A Big Data-Based Method for Impact Assessment of Urban Road Traffic Entrances and Exits

Abstract

This invention discloses a method for evaluating the impact of urban road traffic entrances and exits based on big data, belonging to the field of urban traffic management technology. The method includes: constructing traffic feature vectors; building an urban road traffic network topology map based on the traffic feature vectors, proposing a dynamic weighted graph embedding algorithm, and establishing a road network association mapping model; applying an attention-based graph neural network algorithm to the road network association mapping model to evaluate the impact factors of road nodes and quantitatively analyze the impact degree of road entrances and exits; fusing the impact factor evaluation results and constructing a multi-dimensional traffic impact evaluation model using a cross-domain ensemble learning method; and generating optimized decision-making suggestions for urban road traffic entrances and exits through an intelligent recommendation algorithm based on the performance evaluation results of the multi-dimensional traffic impact evaluation model, thereby achieving precise scheduling of traffic network nodes. This invention develops an intelligent recommendation algorithm based on reinforcement learning, providing reliable optimization suggestions for traffic management decisions.

Description

Technical Field

[0001] This invention relates to the field of urban traffic management technology, and in particular to a method for evaluating the impact of urban road traffic entrances and exits based on big data. Background Technology

[0002] With the acceleration of urbanization, urban road traffic networks are becoming increasingly complex. Traffic entrances and exits, as important node facilities, directly affect the operational efficiency of the entire road network through their layout and management. Traditional methods for assessing the impact of traffic entrances and exits mainly rely on static indicators and empirical judgments, which are insufficient to accurately depict the dynamic impact of entrances and exits on the surrounding road network and cannot adapt to the complex traffic flow characteristics under large-scale road networks.

[0003] Existing traffic impact assessment methods suffer from the following problems: First, the data acquisition channels are limited, making it difficult to comprehensively reflect the operational status of the road network; second, the evaluation indicator system is incomplete, lacking consideration of network resilience and propagation effects; third, the model structure is relatively simple, failing to fully utilize advanced algorithms such as deep learning to mine data features; and fourth, the evaluation results lack dynamic adaptability, making it difficult to provide timely and effective support for traffic management decisions. Summary of the Invention

[0004] In view of the problems of existing traffic entrance and exit impact assessment methods, such as limited data acquisition, incomplete indicator system, simple model structure and insufficient dynamic adaptability, this invention is proposed.

[0005] Therefore, the problem to be solved by this invention is how to construct a traffic entrance and exit impact assessment method based on multi-source data fusion, and extract dynamic features of the traffic network through deep learning algorithms to achieve accurate assessment and optimization decision-making on the impact of entrances and exits.

[0006] To solve the above-mentioned technical problems, the present invention provides the following technical solution:

[0007] In a first aspect, embodiments of the present invention provide a method for evaluating the impact of urban road traffic entrances and exits based on big data. This method includes acquiring raw urban road traffic data through multi-source heterogeneous sensors, preprocessing it, and labeling it with spatiotemporal features to form traffic feature vectors. The raw urban road traffic data includes vehicle traffic density, driving trajectory, and travel time. Based on the traffic feature vectors, an urban road traffic network topology map is constructed. A dynamic weighted graph embedding algorithm is proposed, and through multi-scale feature extraction, the road network structure features and connection weights are adjusted to establish a road network association mapping model. An attention-based graph neural network algorithm is applied to the road network association mapping model to evaluate the impact factors of road nodes, calculate road node centrality, network propagation, and dynamic reachability, and quantitatively analyze the degree of impact of road entrances and exits. The impact factor evaluation results are integrated, and a cross-domain ensemble learning method is used to construct a multi-dimensional traffic impact evaluation model, defining a quantitative indicator system. The quantitative indicator system includes traffic efficiency, network resilience, and propagation influence. Based on the performance evaluation results of the multi-dimensional traffic impact evaluation model, an intelligent recommendation algorithm is used to generate optimization decision suggestions for urban road traffic entrances and exits, completing precise scheduling of traffic network nodes.

[0008] As a preferred embodiment of the big data-based urban road traffic entrance and exit impact assessment method described in this invention, the formation of traffic feature vectors includes the following steps: deploying an urban road traffic sensing device network, integrating vehicle traffic density, driving trajectory, and travel time collected by the sensing device network into raw traffic data; cleaning the raw traffic data using an adaptive time window method, calculating the standard deviation of vehicle traffic density, the average speed of driving trajectory, and the fluctuation coefficient of travel time to generate a preliminary dataset; constructing a spatiotemporal feature matrix based on the preliminary dataset, dividing the road network using a hexagonal honeycomb grid, and extracting time-varying features of traffic flow, spatial distribution features of vehicle trajectory, and travel time variation features; normalizing the time-varying features of traffic flow, spatial distribution features of vehicle trajectory, and travel time variation features, using sparse principal component analysis to reduce dimensionality, extracting the main feature components, and constructing a multidimensional traffic feature vector; spatiotemporally labeling the multidimensional traffic feature vector, adding timestamps and spatial coordinate information based on geographic information grid coding to form the final traffic feature vector.

[0009] As a preferred embodiment of the big data-based urban road traffic entrance and exit impact assessment method described in this invention, the establishment of a road network association mapping model includes the following steps: constructing an initial road network topology map based on traffic feature vectors, setting road intersections as network nodes and road connection segments as network edges, and initializing the node attribute matrix and edge connection matrix; using a hierarchical clustering method to perform multi-scale partitioning of network nodes, constructing a three-layer road network structure, namely the arterial road layer, the secondary arterial road layer, and the branch road layer, recording the connection relationships between nodes at each level; extracting vehicle traffic density, driving trajectory, and travel time features from the traffic feature vectors, calculating the spatiotemporal association strength between nodes, and generating an initial weight matrix; and constructing a dynamic weighted graph embedding algorithm. The dynamic weighted graph embedding algorithm employs a multi-head attention mechanism to calculate the influence relationships between nodes, captures weight evolution features through a temporal information fusion module, and optimizes node vector representations by combining a contrastive learning strategy and residual connections to complete the dynamic modeling of the road network structure. The algorithm is applied to the three-layer road network structure to obtain the inter-layer node mapping relationships and construct a hierarchical road network representation. Network structure features, including node degree distribution, clustering coefficient, and path length distribution, are extracted to calculate the network topology features at each level. Based on the hierarchical road network representation and network topology features, an adaptive weight update mechanism is constructed to dynamically adjust network connection weights. Finally, the hierarchical road network representation, network topology features, and dynamic connection weights are integrated to establish a road network association mapping model.

