A smart traffic congestion relieving method and system for urban road network

Abstract

The present application belongs to the field of urban traffic management and control technology, and particularly relates to a smart traffic congestion dredging method and system for urban road network, which comprises: a multi-source data sensing layer for collecting implicit bottleneck space-time characteristic data, driver interactive behavior data and rail passenger flow coupling correlation data of a target road network; a coupling analysis layer for constructing a coupling analysis model based on the multi-source heterogeneous data output by the sensing layer; a hierarchical risk including at least a single implicit congestion risk, an implicit congestion and behavior game superposition risk, and a rail superposition congestion risk; a hierarchical regulation layer configured to output the hierarchical risk type according to the coupling analysis layer; and a feedback optimization layer for collecting multi-dimensional traffic state feedback data after the dredging instruction is executed. The present application can effectively reveal the hidden physical bottleneck by fusing the road infrastructure state, driver micro interactive behavior, rail traffic macro passenger flow and other multi-source heterogeneous data, and constructing a coupling analysis model.

Description

Technical Field

[0001] This invention belongs to the field of urban traffic management and control technology, specifically relating to a smart traffic congestion relief method and system for urban road networks. Background Technology

[0002] The stable operation of modern urban transportation systems is crucial for supporting socio-economic development and efficient public travel. However, with the continuous growth of motor vehicle ownership and increasingly diversified travel demands, urban road networks are under immense pressure. Traffic congestion is not only manifested in obvious phenomena such as decreased vehicle speed and increased queue lengths, but its causes are often deeply rooted in complex and dynamic factors such as the road's physical environment, driver behavior and psychology, and the interaction of passenger flow between different modes of transportation. These factors change independently yet influence each other, making the evolution of traffic conditions highly nonlinear and uncertain, posing unprecedented challenges to the depth of perception, analytical intelligence, and coordinated response of traffic management technologies.

[0003] Currently, urban traffic congestion management primarily relies on the combination of information technology and automatic control technology. At the perception level, technologies such as geomagnetic induction coils, microwave detectors, and video image recognition are commonly used to collect macroscopic traffic flow parameters such as traffic volume, average speed, and occupancy rate at road cross-sections in real time. At the analysis and decision-making level, mainstream technologies include signal timing scheme optimization based on historical data, single-point or trunk-line adaptive signal control systems based on real-time traffic flow, and regional traffic status assessment and dissemination platforms integrating GPS floating car data. These existing technologies often rely on pre-set fixed models and respond passively to single points through numerous isolated systems. In terms of risk identification, they often remain at the level of "post-event" description of explicit, already occurring congestion, leading to delayed early warnings, inaccurate root cause diagnosis, limited overall efficiency, and an inability to adapt to the dynamic and complex evolutionary characteristics of traffic systems, making it difficult to continuously improve long-term management effectiveness. Summary of the Invention

[0004] To address the problems existing in the prior art, the purpose of this invention is to provide a smart traffic congestion management method and system for urban road networks, which can adapt to dynamic changes in traffic, achieve continuous improvement in long-term management efficiency, and effectively solve the shortcomings of the prior art in adapting to complex evolutionary characteristics.

[0005] The technical solution of this invention is: A smart traffic congestion mitigation method for urban road networks includes the following steps: Collect spatiotemporal characteristic data of hidden bottlenecks in the target road network, driver interaction behavior data, and rail passenger flow coupling correlation data, and perform preprocessing. Based on the preprocessed spatiotemporal characteristic data of hidden bottlenecks, driver interaction behavior data, and rail passenger flow coupling correlation data, a coupling analysis model is constructed to determine the comprehensive risk level. Based on the comprehensive risk classification value and the preset risk assessment parameters, the classification risk types caused and superimposed by the spatiotemporal characteristics of hidden road bottlenecks, driver interaction behavior data and passenger flow coupling correlation data of rail network are assessed. Based on the assessed risk level, a matching graded collaborative diversion instruction set is generated, and each diversion instruction in the graded collaborative diversion instruction set is distributed to the corresponding traffic control system execution terminal to alleviate traffic congestion. The method for constructing the coupling analysis model includes: Based on the preprocessed spatiotemporal characteristic data of hidden bottlenecks, driver interaction behavior data, and rail passenger flow coupling correlation data, the hidden bottleneck interference degree index, traffic flow disorder coefficient, and rail supply-demand imbalance ratio of the target section at the evaluation time are obtained respectively. Based on the degree of influence corresponding to the hidden bottleneck interference degree index, traffic flow disorder coefficient, and rail supply-demand imbalance ratio, their corresponding dynamic weight coefficients are determined and weighted fusion calculations are performed to obtain the comprehensive graded risk value.

[0006] Preferably, the method further includes a feedback optimization step: collecting multi-dimensional traffic status feedback data after the execution of the traffic diversion instructions, and dynamically calibrating the risk assessment parameters of the coupled analysis model and the instruction triggering threshold of the hierarchical collaborative traffic diversion instruction set based on the multi-dimensional traffic status feedback data through an online learning mechanism.

[0007] Preferably, the steps for assessing graded risks include: Based on driver interaction behavior data, the interaction decision-making process between drivers is simulated, and the traffic flow disorder coefficient is output to quantitatively characterize the traffic flow stability risk caused by driver interaction behavior. The driver interaction behavior data includes the frequency of cutting in, following distance, and number of lane changes. Driving style labels are generated by clustering historical driving data. The driving style labels include at least aggressive, conservative, and neutral types. Based on the coupled data of rail passenger flow, a time-series prediction model is used to estimate the distribution of exiting passenger flow at subway stations within a target time period. This is then combined with a matching analysis of the dynamic capacity of the associated road network to quantify and determine the risk of congestion caused by the superposition of rail passenger flow. The matching analysis includes converting the estimated exiting passenger flow at subway stations into an equivalent number of vehicles and comparing this equivalent number of vehicles with the dynamic capacity to obtain a supply-demand imbalance ratio that characterizes the supply-demand relationship of the rail system. The traffic flow disorder coefficient and the supply-demand imbalance ratio are used as core quantitative indicators. They are input together with the hidden bottleneck interference degree index into the weighted fusion algorithm of the coupled analysis model to calculate the comprehensive graded risk value and complete the graded risk assessment including "hidden congestion and behavioral game superposition risk" and "railway superposition congestion risk".

