0055] Assign features to the nodes of each graph structure. The node features of existing roads include the existing features of the roads, and the node features of new roads include the design features of the new roads; As Figure 3 shown is a schematic diagram of a simplified graph structure.

[0056] Use the trained graph neural network evaluation model to process each graph structure, predict the accident risk of each road segment, and calculate the global risk index of each new road construction plan according to the predicted accident risk;

[0057] Based on the global risk index, compare multiple new road construction plans and select the plan with the lowest risk as the optimal new road construction plan.

[0058] Among them, the graph neural network evaluation model uses a graph neural network. A graph neural network is a type of neural network model specifically designed for processing graph-structured data. Its core idea is to perform information propagation and feature aggregation through the connection relationships between nodes, so that each node can fuse the feature information from neighboring nodes after several layers of propagation. In the application of road structure evaluation, the graph neural network first models the entire road network as a graph structure, where each node represents a road segment or intersection, and the features of the node can include road width, traffic flow, pedestrian density, historical accident data, surrounding environment image features, etc., while the edges represent the connection relationships between road segments. During the training process, the system inputs a large number of historical road networks and their corresponding accident rates as supervision labels into the model. The model learns how to predict the accident risk probability of a node based on the features of the node itself and the features of neighboring nodes through the forward propagation process, and continuously adjusts the model parameters through backpropagation and optimization algorithms to minimize the error between the predicted accident rate and the true accident rate. After sufficient training, the graph neural network has strong generalization ability and can accurately predict the accident risk of each node on the new road structure diagram. Therefore, when a new road is added, since the overall road structure has changed, the accident risk can also be predicted again through the graph neural network.

[0059] In the above embodiments, the new road construction plan can be determined by manual selection. However, in other embodiments, the new road construction plan can also be automatically generated by a neural network.

[0060] As Figure 2 shown, specifically, the method for generating a new road construction plan includes:

[0061] Taking the starting point constraint condition, ending point constraint condition, environmental constraint condition of road construction, and the existing road network graph structure as input information;

[0062] Using a graph neural network encoder to jointly encode the input information to obtain a latent space representation vector;

[0063] Inputting the latent space representation vector into a graph structure decoder to generate multiple candidate new road plans that meet the constraint conditions and can be connected to the existing road network. Each plan is represented by a graph structure that only contains the new road part.

[0064] Among them, in a preferred embodiment, the graph structure decoder is a graph variational autoencoder, specifically including:

[0065] An encoding module is used to jointly encode the starting point constraint condition, ending point constraint condition, environmental constraint condition, and the existing road structure into Gaussian distribution parameters in the latent space, and sample the latent space representation vector using the Gaussian distribution parameters;

[0066] A decoding module gradually generates the nodes and connection relationships of the new road through the latent space representation vector to form a new road graph structure that can be topologically connected to the existing road network.

[0067] Among them, the design of the graph neural network encoder is based on graph structure data modeling and information propagation mechanisms. In a graph structure, nodes and edges together form an irregular data structure, different from the regular grid input in traditional convolutional neural networks. Graph neural networks rely on an adjacency-aware calculation method: each node continuously aggregates the information of its neighbor nodes during forward propagation to iteratively update its own representation. This process can effectively capture the spatial topological features of the local structure in the graph and the complex dependencies between nodes. In the present invention, the road network naturally has graph structure characteristics. Nodes can represent road intersections or section endpoints, and edges represent section connection relationships. The starting point constraint, ending point constraint, and environmental constraint can be expressed in an embedded manner in node features or edge weights. Therefore, it is very suitable to use a graph neural network encoder for representation learning. After sending this information into the graph neural network encoder, the encoder effectively fuses local and global structure information through multiple propagation operations, thereby forming a latent space vector representation that can comprehensively reflect road design requirements, geographical conditions, and the existing network structure.