[0010] As a preferred embodiment of the big data-based urban road traffic entrance and exit impact assessment method of the present invention, the road node impact factor assessment includes the following steps: extracting road node features based on a road network association mapping model, including spatial location information, connectivity information, and traffic flow information of the nodes, and constructing a node feature matrix; designing a temporal attention module based on a bidirectional gated cyclic unit to capture the temporal evolution law of node features and generate a dynamic temporal representation of the nodes; constructing a spatial attention layer to calculate the spatial dependency relationship between nodes, and fusing neighborhood information from different ranges to form a spatial association representation of the nodes; combining the temporal attention module and the spatial attention layer to construct a spatiotemporal attention graph neural network model to dynamically aggregate and propagate node features; and calculating road node impact factors based on the spatiotemporal attention graph neural network model. Node centrality indices, including degree centrality, betweenness centrality, and eigenvector centrality, are used to assess the importance of nodes in the network. A network propagation module is constructed based on a spatiotemporal attention graph neural network model to simulate the diffusion process of traffic flow in the road network, calculate the propagation efficiency and propagation range between nodes, and obtain network propagation indices. A dynamic accessibility assessment module is established based on a spatiotemporal attention graph neural network model to calculate the reachability time and reachability range of nodes based on real-time traffic conditions and historical data, and quantify the accessibility level of nodes. By integrating road node centrality indices, network propagation indices, and accessibility levels, a multilayer perceptron network is designed to calculate the comprehensive influence factor of road nodes. For road entrance and exit nodes, the influence of the comprehensive influence factor on the regional traffic network is quantitatively analyzed in combination with the surrounding road network structure characteristics and traffic flow characteristics.

[0011] As a preferred embodiment of the big data-based urban road traffic entrance and exit impact assessment method of the present invention, the method for constructing a multi-dimensional traffic impact assessment model using a cross-domain ensemble learning approach includes the following steps: feature decomposition of the comprehensive impact factors of road nodes, extraction of node dynamic features, structural features, and flow features, and construction of feature vector sets; designing a cross-domain feature mapping module to map data from different feature domains to a unified feature space, and using an autoencoder for feature alignment and dimensionality reduction; constructing a set of base learners, including random forests, gradient boosting trees, and deep neural networks, and training the feature vector sets in parallel; and integrating the prediction results of the base learners using a weighted voting strategy, with the weight coefficients determined through cross-validation optimization to obtain the fusion result. The system integrates prediction results; establishes a traffic efficiency evaluation index, which calculates the traffic efficiency score by comprehensively considering the average travel time of road segments, the rate of change of traffic flow, and the congestion delay index; constructs a network resilience evaluation index, which calculates the network resilience score by comprehensively analyzing the network connectivity index, the number of alternative routes in the road network, and the ability to withstand traffic flow fluctuations; establishes a propagation impact evaluation index, which calculates the propagation impact score by comprehensively assessing the scope of traffic flow diffusion, the speed of congestion propagation, and the duration of regional impact; normalizes the traffic efficiency score, network resilience score, and propagation impact score, and designs the analytic hierarchy process (AHP) to determine the index weights; based on the index scores and index weights, a multi-dimensional traffic impact assessment model is constructed, and a comprehensive evaluation result is output.

[0012] As a preferred embodiment of the big data-based urban road traffic entrance and exit impact assessment method of the present invention, the method for generating optimized decision-making suggestions for urban road traffic entrances and exits through an intelligent recommendation algorithm includes the following steps: Based on the evaluation results of a multi-dimensional traffic impact assessment model, a decision feature matrix is ​​constructed, including traffic efficiency score, network resilience score, and propagation influence score; a multi-objective optimization model is designed, taking maximizing traffic efficiency, improving network resilience, and controlling propagation influence as optimization objectives, and establishing a set of constraints; an intelligent recommendation algorithm based on reinforcement learning is constructed, employing a dual-loop adversarial structure optimization decision-making strategy to dynamically evaluate road entrance and exit scheduling schemes; wherein, the outer loop introduces adaptive encoding of traffic flow features. The system extracts dynamic features of the road network using a temporal attention mechanism, and performs dual sampling optimization by combining a prediction network and a value network in the inner loop. A decision candidate set generation module is established to generate multiple feasible traffic entrance / exit optimization schemes based on historical scheduling experience and real-time traffic conditions. Feasibility assessments are performed on the candidate schemes, calculating the execution cost and expected benefits of each scheme, and selecting the most cost-effective optimization schemes. A scenario simulation system is constructed to simulate and verify the selected optimization schemes, evaluating their impact on the regional traffic network. Based on the simulation results, a scheduling execution plan is designed, including time-segmented traffic flow control strategies and emergency response plans. The final traffic entrance / exit optimization decision recommendations are output, including specific scheduling times, scheduling methods, and supporting measures.

[0013] As a preferred embodiment of the big data-based urban road traffic entrance and exit impact assessment method of the present invention, the following steps are included in the optimization decision-making strategy using a dual-loop adversarial structure: Constructing an outer loop traffic flow feature adaptive encoder, extracting spatiotemporal features of traffic flow using a multi-layer convolutional network, adaptively adjusting feature weights, and outputting a dynamic feature representation of the road network after processing by an attention mechanism; Constructing an outer loop temporal attention module, performing multi-head self-attention processing on the extracted feature representation, and outputting a feature representation with temporal weights through time-step feature aggregation and long short-term memory units; Constructing an inner loop dual-network structure, combining state encoding, action generation, and reward prediction into a prediction network, and combining state evaluation and advantage calculation into a value network, forming an action value evaluation system; Constructing an adversarial training mechanism, outputting a scheduling strategy through a generator, obtaining a strategy feasibility score after evaluation by a discriminator, and guiding the strategy optimization direction based on adversarial losses; Constructing a multi-objective reward calculation system, converting traffic efficiency, network resilience, and propagation influence into reward signals, and training the strategy through experience replay and priority sampling; Constructing a strategy optimization process, iteratively optimizing the strategy and value network through dual sampling based on evaluation indicators and an adaptive learning rate, and outputting the final scheduling strategy.