[0008] Preferably, the comprehensive risk level is determined according to the following formula: , , , , In the formula, This is a comprehensive risk classification value; Index of the degree of interference from hidden bottlenecks; The traffic flow turbulence coefficient represents the degree of impact of driver interaction. To meet the basic traffic needs of the road section; This refers to the equivalent number of vehicles generated from the passenger flow exiting the subway station. This item is for the dynamic traffic capacity of the road section. The overall characterization of the risks associated with supply and demand imbalances under track coupling; , , Each risk item is in Dynamic weighting coefficients at any given time; For the preprocessed first The original quantization parameters of the spatiotemporal characteristics of a hidden bottleneck; This represents the total number of latent bottleneck spatiotemporal characteristics; For the first The weights of the spatiotemporal features of the hidden bottleneck; This represents the total number of driver behavior status categories. For each state category The probability of its occurrence; To predict passenger flow exiting subway stations; The percentage of passengers choosing motor vehicle travel when exiting the station; The average number of occupants in a motor vehicle; This is a correction factor.

[0009] Preferably, the hierarchical collaborative guidance instruction set includes: Level 1 control instructions, targeting a single hidden congestion risk, include at least using signal control to create dynamic lane allocation to avoid bottleneck areas; The secondary control order addresses the combined risks of hidden congestion and behavioral game theory. Based on the primary control order, it adds targeted intervention for high-risk driving behaviors and fine-tuning of local traffic signal parameters. The Level 3 control order addresses the risk of congestion on rail lines by initiating cross-transportation coordinated scheduling based on the Level 2 control order. This includes coordinated control of subway passenger flow, surface connecting transport capacity, and traffic flow on related road networks.

[0010] Preferably, the steps for dynamically calibrating parameters using an online learning mechanism include: Collect multi-dimensional traffic status feedback data after the execution of the traffic diversion command, and quantify the collected data into measurable system status indicators. The indicators include at least the changes in average vehicle speed on road sections, changes in queue length at intersections, and actual traffic flow disorder coefficients generated after the execution of the hierarchical coordinated traffic diversion command. The dynamic weighting coefficients that directly affect the calculation results of the comprehensive risk level in the coupled analysis model, and the risk level thresholds on which the triggering conditions of different levels of instructions in the hierarchical collaborative guidance instruction set depend, are collectively defined as the action parameters to be optimized and adjusted in the learning framework. Based on quantified system state indicators, a reward function is designed and a reward value is obtained; the value of the reward function is positively correlated with the increase in average vehicle speed, the reduction in queue length, and the decrease in traffic flow disorder coefficient in the next cycle. Based on the reward value, the reinforcement learning framework is driven to periodically iteratively optimize the defined action parameters; the coupled analysis model and the hierarchical collaborative guidance instruction set are updated based on the optimized parameters.

[0011] Preferably, it also includes adaptive risk response steps in the event of extreme weather or emergencies, specifically including: Real-time monitoring of meteorological data or emergency alarm information; when extreme weather conditions or emergencies of a specific level are detected, the parameter adaptive adjustment mode is automatically triggered. Based on the currently monitored scene characteristics, historical traffic data and optimized parameter sets for similar scenarios are matched from the historical database; the parameter set includes at least the risk assessment parameters of the coupled analysis model calibrated for similar scenarios and the instruction triggering thresholds of the hierarchical collaborative traffic management instruction set. In the parameter adaptive adjustment mode, the parameter set is used as the initial strategy of the online learning mechanism. During the duration of this scenario, the online learning mechanism continues to be executed, and the risk assessment parameters and instruction trigger thresholds are quickly fine-tuned using real-time collected multi-dimensional traffic status feedback data to dynamically respond to the superimposed risks of hidden bottleneck deterioration, driving behavior distortion and traffic demand fluctuation caused by extreme weather or sudden events.

[0012] A smart traffic congestion mitigation system for urban road networks, used to implement any of the methods described above, includes a multi-source data sensing layer, a coupling analysis layer, a hierarchical control layer and a feedback optimization layer connected in sequence; The multi-source data perception layer is used to collect multi-source heterogeneous data of the target road network and perform preprocessing. The multi-source heterogeneous data includes latent bottleneck spatiotemporal feature data, driver interaction behavior data, and rail passenger flow coupling correlation data. The coupling analysis layer is used to construct a coupling analysis model based on the preprocessed multi-source heterogeneous data, obtain a comprehensive graded risk value, and, in combination with preset risk assessment parameters, assess the graded risks caused and superimposed by the spatiotemporal characteristic data of road hidden bottlenecks, driver interaction behavior data, and rail network passenger flow coupling correlation data, respectively. The graded risk types include at least single hidden congestion risk, hidden congestion and behavioral game superposition risk, and rail superposition congestion risk. The hierarchical control layer is used to generate a hierarchical collaborative diversion instruction set that matches the hierarchical risk type output by the coupling analysis layer, and distribute each diversion instruction in the diversion instruction set to the corresponding traffic control system execution terminal. The feedback optimization layer is used to collect multi-dimensional traffic status feedback data after the execution of traffic diversion instructions. Based on the multi-dimensional traffic status feedback data, the risk assessment parameters of the coupled analysis model and the instruction triggering threshold of the hierarchical collaborative traffic diversion instruction set are dynamically calibrated through an online learning mechanism.

[0013] Preferably, the multi-source data perception layer includes lidar, stress sensors, and video detectors deployed on road infrastructure for collecting multi-source heterogeneous data, as well as an edge computing node network for processing multi-source heterogeneous data; the traffic control system execution terminal includes a traffic signal controller, variable message sign, vehicle guidance terminal, and public transportation operation scheduling platform, for receiving and executing hierarchical collaborative guidance instruction sets.

[0014] Preferably, the coupling analysis layer includes a latent bottleneck assessment unit, a behavioral game simulation unit, and a track coupling assessment unit; the latent bottleneck assessment unit uses a multi-index fusion method based on information entropy to calculate the latent bottleneck entropy value, which is used to assess the risk of a single latent congestion; the behavioral game simulation unit simulates driver interaction decisions based on game rules, which is used to assess the risk of latent congestion and behavioral game superposition; the track coupling assessment unit uses a time-series prediction model to perform passenger flow and road network matching analysis, which is used to assess the risk of track superposition congestion.