[0068] To endow the model with the ability to generate diverse structures, the method of the present invention further introduces the framework of a graph variational autoencoder. Different from a general autoencoder that directly maps to a deterministic vector, the encoding module of the graph variational autoencoder maps the input graph structure to the parameters of a Gaussian distribution in the latent space - the mean and variance vectors, and samples from them to obtain the latent space representation vector. This mechanism introduces probability and generalization ability, enabling multiple candidate solutions that meet the requirements but have different structures to be generated under the same input conditions. This latent vector serves as the input to the decoder, controlling the overall style and distribution range of the subsequent generated graph structure.

[0069] In the decoding stage, the decoder gradually grows a graph structure that only contains the newly built road part based on this latent vector. The generation process follows the principle of topological consistency. The decoder can use a recursive or attention mechanism to determine the generation position of the next node and the connection method with the existing nodes until a complete graph is formed. Since the key constraints in the input graph structure have been modeled during the training process of the latent space, the structure generated by the decoder naturally meets the connection requirements of the starting point and the ending point, and will not violate the input environmental constraints. Moreover, through feature consistency, it ensures topological connectivity with the existing road network. This idea of generating based on a probabilistic graph provides powerful model freedom and structure expression ability in principle, making the road scheme design for constraint optimization not only feasible in structure, but also have good generalization ability and generation efficiency. When generating, in principle, only the newly built road part needs to be generated to improve efficiency.

[0070] In a further embodiment, the method of the present invention further includes fusing the generated newly built road graph structure with the existing road graph structure into a combined graph;

[0071] using the graph neural network evaluation model to perform risk prediction and evaluation on the combined graph;

[0072] using the prediction result as feedback information to optimize the latent space sampling strategy or the graph structure decoder parameters in the structure generation stage.

[0073] In the above embodiment, the predicted risk assessment result is used as feedback in the structure generation stage in reverse. This not only enables the generation model to learn to avoid high-risk areas and connection methods in each round of iteration, but also guides the generation model to be more inclined to output road schemes with reasonable structures, topological stability and lower risks through fine-tuning of the latent space sampling strategy and decoder parameters. This self-optimization mechanism that integrates generation and evaluation breaks the passive process of separating traditional design and verification, enables the scheme generation to have the active optimization ability oriented to the goal, and constructs an intelligent closed loop of perception - generation - feedback - regeneration in principle, significantly improving the adaptability and generation quality of the system in complex traffic planning tasks.

[0074] In further embodiments, the starting point constraint condition, the ending point constraint condition, and the environmental constraint condition include at least one or more of the following:

[0075] The geographical coordinates of the starting point and the ending point;

[0076] The maximum allowable slope;

[0077] The avoidance area;

[0078] The minimum turning radius;

[0079] The regional planning layer information.

[0080] In further embodiments, the node features of the above-mentioned newly built road include any one or more of the following:

[0081] The virtual road image, and the obtaining method is: constructing a road three-dimensional model according to the road design parameters and the terrain elevation data; rendering and generating a virtual image of the road based on the road three-dimensional model;

[0082] The road features, including the road width, the road length, the radius of curvature, and the slope.

[0083] In further embodiments, the node features of the existing road section include image features, road features, historical traffic flow data, and historical accident data.

[0084] In further embodiments, the calculation formula of the global risk index is:

[0085] ;

[0086] Wherein, is the global risk index of the graph structure G, is the number of nodes in the graph structure G, is the accident risk prediction value of the th node, is the weight corresponding to the node, and the weight is comprehensively determined by the road section length, the traffic flow, and the accident severity, or can also be artificially assigned.

[0087] The present invention also discloses a newly built road evaluation and selection system based on a graph neural network, including:

[0088] A scheme acquisition module, configured to acquire multiple newly built road construction schemes, and each scheme includes the layout, connection method, and design parameters of the newly built road;

[0089] A graph construction module, configured to model each newly built road construction scheme as a graph structure ;

[0090] Wherein Represents a set of road segments, including newly built road segments and existing road segments; Represents the connection relationship between road segments;

[0091] A feature assignment module for assigning features to the nodes of each graph structure. The node features of existing roads include the existing features of the roads, and the node features of newly built roads include the design features of the newly built roads;

[0092] An evaluation module, including a trained graph neural network evaluation model, for processing each graph structure, predicting the accident risk of each road segment, and calculating the global risk index of each newly built road construction plan based on the predicted accident risk;

[0093] A comparison module for comparing multiple newly built road construction plans based on the global risk index and selecting the plan with the lowest risk as the optimal newly built road construction plan.