[0014] The beneficial effects of this invention are as follows: This invention collects traffic data through multi-source heterogeneous sensors, establishes a complete evaluation index system, and employs advanced algorithms such as graph neural networks and reinforcement learning to achieve accurate assessment and dynamic optimization of the impact on traffic entrances and exits. Compared with existing technologies, it has the following advantages: First, it constructs a multi-level data collection and processing framework, improving the comprehensiveness and reliability of the data; second, it proposes a dynamic weighted graph embedding algorithm, accurately characterizing the time-varying features of the road network structure; third, it designs an impact factor evaluation method based on an attention mechanism, improving the accuracy of node importance assessment; fourth, it uses cross-domain ensemble learning to construct a multi-dimensional evaluation model, achieving a comprehensive assessment of traffic efficiency, network resilience, and propagation influence; fifth, it develops an intelligent recommendation algorithm based on reinforcement learning, providing reliable optimization suggestions for traffic management decisions. Experimental verification shows that this scheme outperforms traditional methods in both assessment accuracy and decision reliability, providing strong support for refined urban traffic management. Attached Figure Description

[0015] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0016] Figure 1This is a flowchart outlining the framework for a big data-based method for assessing the impact of urban road traffic entrances and exits.

[0017] Figure 2 A flowchart for constructing traffic feature vectors for a big data-based urban road traffic entrance and exit impact assessment method.

[0018] Figure 3 A flowchart illustrating the design of a dual-loop adversarial structure for a big data-based urban road traffic entrance and exit impact assessment method. Detailed Implementation

[0019] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0020] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0021] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in different places in this specification does not necessarily refer to the same embodiment, nor is it a single or selective embodiment that is mutually exclusive with other embodiments.

[0022] Example 1, referring to Figures 1-3 This is the first embodiment of the present invention, which provides a method for assessing the impact of urban road traffic entrances and exits based on big data. The framework flowchart is as follows. Figure 1 As shown, including,

[0023] S1: Obtain raw urban road traffic data through multi-source heterogeneous sensors, and perform preprocessing and spatiotemporal feature annotation to form traffic feature vectors.

[0024] The flowchart for constructing traffic feature vectors is as follows: Figure 2As shown, the specific steps are as follows: A network of urban road traffic sensing devices is deployed, including roadside millimeter-wave radar sensor arrays, infrared optical cameras, and vehicle-mounted positioning terminals. Vehicle traffic density, driving trajectories, and travel times collected by the sensing device network are integrated into raw traffic data. An adaptive time window method is used to clean the raw traffic data. The adaptive time window dynamically adjusts its size based on data fluctuations. The standard deviation of vehicle traffic density, the average speed of driving trajectories, and the fluctuation coefficient of travel time are calculated to generate a preliminary dataset. A spatiotemporal feature matrix is ​​constructed based on the preliminary dataset. The road network is divided using a hexagonal honeycomb grid, and time-varying features of traffic flow, spatial distribution features of vehicle trajectories, and travel time variation features are extracted. These features are normalized, and sparse principal component analysis is used for dimensionality reduction to extract the main feature components and construct a multidimensional traffic feature vector. The multidimensional traffic feature vector is spatiotemporally labeled, and timestamps and spatial coordinate information are added based on geographic information grid coding to form the final traffic feature vector.

[0025] Among the raw data for urban road traffic, vehicle density, travel trajectory, and travel time were selected as the core indicators because these three indicators can comprehensively reflect the road traffic operation status: vehicle density reflects the real-time saturation and congestion level of a road segment, travel trajectory reveals the spatial distribution patterns and route selection characteristics of vehicles, and travel time reflects the service level and traffic efficiency of the road. These three indicators complement and verify each other, reflecting the traffic flow operation status at a macro level and characterizing the behavior of individual vehicles at a micro level, providing a complete data foundation for subsequent traffic impact assessments.

[0026] S2: Based on traffic feature vectors, a topology map of urban road traffic network is constructed. A dynamic weighted graph embedding algorithm is proposed. Through multi-scale feature extraction, the structural features and connection weights of the road network are adjusted to establish a road network association mapping model.

[0027] Specifically, it includes the following steps:

[0028] S2.1: Construct an initial road network topology based on traffic feature vectors, set road intersections as network nodes, road connection segments as network edges, and initialize the node attribute matrix and edge connection matrix.

[0029] S2.2: A hierarchical clustering method is used to divide the network nodes into multiple scales and construct a three-layer road network structure, namely the main road layer, the secondary road layer and the branch road layer, and the connection relationship between the nodes at each level is recorded.

[0030] S2.3: Extract vehicle traffic density, travel trajectory, and travel time features from the traffic feature vector, calculate the spatiotemporal correlation strength between nodes, and generate an initial weight matrix.

[0031] S2.4: Construct a dynamic weighted graph embedding algorithm. The dynamic weighted graph embedding algorithm uses a multi-head attention mechanism to calculate the influence relationship between nodes, captures the weight evolution characteristics through a time-series information fusion module, and combines a contrastive learning strategy and residual connection to optimize the node vector representation, thereby completing the dynamic modeling of the road network structure.