[0015] Compared with existing technologies, the intelligent traffic congestion mitigation method and system for urban road networks of the present invention has the following beneficial effects: This invention constructs a coupled analysis model by collecting multi-source heterogeneous data, including the spatiotemporal characteristics of hidden bottlenecks, driver interaction behavior, and rail passenger flow coupling. This model can accurately assess the risk levels, ranging from single hidden congestion to multiple superimposed risks, thereby proactively identifying the root causes of congestion and overcoming the limitations of traditional passive responses. The generated hierarchical collaborative traffic management instruction set implements coordinated interventions, from dynamic lane allocation to cross-transit system scheduling, for different risk levels, ensuring the accuracy and efficiency of regulation. Furthermore, through the online learning mechanism of the feedback optimization layer, the model parameters and instruction thresholds are dynamically calibrated to form an iterative optimization traffic management mechanism that can adapt to dynamic traffic changes and achieve continuous improvement in long-term management efficiency, effectively solving the shortcomings of existing technologies that cannot adapt to complex evolutionary characteristics. Attached Figure Description

[0016] Figure 1 This is a flowchart of the dredging method in an embodiment of the present invention; Figure 2 This is a structural block diagram of the dredging method in an embodiment of the present invention. Detailed Implementation

[0017] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0018] Based on the embodiments of this invention, all other embodiments obtained by those skilled in the art without inventive effort are within the scope of protection of this invention.

[0019] Furthermore, the technical solutions of the various embodiments of the present invention can be combined with each other, but only if they are feasible for those skilled in the art. If the combination of technical solutions is contradictory or cannot be implemented, it should be considered that such combination of technical solutions does not exist and is not within the scope of protection claimed by the present invention.

[0020] See Figure 1 As shown, in order to adapt to dynamic changes in traffic and achieve continuous improvement in long-term management efficiency, this embodiment effectively solves the shortcomings of existing technologies that cannot adapt to complex evolution characteristics. It provides a smart traffic congestion mitigation system for urban road networks, including a multi-source data perception layer, a coupling analysis layer, a hierarchical control layer and a feedback optimization layer connected in sequence. The multi-source data perception layer is used to collect multi-source heterogeneous data of the target road network, including spatiotemporal characteristic data of hidden bottlenecks, driver interaction behavior data, and rail passenger flow coupling correlation data. The coupling analysis layer constructs a coupling analysis model based on the multi-source heterogeneous data output by the multi-source data perception layer, and presets risk assessment parameters corresponding to the multi-source heterogeneous data in the coupling analysis model. Based on the coupling analysis model and the preset risk assessment parameters, it assesses the graded risks caused and superimposed by the spatiotemporal characteristic data of hidden road bottlenecks, driver interaction behavior data, and rail passenger flow coupling correlation data, with the graded risk types including at least... It includes single hidden congestion risk, hidden congestion and behavioral game-related risk combined risk, and rail-based congestion risk; the hierarchical control layer is configured to generate a matching hierarchical collaborative diversion instruction set based on the hierarchical risk type output by the coupling analysis layer, and distribute each diversion instruction in the diversion instruction set to the corresponding traffic control system execution terminal (i.e., external traffic system); the feedback optimization layer collects multi-dimensional traffic state feedback data after the diversion instructions are executed, and dynamically calibrates the risk assessment parameters of the coupling analysis model and the instruction trigger threshold of the hierarchical collaborative diversion instruction set through an online learning mechanism based on the multi-dimensional traffic state feedback data.

[0021] Furthermore, the multi-source data perception layer includes lidar, stress sensors, and video detectors deployed on road infrastructure for collecting multi-source heterogeneous data, as well as an edge computing node network for processing multi-source heterogeneous data; the traffic control system execution terminal includes traffic signal controllers, variable message signs, vehicle guidance terminals, and public transportation operation scheduling platforms, which are used to receive and execute hierarchical collaborative guidance instruction sets.

[0022] Furthermore, the coupling analysis layer includes a latent bottleneck assessment unit, a behavioral game simulation unit, and a track coupling assessment unit. The latent bottleneck assessment unit employs a multi-index fusion method based on information entropy to comprehensively calculate core feature parameters, generating latent bottleneck entropy values ​​for each road segment. When the latent bottleneck entropy value of a road segment exceeds a dynamically adjusted threshold, a single latent congestion risk is triggered. The behavioral game simulation unit, based on preset game rules and driver interaction behavior data, simulates driver interaction decisions in a virtual traffic scenario containing latent bottlenecks, outputting a traffic flow turbulence coefficient characterizing the risk of traffic flow stability. The behavioral game simulation unit simulates differentiated decision-making strategies for drivers with different driving styles in high-risk scenarios. The track coupling assessment unit uses a time-series prediction model to perform passenger flow and road network matching analysis. Specifically, it predicts the distribution of subway exit passenger flow in future periods based on the time-series prediction model and matches it with the dynamic capacity of the associated road network to determine the risk of overlapping congestion due to track passenger flow.

[0023] Based on the above system setup, this invention provides a smart traffic congestion mitigation method for urban road networks, comprising the following steps: Step 1: Collect multi-source heterogeneous data of the target road network, including spatiotemporal characteristic data of hidden bottlenecks, driver interaction behavior data, and rail passenger flow coupling correlation data. Hidden bottleneck spatiotemporal characteristic data includes data on abrupt changes in road slope, manhole cover distribution, and interference from non-motorized vehicle traffic. Driver interaction behavior data includes cut-in frequency, following distance, lane change frequency, and driving style tags. Rail passenger flow coupling correlation data includes real-time traffic flow data of subway station entry and exit passenger flow and surrounding road network connections.

[0024] Step 2: Construct a coupling analysis model based on the collected multi-source heterogeneous data. Method for constructing the coupling analysis model:

[0025] (1) Data source definition and unified spatiotemporal benchmark The coupling analysis model of this invention uses urban road networks as its basis, and the input multi-source heterogeneous data includes at least: ① Road network operation data: vehicle trajectory / floating car speed, road segment traffic flow, occupancy rate, queue length, signal phase and timing, road segment topology, etc.; ② Driver interaction behavior data: lane change / cut in / sudden braking / following time, navigation route preference, detour selection, travel time preference, etc. (can be extracted from vehicle terminal, navigation platform or sensing device); ③ Coupling data of rail passenger flow: passenger flow entering and exiting rail stations, transfer intensity, attraction / evacuation intensity of roads around stations, arrival and departure timetables, etc.

[0026] To achieve cross-source fusion, a unified spatiotemporal reference must first be established: Time alignment: Mapping data with different sampling frequencies to a unified time window (e.g., 30s / 1min / 5min), missing segments are filled using interpolation, moving average, or Kalman filtering; Spatial alignment: Mapping trajectory points, stations, events, etc., to the road network map. The nodes / edges are used to complete the attribution of "data → road segment / intersection / station service area".