[0094] In an alternative embodiment, the system further includes:

[0095] A structure generation module for generating multiple candidate newly built road structure plan graphs based on the starting point constraint conditions, ending point constraint conditions, environmental constraint conditions, and existing road structure information of road construction;

[0096] A feedback optimization module for using the global risk index output by the evaluation module as a feedback signal to optimize and update the potential space sampling strategy, generation model parameters, or path selection strategy of the structure generation module, thereby realizing the closed-loop self-optimization of structure generation and evaluation.

[0097] In terms of the hardware platform, the system can use high-performance computing devices, such as servers or dedicated workstations, specifically including high-performance CPUs, paired with high-performance GPUs, to efficiently support the training and inference calculations of graph neural networks. In addition, the server also needs to be equipped with sufficient network interfaces to achieve data interaction with external data platforms or design departments.

[0098] In terms of software implementation, this system can be deployed based on the Linux system to ensure system stability and efficiency. The software architecture design of the system mainly includes a data acquisition and management subsystem, a road structure modeling subsystem, a graph neural network evaluation subsystem, a structure plan generation subsystem, and a closed-loop feedback optimization subsystem.

[0099] Specifically, the software implementation of the plan acquisition module can be developed using Python, paired with a database to effectively store, manage, and retrieve data on multiple newly built road plans. The data sources can include planning and design documents, CAD files, GIS vector data, DEM data, etc., which are imported in batches through Web API interfaces or local files.

[0100] In terms of the software implementation of the map construction module, the Python language can be used in combination with graph computing libraries such as NetworkX or DGL to quickly convert the road plan data into a graph structure, complete the construction of nodes and edges, and express the topological relationship.

[0101] The implementation of the feature assignment module also uses the Python language. In data processing, libraries such as Pandas, NumPy, and GeoPandas can be integrated to complete the extraction and normalization of node feature data. For the processing of visual image features, computer vision libraries such as OpenCV and PyTorch can be integrated for image feature extraction and processing; for the preprocessing of historical traffic flow data and accident data, data cleaning and normalization techniques are adopted to form a unified data input format.

[0102] For the software implementation of the evaluation module, the PyTorch or TensorFlow framework can be used to develop, train, and deploy a graph neural network model. After pre-training and tuning, the model can achieve efficient inference on the GPU through CUDA technology, quickly calculate the accident risk of each node and the global risk index. The model evaluation results can be efficiently called through the RESTful API interface.

[0103] The software implementation of the comparison module is carried out using Python. Specifically, the Pandas library is used to sort and compare the risk indicators, and automatically select the plan with the lowest risk. The output result of this module directly generates an interactive chart that is easy to interpret (such as generated through the Plotly library), which is convenient for decision-makers to conduct intuitive comparative analysis.

[0104] The structure generation module can be implemented using a deep learning framework based on the Graph-Variational Autoencoder (Graph-VAE), using the Python language in combination with the PyTorch Geometric library to automatically generate the new road graph structure through the encoder and decoder structures.

[0105] The implementation method of the feedback optimization module is a feedback control module written in the Python language. By monitoring the risk results output by the evaluation module, and using reinforcement learning algorithms (such as DDPG or PPO) or optimization algorithms (such as Bayesian optimization) to dynamically adjust the latent space sampling strategy, generate model parameters, or path selection strategy, so as to achieve the closed-loop adaptive optimization of the model. During the operation of the system, the feedback optimization module regularly performs automatic parameter calibration to achieve the gradual optimization of the generated structure and the reduction of risk.

[0106] As described above, it is only the preferred specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art within the technical scope disclosed by the present invention, according to the technical solution and inventive concept of the present invention, makes equivalent substitutions or changes, and should be covered by the protection scope of the present invention.