[0032] Specifically, a multi-head attention computation unit is constructed to generate query features, key features, and value features from node features through linear transformation, and to calculate the attention weight coefficients between nodes. Multiple attention heads are set up, each independently calculating the influence relationships between nodes, resulting in multiple sets of attention feature representations. These multiple sets of features are concatenated and linearly mapped to obtain the multi-head attention output features. A temporal information fusion module is constructed, using a bidirectional long short-term memory network to process the historical feature sequences of nodes, extracting evolution patterns in the time dimension, and generating temporal dynamic features. The temporal dynamic features are fused with the multi-head attention output features, and a gating mechanism is used to adaptively adjust the weight ratio of the two types of features to obtain the fused features. A contrastive learning module is constructed to perform data augmentation on the fused features, generating two sets of feature representations that are mutually positive samples. The system optimizes feature representation by minimizing the contrastive loss function; it employs a deep feedforward neural network to perform nonlinear transformation on the fused features, generating intermediate features, and then adds the original features to the transformed features through residual connections to obtain the optimized node vector representation; it calculates the similarity of nodes within the main road layer, secondary road layer, and branch road layer respectively, and dynamically updates the connection weights between nodes within the layer based on node similarity; it constructs inter-layer mapping relationships, calculates the association strength between nodes at different levels through an attention mechanism, and establishes dynamic connection relationships between nodes between layers; it designs an adaptive weight update mechanism to dynamically adjust intra-layer and inter-layer weights according to changes in traffic flow, maintaining the time-varying characteristics of the network structure; and it integrates intra-layer node representations, intra-layer weights, and inter-layer mapping relationships to construct a complete dynamic road network representation, realizing dynamic modeling of the road network structure.

[0033] Preferably, compared with existing technologies, the dynamic weighted graph embedding algorithm proposed in this invention has significant advantages in road network structure modeling: by combining a multi-head attention mechanism and a temporal information fusion module, it achieves accurate modeling of the dynamic relationships between road network nodes, overcoming the limitation of traditional static graph embedding methods in capturing time-varying features; it innovatively introduces a contrastive learning strategy and a residual connection mechanism, improving the discriminativeness and stability of node vector representations; and it adopts a hierarchical modeling and adaptive weight update approach to accurately reflect the hierarchical relationships and dynamic changes between roads of different levels. Experimental results show that compared with traditional graph embedding methods, this scheme improves the accuracy of road network structure representation by 18%, significantly enhances adaptability in dynamic scenarios, and provides more reliable network structure features for subsequent traffic impact assessment.

[0034] S2.5: Apply the dynamic weighted graph embedding algorithm to the three-layer road network structure to obtain the inter-layer node mapping relationship and construct a hierarchical road network representation.

[0035] S2.6: Extract network structure features, including node degree distribution, clustering coefficient, path length distribution, and calculate the network topology features at each level.

[0036] S2.7: Based on hierarchical road network representation and network topology characteristics, an adaptive weight update mechanism is constructed to dynamically adjust network connection weights.

[0037] S2.8: Integrate hierarchical road network representation, network topology features, and dynamic connection weights to establish a road network association mapping model.

[0038] S3: Apply an attention-based graph neural network algorithm to the road network association mapping model to evaluate the impact factors of road nodes, calculate road node centrality, network propagation and dynamic reachability, and quantitatively analyze the impact of road entrances and exits.

[0039] Specifically, it includes the following steps:

[0040] S3.1: Extract road node features based on the road network association mapping model, including the spatial location information, connectivity information and traffic flow information of the nodes, and construct a node feature matrix.

[0041] S3.2: Design a temporal attention module based on bidirectional gated recurrent units to capture the temporal evolution of node features and generate dynamic temporal representations of nodes.

[0042] S3.3: Construct a spatial attention layer to calculate the spatial dependencies between nodes, integrate neighborhood information from different ranges, and form a spatial association representation of nodes.

[0043] S3.4: Combining the temporal attention module and the spatial attention layer, a spatiotemporal attention graph neural network model is constructed to dynamically aggregate and propagate node features.

[0044] Specifically, based on the dynamic temporal representation output by the temporal attention module and the spatial correlation representation output by the spatial attention layer, a dual-stream feature fusion layer is constructed, using adaptive gating units to dynamically adjust the fusion ratio of the two types of features; a graph convolutional layer is constructed, a learnable adjacency matrix is ​​designed, node features are spatially aggregated, local structural information is extracted, and batch normalization is used to enhance feature representation capabilities; a spatiotemporal attention computation module is designed, inputting the temporal feature sequence and spatial feature sequence into independent attention sub-networks to generate attention weights in the temporal and spatial dimensions; a multi-layer graph convolutional network structure is constructed, with each layer containing residual connections and skip connections, gradually expanding the receptive field by stacking multiple graph convolutional layers to capture multi-scale node interaction features; a feature aggregation module is designed, adaptively weighting and aggregating the features output by different layers of graph convolutions, using attention mechanisms. The importance weights of features at each layer are determined; an edge feature update mechanism is introduced to dynamically adjust the edge weight coefficients based on the evolution of node features, thereby achieving dynamic updates of the network topology; a spatiotemporal feature decoupling module is constructed to decompose the fused features into temporal and spatial feature components, facilitating the extraction of relevant feature information in subsequent tasks; a multi-task learning head is designed, including a node classification head, a link prediction head, and a temporal prediction head, sharing the underlying feature extraction network to enhance the model's feature representation capability; the model is trained in an end-to-end manner, and a joint loss function is designed, including node classification loss, link prediction loss, and temporal prediction loss, with model parameters optimized through backpropagation; an inference module is constructed to apply the trained spatiotemporal attention graph neural network model to road node impact factor evaluation, providing basic feature support for subsequent centrality calculation, propagation analysis, and reachability assessment.

[0045] S3.5: Calculate road node centrality indices based on the spatiotemporal attention graph neural network model, including degree centrality, betweenness centrality, and eigenvector centrality, to evaluate the importance of nodes in the network.

[0046] S3.6: Based on the spatiotemporal attention graph neural network model, a network propagation module is constructed to simulate the diffusion process of traffic flow in the road network, calculate the propagation efficiency and propagation range between nodes, and obtain network propagation index.

[0047] S3.7: A dynamic accessibility assessment module is established based on a spatiotemporal attention graph neural network model. Based on real-time traffic conditions and historical data, the module calculates the reachability time and range of nodes and quantifies the accessibility level of nodes.

[0048] S3.8: Integrate road node centrality indicators, network propagation indicators, and accessibility levels to design a multilayer perceptron network and calculate the comprehensive influence factor of road nodes.

[0049] S3.9: For road entrance and exit nodes, combine the surrounding road network structure characteristics and traffic flow characteristics to quantitatively analyze the degree of influence of comprehensive influencing factors on the regional transportation network.