[0027] (2) Heterogeneous data cleaning and feature construction (multi-channel feature tensor) The aligned data is cleaned and characterized to form a multi-channel feature set for coupled modeling: Road segment / intersection operation characteristics : velocity, flow rate, density, saturation, phase pressure, queue growth rate, etc.; Behavioral characteristics Lane change intensity, lane cutting probability, path deviation rate, and car-following instability index, etc. Track coupling characteristics : Station passenger flow intensity, entrance and exit tidal direction, transfer event trigger intensity, station-road segment impact coefficient, etc.; Topological features : Road segment length, number of lanes, speed limit, adjacent connectivity, importance centrality, etc.

[0028] Latent bottleneck spatiotemporal characteristics The following factors are considered: the intensity of abrupt changes in the longitudinal profile of the road section, the density of manhole covers, the proportion of non-motorized vehicles mixed in the road, and the speed fluctuation interference index.

[0029] The above features in each time window The following can be represented as: , (3) Coupling relationship modeling: Constructing a "multi-layer coupling graph / hypergraph" To characterize the combined effects of "hidden bottlenecks, behavioral disturbances, and track coupling" on congestion evolution, this invention constructs a multi-layered coupling structure. At least including: ① Road topology coupling layer Using road segments as edges and intersections as nodes (or road segments as nodes), the edge weights reflect the adjacent propagation intensity (which can be estimated by historical correlation coefficients, shock wave propagation speed, and queue overflow probability).

[0030] ② Behavior affects the coupling layer A graph is constructed using the causal or correlational relationship of "behavioral event → change in road segment status", such as an increase in lane change intensity leading to speed fluctuations in adjacent road segments.

[0031] ③ Track coupling layer The station and surrounding road sections form a bipartite graph or hyperedge structure, and the changes in passenger flow at the station are mapped to the demand disturbance of the road sections.

[0032] The final result is a multi-layered coupled structure: , (4) The core of the coupled analysis model: fusion inference and state prediction Based on multi-channel features With coupling structure Construct a coupling analysis model The output should include at least: Road congestion status estimation (Congestion level, queuing risk, fluctuation intensity); Hidden bottleneck strength (Risks that are not overtly congested but have the conditions to trigger them); Forecasting of transmission trends ( (for predicting step size) One implementation involves encoding different data channels separately (linear transformation, temporal convolution, recurrent networks, or attention mechanisms can all be used as alternatives) to obtain the hidden representations for each channel. And hidden bottleneck exclusive hidden representation Subsequently, these implicit representations are concatenated or interact with through self-attention at the node (road segment / intersection) dimension, and then based on a multi-layered coupling structure... To aggregate and integrate:

[0033] , , in, : Indicates within a time window Internally, it is the comprehensive latent representation (i.e., high-dimensional feature vector) output after the feature fusion of multi-source heterogeneous data. The implicit representation obtained after extracting the macroscopic traffic operation characteristics (such as vehicle speed, flow rate, density, etc.) of a road segment or intersection through a coding network; : This represents the hidden representation obtained after the driver's micro-interaction behavior characteristics (such as lane change intensity, cut-in probability, etc.) are extracted by the coding network; : This represents the implicit representation obtained after the coupling and correlation characteristics of rail passenger flow (such as the intensity of exiting passenger flow, conversion rate, etc.) are extracted by the coding network; : This represents the specific latent representation obtained after the implicit bottleneck spatiotemporal features (such as slope change intensity, mixed traffic interference index, etc.) are extracted by the coding network; : Represents a constructed multi-layered coupled graph or hypergraph structure, which includes road topology associations, propagation paths of behavioral influences, and track supply and demand mapping relationships; : Represents the future output of the system The prediction results at any given time, namely the estimated traffic congestion status or the superimposed risk level; : Indicates the time step for predicting the future; by introducing The model can strongly couple road physical defects with micro traffic flow fluctuations, thereby accurately calculating single and superimposed risks.

[0034] Specifically, the mathematical expression and calculation process of the coupling analysis model are as follows: First, the implicit representations of each channel are concatenated at the node (road segment or intersection) level, and the initial fusion features of each node are obtained through linear mapping. : In the formula, This represents the concatenation operation of feature vectors; and These are the learnable input weight matrix and the bias term, respectively; It is a non-linear activation function.

[0035] Subsequently, based on the multi-layer coupling structure Feature aggregation and fusion are performed, and the specific expression for spatial aggregation is as follows: In the formula, For the first The implicit representation of the layer, when hour, That is, it is composed of each node The initial fusion feature matrix is ​​formed; A self-loop coupling structure adjacency matrix is ​​added to characterize the relationship between road topology, behavioral influences, and track coupling. This is the corresponding degree matrix; For the first Layer weight parameters. After... After propagation through the layered graph network, the final synthesized hidden representation is output. .

[0036] Ultimately, the prediction of dissemination trends The specific expression is: In the formula, and These are the prediction weight matrix and bias term of the output layer, respectively; The function is used to output the graded probability prediction results of traffic congestion status.

[0037] Furthermore, in order to map the prediction results output by the aforementioned deep learning model to the specific physical operation mechanism of traffic flow and overcome the "black box" defect of a simple neural network, the coupled analysis model constructs a weighted fusion evaluation formula with explicit physical meaning, which is used to calculate the comprehensive graded risk value that ultimately triggers the coordinated instruction. : In the formula, The comprehensive risk level output by the system assessment has a value range of [0,1] and is used to trigger the corresponding level of graded collaborative evacuation instructions; The normalized index of the degree of hidden bottleneck interference represents the impact of a single hidden congestion risk. The normalized traffic flow turbulence coefficient characterizes the degree of influence of driver interaction. To meet the basic traffic needs of the road section, This refers to the equivalent number of vehicles generated from the passenger flow exiting the subway station. This item is for the dynamic traffic capacity of the road section. The overall characterization of the risks associated with supply and demand imbalances under track coupling; , , Each risk item is in The dynamic weight coefficients at each time step are determined by the comprehensive implicit representation output by the aforementioned graph network. Mapping generation.

[0038] Through this fusion formula, the system will capture the high-dimensional hidden features ( ) from the graph network. This is transformed into dynamic weights that regulate physical traffic flow indicators, realizing a dual-drive coupling of data-driven (AI prediction) and mechanism model (traffic flow theory).