Claims

1. A method for evaluating and selecting a newly built road based on a graph neural network, characterized in that, It includes the following steps: Obtain multiple new road construction plans, each plan including the layout, connection method, and design parameters of the new road; Model each newly built road construction plan as a graph structure ; Among them represents a set of road segments, including newly built road segments and existing road segments; represents the connection relationship between road segments; Assign features to the nodes of each graph structure. The node features of the existing roads include the existing features of the roads, and the node features of the new roads include the design features of the new roads; Use the trained graph neural network evaluation model to process each graph structure, predict the accident risk of each road section, and calculate the global risk index of each new road construction plan according to the predicted accident risk; Based on the global risk index, compare multiple new road construction plans and select the plan with the lowest risk as the optimal new road construction plan; Among them, the method for generating a new road construction plan includes: Take the starting point constraint conditions, ending point constraint conditions, environmental constraint conditions, and the existing road network graph structure as input information; Use a graph neural network encoder to jointly encode the input information to obtain a latent space representation vector; Input the latent space representation vector into a graph structure decoder to generate multiple candidate new road plans that meet the constraint conditions and can be connected to the existing road network, and each plan is represented by a graph structure.

2. The method according to claim 1, wherein The graph structure decoder is a graph variational autoencoder, specifically including: The encoding module is used to jointly encode the starting point constraint conditions, ending point constraint conditions, environmental constraint conditions, and the existing road structure into Gaussian distribution parameters in the latent space, and sample the latent space representation vector using the Gaussian distribution parameters; The decoding module gradually generates the nodes and connection relationships of the new road through the latent space representation vector to form a new road graph structure that can be topologically connected to the existing road network.

3. The method according to claim 1, wherein The method further includes fusing the generated new road graph structure with the existing road graph structure into a combined graph; Use the graph neural network evaluation model to perform risk prediction and evaluation on the combined graph; Take the prediction result as feedback information to optimize the latent space sampling strategy or the graph structure decoder parameters in the structure generation stage.

4. The method according to claim 3, characterized in that, The starting point constraint conditions, ending point constraint conditions, and environmental constraint conditions include at least one or more of the following: The geographical coordinates of the starting point and the ending point; The maximum allowable slope; The avoidance area; The minimum turning radius; The regional planning layer information.

5. The method according to claim 1, wherein The node features of the new road include any one or more of the following: The virtual road image, and the obtaining method is: construct a 3D road model according to the road design parameters and terrain elevation data; Render the virtual image of the road based on the 3D road model; The road features, including the road width, road length, curvature radius, and slope.

6. The method according to claim 1, characterized in that, The node features of the existing road section include image features, road features, historical traffic flow data, and historical accident data.

7. The method according to claim 1, characterized in that The calculation formula of the global risk index is: ； in, is the global risk index of the graph structure G, is the number of nodes in the graph structure G, For the The accident risk prediction value of each node, is the weight of the corresponding node.

8. A new road evaluation and selection system based on a graph neural network for implementing claim 1, characterized in that, It includes: The plan acquisition module is used to obtain multiple new road construction plans, each plan including the layout, connection method, and design parameters of the new road; A map building module, configured to model each newly-built road construction plan as a graph structure ; Among them represents a set of road segments, including newly built road segments and existing road segments; represents the connection relationship between road segments; The feature assignment module is used to assign features to the nodes of each graph structure. The node features of the existing roads include the existing features of the roads, and the node features of the new roads include the design features of the new roads; An evaluation module, including a trained graph neural network evaluation model, is used to process each graph structure, predict the accident risk of each road section, and calculate the global risk index of each new road construction plan according to the predicted accident risk; A comparison module is used to compare multiple new road construction plans based on the global risk index and select the plan with the lowest risk as the optimal new road construction plan.

9. The system according to claim 8, wherein The system further includes: A structure generation module is used to generate multiple candidate new road structure plan graphs based on the starting point constraint conditions, ending point constraint conditions, environmental constraint conditions of road construction and the existing road structure information; A feedback optimization module is used to use the global risk index output by the evaluation module as a feedback signal to optimize and update the potential space sampling strategy, generation model parameters or path selection strategy of the structure generation module, so as to realize the closed-loop self-optimization of structure generation and evaluation.

Images