[0050] Preferably, compared with existing technologies, the attention-based graph neural network algorithm proposed in this invention has unique advantages in evaluating the impact factors of road nodes: by designing a spatiotemporal attention graph neural network model that fuses dual-flow features, it innovatively combines the temporal feature extraction of bidirectional gated recurrent units with the fusion of neighborhood information for spatial attention, achieving accurate capture of the dynamic evolution features of road nodes; the architecture design of multi-layer graph convolutional networks and feature aggregation modules significantly improves the ability to express multi-scale node interaction features; through dynamic edge feature updates and spatiotemporal feature decoupling mechanisms, the model can adaptively adjust the network topology, more accurately reflecting the dynamic changes of the actual traffic network. Experimental results show that compared with traditional static network analysis methods, this scheme improves the accuracy of node importance assessment by 25% and the accuracy of propagation effect prediction by 30%, and can effectively handle the complex spatiotemporal dependencies of large-scale road networks, providing more reliable technical support for the refined management of urban traffic networks.

[0051] S4: Integrating the assessment results of impact factors, a multi-dimensional traffic impact assessment model is constructed using a cross-domain ensemble learning method, and a quantitative indicator system is defined. The quantitative indicator system includes traffic efficiency, network resilience, and dissemination influence.

[0052] Specifically, it includes the following steps:

[0053] S4.1: Perform feature decomposition on the comprehensive influencing factors of road nodes, extract node dynamic features, structural features and flow features, and construct feature vector groups.

[0054] S4.2: Design a cross-domain feature mapping module to map data from different feature domains to a unified feature space, and use an autoencoder for feature alignment and dimensionality reduction.

[0055] Specifically, a feature preprocessing layer is constructed to standardize node dynamic features, structural features, and flow features respectively, eliminating scale differences between different feature domains; an independent feature encoder is designed, with a corresponding encoding network structure for each type of feature, mapping the original features to the latent feature space; a feature alignment layer is constructed, employing adversarial learning to reduce distribution differences between different feature domains, and constraining the distribution consistency of latent features through a discriminator network; an autoencoder network is designed, consisting of an encoder and a decoder, where the encoder compresses the aligned features to a low-dimensional space, and the decoder restores the compressed features to the original space; a reconstruction loss function is introduced, minimizing the difference between the input features and the reconstructed features. To ensure information preservation during feature compression, a feature fusion module is designed, employing an attention mechanism to adaptively fuse low-dimensional representations from different feature domains, generating a unified feature representation. A feature enhancement layer is constructed, enriching feature expression and improving the model's feature extraction capability through residual connections and feature combinations. A feature selection mechanism is designed, selecting the most representative feature combinations based on feature importance scores to reduce feature redundancy. A domain adaptation module is established, improving feature transfer capability by minimizing the maximum average difference between the source and target domains. All processing modules are integrated to construct an end-to-end feature mapping network, outputting standardized feature representations in a unified feature space, providing high-quality input data for subsequent base learner training.

[0056] S4.3: Construct a set of base learners, including random forests, gradient boosting trees, and deep neural networks, and train the feature vector sets in parallel.

[0057] S4.4: A weighted voting strategy is used to integrate the prediction results of the base learners. The weight coefficients are determined by cross-validation optimization to obtain the fused prediction results.

[0058] S4.5: Establish traffic efficiency evaluation indicators, and obtain traffic efficiency scores by comprehensively calculating the average travel time of road segments, traffic flow change rate, and congestion delay index.

[0059] S4.6: Construct network resilience evaluation indicators. The network resilience score is obtained through a comprehensive analysis of network connectivity index, number of alternative routes in the road network, and traffic flow fluctuation tolerance.

[0060] S4.7: Establish a communication impact evaluation index, and obtain a communication impact score through a comprehensive assessment of the traffic flow diffusion range, congestion spread speed and regional impact duration.

[0061] S4.8: Normalize the scores for traffic efficiency, network resilience, and dissemination influence, and design the analytic hierarchy process (AHP) to determine the weights of the indicators.

[0062] S4.9: Based on indicator scores and indicator weights, construct a multi-dimensional traffic impact assessment model and output comprehensive evaluation results.

[0063] Preferably, compared with existing technologies, the cross-domain ensemble learning method proposed in this invention has significant advantages in constructing a multi-dimensional traffic impact assessment model: Through feature decomposition and cross-domain feature mapping modules, it achieves unified expression and alignment of different feature domains (dynamic features, structural features, and flow features), overcoming the evaluation bias caused by feature heterogeneity in traditional methods; it employs a combination of autoencoders and adversarial learning for feature dimensionality reduction and alignment, ensuring sufficient preservation of feature information and inter-domain consistency; through parallel training and weighted ensemble of multiple base learners, it significantly improves the model's generalization ability and prediction stability; and it innovatively constructs a multi-dimensional evaluation system encompassing traffic efficiency, network resilience, and propagation influence, comprehensively reflecting the integrated impact of traffic entrances and exits. Experimental verification shows that compared with traditional single-model evaluation methods, the prediction accuracy of this scheme is improved by 20%, and it exhibits good transferability under different urban road network conditions, providing a more reliable evaluation basis for urban traffic management decisions.

[0064] S5: Based on the performance evaluation results of the multi-dimensional traffic impact assessment model, generate optimization decision suggestions for urban road traffic entrances and exits through intelligent recommendation algorithms to achieve precise scheduling of traffic network nodes.

[0065] Specifically, it includes the following steps:

[0066] S5.1: Based on the evaluation results of the multi-dimensional traffic impact assessment model, construct a decision feature matrix, including traffic efficiency score, network resilience score, and dissemination influence score.

[0067] S5.2: Design a multi-objective optimization model, taking maximizing traffic efficiency, improving network resilience, and controlling propagation influence as optimization objectives, and establish a set of constraints.

[0068] S5.3: Construct an intelligent recommendation algorithm based on reinforcement learning, and adopt a dual-loop adversarial structure to optimize the decision-making strategy and dynamically evaluate the road entrance and exit scheduling scheme.

[0069] The outer loop introduces a traffic flow feature adaptive encoder and a temporal attention mechanism to extract dynamic features of the road network, while the inner loop combines a prediction network and a value network for dual sampling optimization.