[0039] (5) Model training / calibration and online update The parameter acquisition of the coupled analysis model includes two stages: offline calibration and online updating. Offline phase: Use historical data and congestion labels (generated by speed thresholds, delays, queue lengths or manual annotations) to train model parameters to minimize prediction error; Online Phase: Receive feedback data in real time (speed recovery rate after adjustment, queue dissipation time, etc.), and use online learning or reinforcement learning to optimize key parameters (such as...). Risk threshold, edge weight Dynamic calibration is performed to adapt to unstable scenarios such as holidays, emergencies, and construction.

[0040] Finally, the risk assessment parameters for multi-source heterogeneous data are configured in the trained coupled analysis model. These risk assessment parameters include at least the following: Latent bottleneck spatiotemporal risk parameters: 1. Weighting of Multiple Indicators: When calculating the "hidden bottleneck entropy value," weights are assigned to different characteristic indicators such as "abrupt changes in road slope," "density of manhole cover distribution," and "degree of interference from non-motorized vehicle traffic." These weights determine the magnitude of each factor's impact on the final risk value.

[0041] 2. Baseline value and adjustment coefficient of dynamic threshold: The entropy threshold used to trigger "single hidden congestion risk" (such as 0.65 in Case 1) is managed by the feedback optimization layer, and its baseline value and adjustment rules under different times and weather conditions are managed by the feedback optimization layer.

[0042] Driver interaction behavior risk parameters: 1. Benefit / Cost Coefficients in Driver Game Rules: Values ​​set in the model for virtual benefits or costs such as "saving time," "avoiding collisions," and "psychological stress" when simulating driver decisions such as cutting in, following other vehicles, and changing lanes.

[0043] 2. Behavioral preference parameters for different driving styles: For "aggressive", "conservative" and "neutral" drivers, the model defines the probability distribution parameters of their specific decisions (such as forced lane changes or conservative following) in risky scenarios.

[0044] 3. Calculation formula for traffic flow disorder coefficient: The weight of each behavioral indicator (such as the frequency of cutting in and the variance of following distance) when aggregating micro-level game behavior into the macro-level "traffic flow disorder coefficient".

[0045] Risk parameters related to the coupling of rail passenger flow: 1. Passenger-vehicle conversion coefficient: The coefficient used to convert the predicted number of subway exit passengers into the equivalent number of road vehicles, which directly affects the accuracy of demand forecasting.

[0046] 2. Attenuation factor in dynamic capacity calculation model: When calculating the "remaining capacity" that takes into account real-time traffic conditions and the impact of hidden bottlenecks, the capacity reduction factor caused by various disturbance factors (such as weather and accidents) is used.

[0047] 3. Risk assessment threshold for supply and demand imbalance: As mentioned in Example 3 of the document, the ratio of "equivalent number of vehicles" to "remaining capacity" (88%) is compared with a preset threshold (80%). This threshold is the key parameter.

[0048] Step 3: Based on the coupling analysis model and its configured risk assessment parameters, assess the graded risks caused and superimposed by the spatiotemporal characteristics of road hidden bottlenecks, driver interaction behavior data, and passenger flow coupling correlation data of rail network. The graded risk types include at least single hidden congestion risk, hidden congestion and behavioral game superposition risk, and rail superposition congestion risk.

[0049] Step 4: Based on the assessed risk level, generate a matching graded collaborative traffic management instruction set, and distribute each traffic management instruction in the graded collaborative traffic management instruction set to the corresponding traffic control system execution terminal (external traffic system) to manage traffic congestion.

[0050] The tiered and coordinated control instruction set includes: Level 1 control instructions, Level 2 control instructions, and Level 3 control instructions. Specific control methods:

[0051] 1. For single hidden congestion risks, implement Level 1 control instructions, including at least using signal control to form dynamic lane allocation to avoid bottleneck areas; 2. To address the combined risks of hidden congestion and behavioral game theory, implement secondary control instructions, which, based on primary control instructions, include targeted interventions for high-risk driving behaviors and fine-tuning of local traffic signal parameters; 3. In response to the risk of congestion caused by overlapping rail lines, a three-level control order will be implemented. Based on the two-level control order, cross-transportation coordinated scheduling will be initiated, including the coordinated control of subway passenger flow, ground connecting transport capacity and related road network traffic.

[0052] Triggering conditions and decision-making logic of the hierarchical collaborative guidance instruction set: Risk indicators and grading thresholds: The risk values ​​output by the aforementioned risk assessment unit are denoted as... Or, combined with the remaining capacity ratio: , in .

[0053] In the formula, The ratio of remaining capacity; For road section In time The remaining capacity; = For road section In time Dynamic passage capability; Section In time Traffic demand (number of vehicles / time); This is the risk value.

[0054] Risk level definition (example): Based on risk value : L0 (No Risk): .

[0055] L1 (Low Risk): .

[0056] L2 (Medium Risk): .

[0057] L3 (High Risk): .

[0058] L4 (Extremely High / Emergency): .

[0059] Step 5: Collect multi-dimensional traffic status feedback data after the execution of traffic diversion instructions. Based on this data, dynamically calibrate the risk assessment parameters of the coupled analysis model and the instruction trigger thresholds of the hierarchical collaborative traffic diversion instruction set through an online learning mechanism. The multi-dimensional traffic status feedback data includes at least the following categories: 1. Macro-level traffic flow efficiency indicators: average vehicle speed, travel time / delay; 2. Micro-level driving behavior and safety indicators: frequency of high-risk behaviors (e.g., number of times vehicles decelerate suddenly, cutting in line), traffic flow stability (traffic flow stability is actual performance data related to the traffic flow disorder coefficient); 3. Indicators of the effectiveness and compliance rate of control instruction execution: compliance rate of guidance information, capacity scheduling efficiency, and lane management effectiveness; 4. System collaborative operation status indicators: cross-transportation scheduling matching degree (interconnection efficiency data between subway passenger flow, connecting buses, and road network traffic flow), and intersection performance after signal control parameter adjustment (i.e., the dissipation of vehicle queue length at upstream intersections after fine-tuning the signal).

[0060] Furthermore, the spatiotemporal characteristic data of hidden bottlenecks mainly includes data on abrupt slope changes, manhole cover distribution, and interference from non-motorized vehicle traffic on the monitored road sections. The steps for collecting this data include: real-time monitoring of abrupt slope changes, manhole cover distribution, and interference from non-motorized vehicle traffic using roadside sensors; and extracting core feature parameters by real-time processing of sensor data using edge computing nodes. Based on the data on abrupt slope changes, manhole cover distribution, and interference from non-motorized vehicle traffic, a set of hidden bottleneck factor parameters is calculated according to preset quantification rules. The implicit bottleneck interference level index is obtained through normalization and weighted fusion. Among them, the slope change parameter is determined by the slope change intensity of the longitudinal profile of the road segment. The calculation of non-motorized vehicle traffic interference parameters is jointly measured by the proportion of traffic mixing, speed fluctuation, and collision frequency; the corresponding thresholds can be set using rule-based thresholds or adaptive thresholds based on historical data quantiles. Details are as follows:

[0061] Definition: Implicit Bottleneck Interference Index (IBI): For each road segment In the time window Internally, define the index of the degree of interference from hidden bottlenecks: , : No. The original quantitative parameters of the spatiotemporal characteristics of a hidden bottleneck (such as abrupt slope changes, mixed-line interference, etc.).

[0062] Normalized parameters (e.g., mapped to) ).

[0063] Weights can be dynamically calibrated through historical calibration or online learning (without being limited to a specific learning algorithm).

[0064] Normalization method: .

[0065] The lower / upper limit of this parameter can be set by historical data quantiles, industry standard recommendations, or engineering experience.

[0066] in, The objective weights of each hidden bottleneck factor are calculated using a multi-index fusion method based on information entropy. This calculation process directly adapts the weight allocation based on the inherent characteristics of the data structure of multi-source heterogeneous data, avoiding subjective bias. The specific steps are as follows:

[0067] 1. Constructing a data matrix and standardization Assume the system is in a time window Internal monitoring of target road network Each road segment is extracted. The original data matrix consists of several implicit bottleneck parameters (such as abrupt slope changes, manhole cover distribution density, and the degree of interference from non-motorized vehicle traffic). ,in For the first The first section of the road One parameter.

[0068] The matrix is ​​standardized to obtain the normalized parameter matrix. .in For the first The first section of the road One parameter.

[0069] Calculate the first The first section of the road The proportion of features under each bottleneck factor Introducing a minimum constant To prevent logarithms from being meaningless: , 2. Calculate the information entropy of each indicator. Calculate the first Information entropy of hidden bottleneck factors : , Information entropy The smaller the value, the more likely it is that the first... The greater the difference in the degree of latent factors between different road sections, the more information they provide.

[0070] 3. Calculate the information utility value and objective weight. Calculate the first Information utility value of each factor Based on this, the final weights of each hidden bottleneck factor are determined. :

[0071] , Calculate the objective weights of each item Substitute into the aforementioned formula The road segment can then be determined. In the time window The comprehensive index of the degree of interference of hidden bottlenecks (i.e., the hidden bottleneck entropy value).

[0072] The original quantitative parameters of each of the aforementioned hidden bottleneck factors (i.e., in the formula) The specific quantitative definition and measurement rules are as follows: Define slope and slope abrupt change: Distance along the centerline of the road segment Elevation is .slope Defined as:

[0073] , Select length as A sliding window (e.g., 50m~200m, which can be set according to the road grade) defines the slope change intensity. : , in For window Local average of the inner slope.

[0074] Judgment threshold: when At that time, it was determined that the road section had a hidden bottleneck factor of "sudden slope change", and the following was taken: , In the formula, Indicates road segment In the time window The parameter values ​​of the hidden bottleneck factors caused by the sudden change in slope due to internal factors.

[0075] threshold Setting rules: Data-driven approach: Using historical road segment slope abrupt change intensity sets The quantile setting threshold, for example , (That is, take the 85% / 90% percentile to ensure that the threshold automatically adapts to the urban terrain) Measurement standards for "non-motorized vehicle traffic interference" Mixed traffic interference index: In the time window Inside, on the road section definition: Non-motorized vehicle traffic (vehicles / hour or vehicles / minute).

[0076] Traffic flow of motor vehicles.

[0077] Mixed banking ratio ( To prevent the denominator from being divided by a constant (to prevent division by zero).

[0078] Standard deviation of motor vehicle speed (reflecting disturbance fluctuations).

[0079] : Number of conflicts (e.g., the number of events with TTC less than a threshold, from the video trajectory).

[0080] Define the interfering intensity: , in These are calibration coefficients.

[0081] Threshold determination: ⇒ There is a hidden bottleneck of "mixed traffic interference". threshold Take history The 85% / 90% quantiles were used as the threshold (adaptive to city differences).

[0082] Furthermore, the steps for assessing graded risks include: simulating the driver interaction decision-making process in a virtual traffic scenario containing hidden bottlenecks based on pre-defined game rules and driver interaction behavior data, and outputting a traffic flow turbulence coefficient to characterize traffic flow stability risk; driver interaction behavior data includes data such as cut-in frequency, following distance, lane change frequency, and risk event frequency, and driving style labels are generated through historical driving data clustering. These driving style labels, generated by analyzing historical driving data clustering, include at least aggressive, conservative, and neutral styles. Specifically, after standardizing the acquired data such as cut-in frequency, following distance, lane change frequency, and risk event frequency, unsupervised clustering using K-means / GMM / DBSCAN is used to form style clusters, and each cluster is mapped to conservative, stable, or aggressive driving style labels based on cluster center characteristics or intensity score quantile rules. Specifically, the definition and calculation method of the traffic flow turbulence coefficient are as follows:

[0083] (1) Mathematical definition of traffic flow turbulence coefficient: within a uniform time window Inside, on the road section Collect vehicle trajectory sequences The micro-state of traffic flow is discretized into several "state categories," which may consist of, for example, the following discrete variables:

[0084] Lane status: .

[0085] Speed ​​binning: Divided into intervals kind.

[0086] Timing of the front of the vehicle: (or following distance) Divided into intervals kind.

[0087] Accelerometer bins: Divided into intervals kind.

[0088] For each vehicle The states within are statistically analyzed to obtain the state categories. probability of occurrence : , In the formula, Indicates a unified time window Inner section The statistical results show that the categories are in a specific state. The number of vehicles (or the frequency of occurrence); This indicates the total frequency of vehicles in all status categories within this time period and road segment.

[0089] Based on this, the traffic flow disorder coefficient is defined. (Normalized Shannon entropy): , in This represents the total number of driver behavior status categories. The larger the value, the more dispersed and disordered the traffic flow is, and the higher the stability risk.

[0090] Threshold setting method: When The situation is assessed as "traffic flow stability risk". Threshold This can be determined in the following ways: rule setting: for example Data adaptation: based on historical stable operating conditions The distribution is taken from quantiles, such as or .

[0091] How to generate driving style tags: (1) Input Data and Sample Construction: Based on historical driving data (e.g., vehicle trajectory, CAN / OBD, navigation interaction, ADAS events, or behavioral events extracted from video / radar), a sample set is constructed using "vehicle-time window" or "driver-trip" as the sample unit. Let the first... The feature vector corresponding to each sample is: Time window is acceptable. =30s / 1min / 5min (choose one or more scales).