[0070] Specifically, the flowchart for the dual-loop antagonistic structure design is as follows: Figure 3As shown, the implementation includes: constructing an adaptive encoder for outer loop traffic flow features, extracting spatiotemporal features of traffic flow using a multi-layer convolutional network, adaptively adjusting feature weights, and outputting a dynamic feature representation of the road network after processing by an attention mechanism; constructing an outer loop temporal attention module, performing multi-head self-attention processing on the extracted feature representation, and outputting a feature representation with temporal weights through time-step feature aggregation and long short-term memory units; constructing an inner loop dual-network structure, combining state encoding, action generation, and reward prediction into a prediction network, and combining state evaluation and advantage calculation into a value network, forming an action value evaluation system; constructing an adversarial training mechanism, outputting scheduling strategies through a generator, obtaining strategy feasibility scores after evaluation by a discriminator, and guiding the strategy optimization direction based on adversarial loss; constructing a multi-objective reward calculation system, converting traffic efficiency, network resilience, and propagation influence into reward signals, and training strategies through experience replay and priority sampling; and constructing a strategy optimization process, iteratively optimizing the strategy and value network through double sampling based on evaluation indicators and an adaptive learning rate, and outputting the final scheduling strategy.

[0071] Compared with existing technologies, the reinforcement learning algorithm with a dual-loop adversarial structure proposed in this invention has significant advantages: by introducing an adaptive encoder and temporal attention mechanism in the outer loop, it can accurately capture the dynamic spatiotemporal characteristics and long-term dependencies of traffic flow, improving the accuracy of feature extraction; through the dual evaluation mechanism of the inner loop prediction network and value network, combined with adversarial training and multi-objective reward design, it not only improves the feasibility and robustness of the scheduling strategy, but also achieves balanced optimization in multiple dimensions such as traffic efficiency, network resilience, and propagation influence; by employing mechanisms such as experience replay and priority sampling, it significantly improves the training efficiency and convergence speed of the algorithm. Experimental results show that compared with traditional single-network structure reinforcement learning methods, this scheme improves the accuracy of the scheduling strategy by 15%, the convergence speed by 30%, and exhibits better generalization ability and adaptability under complex road conditions.

[0072] S5.4: Establish a decision candidate set generation module to generate multiple feasible traffic entrance and exit optimization schemes based on historical scheduling experience and real-time traffic conditions.

[0073] S5.5: Conduct a feasibility assessment of the candidate solutions, calculate the execution cost and expected benefits of each solution, and select the most cost-effective optimization solution.

[0074] S5.6: Construct a scenario simulation system to simulate and verify the selected optimization schemes and evaluate their impact on the regional transportation network.

[0075] S5.7: Design scheduling and execution schemes based on simulation results, including time-segmented traffic flow control strategies and emergency response plans.

[0076] S5.8: Output the final traffic entrance and exit optimization decision recommendations, including specific scheduling time, scheduling method and supporting measures.

[0077] In summary, this invention collects traffic data using multi-source heterogeneous sensors, establishes a complete evaluation index system, and employs advanced algorithms such as graph neural networks and reinforcement learning to achieve accurate assessment and dynamic optimization of the impact on traffic entrances and exits. Compared with existing technologies, it has the following advantages: First, it constructs a multi-level data collection and processing framework, improving the comprehensiveness and reliability of the data; second, it proposes a dynamic weighted graph embedding algorithm, accurately characterizing the time-varying features of the road network structure; third, it designs an impact factor evaluation method based on an attention mechanism, improving the accuracy of node importance assessment; fourth, it uses cross-domain ensemble learning to construct a multi-dimensional evaluation model, achieving a comprehensive assessment of traffic efficiency, network resilience, and propagation influence; and fifth, it develops an intelligent recommendation algorithm based on reinforcement learning, providing reliable optimization suggestions for traffic management decisions. Experimental verification shows that this scheme outperforms traditional methods in both assessment accuracy and decision reliability, providing strong support for refined urban traffic management.

[0078] Example 2, refer to Figures 1-3 This is the second embodiment of the present invention, which provides a method for evaluating the impact of urban road traffic entrances and exits based on big data. In order to verify the beneficial effects of the present invention, scientific demonstration is carried out through economic benefit calculation and simulation experiments.

[0079] To verify the effectiveness of the proposed big data-based urban road traffic entrance and exit impact assessment method, the research team selected the traffic network of a core area of ​​a megacity as the experimental scenario. The experiment covered an area of ​​approximately 50 square kilometers in the city center, comprising 3 arterial road layers, 5 secondary arterial road layers, and 15 local road layers, totaling over 200 road nodes. Data acquisition employed a multi-source heterogeneous sensor network, deploying 32 roadside millimeter-wave radar sensors (Bosch MMR3 24GHz), 48 high-resolution infrared optical cameras (Hikvision DS-2CD3T36 series), and over 300 vehicle-mounted BeiDou / GPS positioning terminals. Data acquisition spanned seven consecutive days, covering weekdays and weekends, with a sampling frequency of once every 5 minutes, resulting in the collection of over 50TB of raw traffic data.

[0080] In the data preprocessing stage, the research team applied the adaptive time window method for the first time to meticulously clean the raw traffic data. By dynamically adjusting the window size, outlier data points were effectively removed. In the standard deviation calculation, the fluctuation coefficient of vehicle traffic density was controlled within ±15%, the average speed deviation of driving trajectory was controlled within ±10km / h, and the fluctuation coefficient of travel time was controlled within ±12%. After processing, the final effective dataset coverage reached 98.7%, providing a high-quality data foundation for subsequent analysis.

[0081] Table 1 shows a comparison of the performance of graph embedding algorithms. This table details the comparison between the dynamic weighted graph embedding algorithm proposed in this invention and the traditional static graph embedding method in terms of the accuracy of road network structure representation.

[0082] Table 1 Performance Comparison of Dynamic Weight Graph Embedding Algorithms

[0083] Evaluation indicators Traditional static graph embedding methods This invention Improvement rate Structure representation accuracy 69.1％ 89.1％ 18％ Dynamic scene adaptability Low high - Time-varying feature capture capability weak powerful -

[0084] The attention-based graph neural network algorithm demonstrates superior performance in evaluating node influence factors. Through multi-scale analysis of road networks, the algorithm achieves an accuracy of 89.5% in assessing node importance, a 25% improvement over traditional static network analysis methods. In predicting propagation effects, this method achieves 87.3% accuracy, a 30% improvement over previous methods.

[0085] In the construction phase of the multi-dimensional traffic impact assessment model, the cross-domain ensemble learning method demonstrated significant advantages. Through a unified mapping of dynamic features, structural features, and flow characteristics, the model's prediction accuracy reached 79.6%, a 20% improvement over traditional single-model evaluation methods. In transferability tests under different urban road network conditions, the model maintained good stability and adaptability.

[0086] The reinforcement learning-based intelligent recommendation algorithm also demonstrates remarkable performance in traffic entrance and exit optimization decisions. Compared to traditional reinforcement learning methods with a single network structure, this approach improves scheduling strategy accuracy by 15% and convergence speed by 30%. In complex road condition simulations, the algorithm exhibits excellent generalization ability and can quickly respond to dynamic changes in the traffic environment.

[0087] Ultimately, through a comprehensive assessment of the urban core area's transportation network, the research team not only verified the technical feasibility of the proposed method but also provided a practical technical path for refined urban traffic management. This research breaks through the limitations of traditional traffic impact assessment methods and provides important theoretical support and technical reference for future intelligent transportation construction.

[0088] To fully demonstrate the advanced nature of the technical solution of this invention, the research team designed Table 2, a performance comparison table of technical solutions. This table compares the performance differences between the prior art and the method of this invention from multiple dimensions.

[0089] Table 2 Comparison of Overall Performance of Technical Solutions

[0090] Evaluation indicators Existing technology This invention Comprehensiveness of data collection single sensor Multi-source heterogeneous sensor networks Road network structure modeling Static graph embedding Dynamic weighted graph embedding algorithm Node impact factor assessment Traditional network analysis Attention-based graph neural network Traditional effect prediction Single indicator assessment Multidimensional traffic impact assessment model Decision-making intelligent recommendation Experience-driven Reinforcement learning intelligent recommendation algorithm Overall assessment accuracy lower Multi-dimensional, dynamic, intelligent

[0091] Table 2 comprehensively compares the performance differences between the technical solution of this invention and existing technologies in the field of urban traffic evaluation. Through multi-dimensional evaluation, the research team demonstrated the significant advantages of this invention in key aspects such as data collection, road network modeling, node analysis, propagation prediction, and intelligent decision-making. Compared with traditional technologies, this solution achieves significant improvements in data comprehensiveness, model accuracy, and intelligent recommendation capabilities, providing a more intelligent, dynamic, and efficient technical solution for refined urban traffic management.

[0092] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A method for assessing the impact of urban road traffic entrances and exits based on big data, characterized in that: include, Raw urban road traffic data is acquired through multi-source heterogeneous sensors, and then preprocessed and labeled with spatiotemporal features to form a traffic feature vector; the raw urban road traffic data includes vehicle traffic density, driving trajectory, and travel time. Based on the traffic feature vectors, an urban road traffic network topology map is constructed. A dynamic weighted graph embedding algorithm is proposed. Through multi-scale feature extraction, the road network structure features and connection weights are adjusted to establish a road network association mapping model. The road network association mapping model is applied to a graph neural network algorithm based on an attention mechanism to evaluate the road node influence factors, calculate the road node centrality, network propagation and dynamic reachability, and quantitatively analyze the degree of influence of road entrances and exits; By integrating the assessment results of impact factors, a multi-dimensional traffic impact assessment model is constructed using a cross-domain ensemble learning method, and a quantitative indicator system is defined; the quantitative indicator system includes traffic efficiency, network resilience, and dissemination influence. Based on the performance evaluation results of the multi-dimensional traffic impact assessment model, intelligent recommendation algorithms are used to generate optimization decision-making suggestions for urban road traffic entrances and exits, thereby achieving precise scheduling of traffic network nodes. The establishment of the road network association mapping model includes the following steps: Construct an initial road network topology based on traffic feature vectors, set road intersections as network nodes, road connection segments as network edges, and initialize node attribute matrices and edge connection matrices. A hierarchical clustering method was used to divide network nodes into multiple scales and construct a three-layer road network structure, namely the main road layer, the secondary road layer and the branch road layer, and the connection relationship between nodes at each level was recorded. Extract vehicle traffic density, travel trajectory, and travel time features from the traffic feature vector, calculate the spatiotemporal correlation strength between nodes, and generate an initial weight matrix; A dynamic weighted graph embedding algorithm is constructed. The algorithm uses a multi-head attention mechanism to calculate the influence relationship between nodes, captures the weight evolution characteristics through a temporal information fusion module, and optimizes the node vector representation by combining a contrastive learning strategy and residual connection to complete the dynamic modeling of the road network structure. The dynamic weighted graph embedding algorithm is applied to the three-layer road network structure to obtain the inter-layer node mapping relationship and construct a hierarchical road network representation. Extract network structure features, including node degree distribution, clustering coefficient, and path length distribution, and calculate the network topology features at each level; Based on the hierarchical road network representation and the network topology features, an adaptive weight update mechanism is constructed to dynamically adjust the network connection weights. By integrating the hierarchical road network representation, network topology features, and dynamic connection weights, a road network association mapping model is established. The assessment of road node impact factors includes the following steps: Based on the road network association mapping model, road node features are extracted, including the spatial location information, connectivity information and traffic flow information of the nodes, and a node feature matrix is ​​constructed. Design a temporal attention module based on bidirectional gated recurrent units to capture the temporal evolution of node features and generate dynamic temporal representations of nodes; Construct a spatial attention layer to calculate the spatial dependencies between nodes, integrate neighborhood information from different ranges, and form a spatial association representation of nodes; By combining the temporal attention module and the spatial attention layer, a spatiotemporal attention graph neural network model is constructed to dynamically aggregate and propagate node features; The spatiotemporal attention graph neural network model is used to calculate road node centrality indices, including degree centrality, betweenness centrality, and eigenvector centrality, to evaluate the importance of nodes in the network. Based on the spatiotemporal attention graph neural network model, a network propagation module is constructed to simulate the diffusion process of traffic flow in the road network, calculate the propagation efficiency and propagation range between nodes, and obtain network propagation index. A dynamic accessibility assessment module is established based on the spatiotemporal attention graph neural network model. Based on real-time traffic conditions and historical data, the reachability time and reachability range of nodes are calculated, and the accessibility level of nodes is quantified. By integrating the road node centrality index, network propagation index, and accessibility level, a multilayer perceptron network is designed to calculate the comprehensive influence factor of road nodes. For road entrance and exit nodes, the impact of comprehensive influencing factors on the regional transportation network is quantitatively analyzed in combination with the characteristics of the surrounding road network structure and traffic flow. The method of constructing a multi-dimensional traffic impact assessment model using cross-domain ensemble learning includes the following steps: The comprehensive influencing factors of road nodes are decomposed into features to extract dynamic features, structural features and flow features of the nodes, and feature vector sets are constructed. Design a cross-domain feature mapping module to map data from different feature domains to a unified feature space, and use an autoencoder for feature alignment and dimensionality reduction; Construct a set of base learners, including random forest, gradient boosting tree and deep neural network, and train the feature vector set in parallel; A weighted voting strategy is used to integrate the prediction results of the base learners, and the weight coefficients are determined by cross-validation optimization to obtain the fused prediction results. Establish traffic efficiency evaluation indicators, and obtain traffic efficiency scores by comprehensively calculating the average travel time of road segments, traffic flow change rate, and congestion delay index. A network resilience evaluation index is constructed, and a network resilience score is obtained through a comprehensive analysis of the network connectivity index, the number of alternative routes in the road network, and the ability to withstand traffic flow fluctuations. Establish a communication impact evaluation index, and obtain a communication impact score through a comprehensive assessment of the traffic flow diffusion range, congestion spread speed and regional impact duration; The traffic efficiency score, network resilience score, and propagation influence score are normalized, and the weights of the indicators are determined by the analytic hierarchy process. Based on indicator scores and indicator weights, a multi-dimensional traffic impact assessment model is constructed, and a comprehensive evaluation result is output.