[0092] (2) Feature selection (example feature set, which can be added or removed according to data conditions): In order to characterize the differences in driving style, statistical features are extracted from speed, acceleration and deceleration, following, lane changing and risk events to form a feature vector. .

[0093] One feasible feature selection method is as follows (including at least some of the following): ① Speed ​​and Fluctuation Category: Average Speed Speed ​​standard deviation velocity variation coefficient ② Acceleration / Deceleration and Intensity: Average Acceleration Acceleration standard deviation ; Jerk intensity of acceleration: or Number of rapid accelerations / decelerations: , .

[0094] ③ Following distance and safety margin (if information of the vehicle in front is available): Average headway Coefficient of variation over time Minimum TTC (Time-to-Collision) or number of times TTC is less than the threshold: .

[0095] ④ Lane changing / interaction behavior: Lane change rate: ; Frequency of path deviation / detour, number of insertion / being inserted events (can be inferred from the trajectory and lane lines). Note: If certain data sources are missing (such as TTC, lane-level trajectory), clustering can still be completed using available alternative features (such as speed fluctuation, acceleration jitter, lane change rate).

[0096] (3) Preprocessing and dimensionality reduction: Outlier handling: Winsorize or remove 3σ values ​​for velocity / acceleration, etc.; Missing value imputation: Interpolation; Normalization: Standardize each feature using z-score. (4) Clustering algorithm type: for sample representation (or use directly) Unsupervised clustering is performed to obtain driving style clusters. Clustering algorithms may be, but are not limited to, any one of the following or a combination thereof:

[0097] K-means / K-medoids: Suitable for large-scale samples, simple to implement; Gaussian Mixture Model (GMM): Outputs soft label probabilities, suitable for cluster boundary overlap; DBSCAN / HDBSCAN: Suitable for detecting noise and abnormal driving clusters; Hierarchical clustering: facilitates the formation of style tree structures (conservative / general / radical).

[0098] Cluster number / parameter determination method: Maximize the silhouette coefficient.

[0099] Furthermore, the steps for assessing graded risks also include: estimating the distribution of passenger flow exiting subway stations based on time-series prediction models, and performing a matching analysis between the distribution of passenger flow exiting subway stations and the dynamic capacity of the associated road network to determine the risk of congestion caused by the superposition of rail passenger flow; wherein, dynamic capacity is the remaining capacity of the road segment after considering real-time traffic conditions and the impact of hidden bottlenecks, and the matching analysis includes converting the estimated distribution of passenger flow exiting subway stations into an equivalent number of vehicles and comparing it with the dynamic capacity.

[0100] The specific matching analysis process: Matching analysis includes: estimating passenger flow exiting subway stations Based on the proportion of motor vehicle trips Average number of passengers and correction factor Converted to the equivalent number of vehicles, as shown in the following formula: (1) Basic definition: in the time window Inside, subway station The estimated passenger flow exiting the subway station is (people / ). Convert it into road sections Equivalent number of vehicles (vehicles / ) ).

[0101] Person-to-Car Equivalent Travel Demand (Number of Trips) Assigned to each road segment (OD / route assignment) (2) Conversion formula: , In the formula, : The proportion of passengers choosing motor vehicle travel upon exiting the station (0~1, which can be set / calibrated according to time of day, weather, and station type); Average number of occupants per motor vehicle (persons / vehicle, which can be set according to city statistics, site surveys or historical data). : Correction factor (used to account for the "vehicle number magnification / reduction" effect caused by empty driving of ride-hailing / taxi, carpooling, etc., default 1, can be calibrated).

[0102] And through allocation ratio Mapping to road segment demand; simultaneously, basic traffic capacity Dynamic traffic capacity is calculated in real time by combining reduction factors based on weather, accidents, construction, and control measures. The remaining capacity is then compared with the total demand of the road segment to determine the risk assessment.

[0103] Furthermore, the steps for dynamically calibrating parameters through an online learning mechanism include: based on a reinforcement learning algorithm, comparing the predicted risk values ​​with the actual congestion reduction effect, and periodically iteratively optimizing the weight parameters and risk level thresholds in the coupled analysis model. To adapt to the continuous parameter adjustment space, the online learning mechanism specifically adopts the Deep Deterministic Policy Gradient Algorithm (DDPG) or the Proximal Policy Optimization Algorithm (PPO), and its specific reinforcement learning framework and iterative update process are designed as follows:

[0104] 1. State Space The state space consists of multi-dimensional traffic state feedback data after the execution of traffic management instructions. Time, State ,in The average speed of the road segment. Queue length at the intersection The microscopic traffic flow turbulence coefficient, This refers to the frequency of high-risk driving behaviors.

[0105] 2. Movement space Action space refers to the adjustment amount by which the system adjusts the internal parameters of the coupled analysis model and the trigger threshold of the control commands. ,in This is a fine-tuning amount for the weighting of multiple indicators that address hidden bottlenecks. The adjustment amount of the risk threshold triggered by the coordinated guidance instructions at all levels. This is the threshold adjustment amount for determining mixed-line interference.

[0106] 3. Reward Function Design: The reward function is quantitatively designed with "actual congestion dissipation effect" and "traffic flow safety and stability" as its guiding principles. Definition Instant rewards after performing an action for:

[0107] , In the formula, , , All are positive weighting coefficients. If parameter adjustments (actions) result in an increase in the average vehicle speed, a reduction in queue length, and a decrease in traffic flow disorder coefficient in the next cycle, the system receives a positive high reward; otherwise, it receives a negative penalty.

[0108] 4. Algorithm Iteration Optimization Cycle and Training Mechanism: The system continuously collects data using a fixed time window as a time step. Experience tuples are stored in the experience replay pool. To avoid frequent adjustments that could cause oscillations in the traffic control system, the iterative optimization cycle is set to daily off-peak hours or weekly weekends. Within the iteration cycle, the algorithm randomly samples batches of data from the experience replay pool to update the neural network parameters. After verifying in a simulation environment that the expected reward value is positive, the updated parameters are then sent to the coupled analysis model.

[0109] This method also includes adaptively adjusting risk assessment parameters and command triggering thresholds under extreme weather or emergencies to address the combined risks of worsening hidden bottlenecks, distorted driving behavior, and fluctuating traffic demand; the adjustment is based on historical data and learning mechanisms.