2. The method for assessing the impact of urban road traffic entrances and exits based on big data as described in claim 1, characterized in that: The process of forming traffic feature vectors includes the following steps: A network of urban road traffic sensing devices is deployed, and the vehicle traffic density, driving trajectory, and travel time collected by the sensing device network are integrated into raw traffic data. The raw traffic data is cleaned using an adaptive time window method, and the standard deviation of vehicle traffic density, the average speed of the driving trajectory, and the fluctuation coefficient of travel time are calculated to generate a preliminary dataset. Based on the preliminary dataset, a spatiotemporal feature matrix is ​​constructed, and a hexagonal honeycomb grid is used to divide the road network to extract time-varying features of traffic flow, spatial distribution features of vehicle trajectories, and features of changes in travel time. The time-varying characteristics of traffic flow, spatial distribution characteristics of vehicle trajectory, and characteristics of travel time variation are normalized, and sparse principal component analysis is used to reduce dimensionality, extract the main feature components, and construct a multidimensional traffic feature vector. The multidimensional traffic feature vector is spatiotemporally labeled, and timestamps and spatial coordinate information are added based on geographic information grid coding to form the final traffic feature vector.

3. The method for assessing the impact of urban road traffic entrances and exits based on big data as described in claim 1, characterized in that: The process of generating optimization decision-making suggestions for urban road traffic entrances and exits using intelligent recommendation algorithms includes the following steps: Based on the assessment results of the multi-dimensional traffic impact assessment model, a decision feature matrix is ​​constructed, including traffic efficiency score, network resilience score, and propagation impact score. A multi-objective optimization model is designed, taking maximizing traffic efficiency, improving network resilience, and controlling propagation influence as optimization objectives, and a set of constraints is established. A reinforcement learning-based intelligent recommendation algorithm is constructed, and a dual-loop adversarial structure is adopted to optimize the decision-making strategy and dynamically evaluate the road entrance and exit scheduling scheme. The outer loop introduces a traffic flow feature adaptive encoder and a temporal attention mechanism to extract dynamic features of the road network, while the inner loop combines a prediction network and a value network for dual sampling optimization. Establish a decision candidate set generation module to generate multiple feasible traffic entrance and exit optimization schemes based on historical scheduling experience and real-time traffic conditions; Feasibility assessments are conducted on candidate solutions, the execution costs and expected benefits of each solution are calculated, and the most cost-effective optimization solution is selected. A scenario simulation system was constructed to conduct simulation verification of the selected optimization schemes and to evaluate the impact of the optimization schemes on the regional transportation network. Based on simulation results, a scheduling execution plan is designed, including time-segmented traffic flow control strategies and emergency response plans; Output the final traffic entrance and exit optimization decision recommendations, including specific scheduling time, scheduling method and supporting measures.

4. The method for assessing the impact of urban road traffic entrances and exits based on big data as described in claim 3, characterized in that: The optimization decision-making strategy employing a dual-loop adversarial structure includes the following steps: An adaptive encoder for external circulation traffic flow features is constructed. Multi-layer convolutional networks are used to extract the spatiotemporal features of traffic flow. The feature weights are adaptively adjusted, and the output is a dynamic feature representation of the road network after processing by an attention mechanism. An outer loop temporal attention module is constructed to perform multi-head self-attention processing on the extracted feature representations. Through temporal step feature aggregation and long short-term memory units, a feature representation with temporal weights is output. Construct an inner-loop dual-network structure, combining state encoding, action generation, and reward prediction into a prediction network, and state evaluation and advantage calculation into a value network, to form an action value evaluation system; An adversarial training mechanism is constructed, which outputs scheduling strategies through a generator, obtains a strategy feasibility score through evaluation by a discriminator, and guides the strategy optimization direction based on the adversarial loss. A multi-objective reward calculation system is constructed, which transforms traffic efficiency, network resilience, and propagation influence into reward signals, and the strategy is trained through experience replay and priority sampling. A policy optimization process is constructed. Based on the evaluation metrics and adaptive learning rate, the policy and value network are iteratively optimized through double sampling to output the final scheduling policy.

Images