[0110] Based on the above traffic congestion mitigation methods, the following are some relevant case studies: Case 1 Identification and basic traffic management of hidden bottleneck sections: Purpose of implementation: To demonstrate how the system can automatically identify single congestion risks caused by hidden road defects (such as uneven road surfaces) and perform basic traffic management operations.

[0111] System Implementation: The system will be deployed along the city's main thoroughfare, "Financial Avenue." Hardware includes lidar, pavement stress sensors, intersection video detectors, and edge computing nodes deployed along this route. The software system encompasses all modules related to data sensing, coupled analysis, hierarchical control, and feedback optimization.

[0112] Implementation steps: Data perception and processing: Edge computing nodes process sensor data in real time. Analysis revealed that the middle section of Financial Avenue has the characteristics of abrupt slope changes (average 4.5 degrees), dense manhole covers (2.8 per 100 square meters), and a high proportion of non-motorized vehicles mixed with traffic during peak hours (up to 35%).

[0113] Risk Coupling Analysis: The hidden bottleneck assessment unit uses a multi-index fusion method to calculate the comprehensive disturbance entropy value of the road segment (the calculation result is 0.72). The system compares this value with the dynamic threshold (0.65) and determines that the road segment has a significant single hidden congestion risk.

[0114] Tiered and coordinated control: The system initiates Level 1 control. First, the roadside traffic signal controller dynamically sets up a "mini tidal flow lane" in this section of road to guide vehicles around the lane with the highest concentration of potential hazards. Second, guidance information for non-motorized vehicles is disseminated through roadside information boards. Simultaneously, road warning information is pushed to navigation users traveling through this area.

[0115] Results Feedback and Optimization: Data after one week showed that the average vehicle speed during the evening rush hour on this section of road increased by approximately 27%, and the number of sudden vehicle decelerations decreased by 40%. The feedback optimization layer confirmed that the current risk threshold settings were reasonable and the system was operating stably.

[0116] Implementation Results: The system successfully identified a potential hidden congestion point that is often overlooked by traditional monitoring, and effectively improved the traffic efficiency and safety of the road section through low-cost lane management and information guidance, verifying the accuracy of basic risk identification.

[0117] Case 2 Escalation and guidance of superimposed driver game-theoretic behavior: Objective: To demonstrate how the system identifies overlapping risks and initiates differentiated interventions based on driver behavior when a large number of high-risk driving behaviors occur in hidden bottleneck road sections.

[0118] System Implementation: On the same road section as Case 1, the system was equipped with an in-vehicle terminal data interface and a high-precision video behavior analysis module to capture driver interaction behavior.

[0119] Implementation steps: Data perception and processing: The system detected that in the hidden bottleneck area, some aggressive drivers frequently changed lanes and cut in, while some conservative drivers, out of concern for risk, excessively increased the distance between vehicles, resulting in unstable traffic flow.

[0120] Risk Coupling Analysis: The behavioral game simulation unit simulated this road segment as a scenario. The simulation results show that the physical bottleneck and behavioral game exacerbate each other, causing the traffic flow turbulence coefficient to climb to 0.75 (threshold 0.6). The coupling analysis layer, combining the "high hidden bottleneck entropy value" and the "high traffic flow turbulence coefficient," determined that the current risk is a combination of hidden congestion and behavioral game.

[0121] Tiered and coordinated control: The system has been upgraded to level two control. Building upon level one measures, two key operations have been added: first, strong safety warnings are sent to the terminals of aggressive drivers, and following advice is sent to conservative drivers; second, the signals at upstream intersections are fine-tuned, extending the green light phase by 8 seconds to alleviate traffic congestion.

[0122] Results and Optimization: After the adjustments, cutting in line decreased by approximately 30%, and traffic flow became more orderly. The collected data was used to train a behavioral game theory model, making its predictions of driver behavior more accurate.

[0123] Implementation Results: The system represents an upgrade from "managing things" to "influencing people." By accurately identifying behavioral game risks and implementing differentiated guidance, it effectively quelled micro-level traffic flow disturbances, prevented chain congestion caused by individual driving behaviors, and demonstrated the depth of human-vehicle-road collaboration.

[0124] Case 3 Cross-transit system collaborative management to alleviate large passenger overflows in subway systems: Purpose of implementation: To demonstrate how the system initiates the highest level of cross-transportation mode collaborative scheduling when a subway station experiences a sudden surge in passenger flow, and the spillover effect is compounded by the insufficient carrying capacity of the ground road network.

[0125] System Implementation: The system covers the "Central Park Station" subway station and a surrounding 1.5-kilometer road network. The data source for sensing is extended to the subway operation system, integrating real-time passenger flow data.

[0126] Implementation steps: Data Sensing and Processing: The system receives a subway warning indicating that the peak passenger flow exiting the station will reach 5,000 people per hour in the next 20 minutes. Simultaneously, road network data shows that the saturation level of surrounding arterial roads has reached 0.75, and there are known hidden bottlenecks.

[0127] Risk Coupling Analysis: The track coupling assessment unit predicts the direction of passenger flow evacuation and converts it into equivalent traffic volume. Calculations show that the instantaneous traffic demand of the critical connecting road "Financial Avenue" will approach 88% of its remaining capacity, far exceeding the preset threshold of 80%, and the system determines this as a risk of track-related congestion.

[0128] Tiered and coordinated control: The system initiates three levels of control. Instructions are simultaneously issued to multiple parties: subway stations guide passengers to exit the station and control the speed of turnstiles; the bus dispatching platform increases the frequency of micro-circulation shuttle buses to once every 5 minutes; the traffic guidance system pushes detour suggestions to private cars to avoid congested hotspots.

[0129] Feedback and Optimization: No severe traffic congestion occurred around the subway station during the incident. Data shows that approximately 40% of private cars accepted the guided detours, and public transportation handled over 30% of the exiting passengers. This successful handling provides a key case study for optimizing the passenger-vehicle flow coupling threshold.

[0130] Implementation Results: The system transcends the scope of single-mode ground transportation management, achieving "supply-demand matching" and "time-space peak avoidance" between rail transit and ground transportation. Through cross-modal resource scheduling and passenger flow guidance, it successfully mitigated short-term, concentrated surges in passenger flow, ensuring the normal operation of urban hub functions.

[0131] Case 4 Comprehensive emergency response to multiple risks under extreme weather conditions: Purpose of implementation: To demonstrate the system's comprehensive handling and enhanced control capabilities in the face of multiple risks (deterioration of hidden bottlenecks, distortion of driving behavior, and fluctuations in traffic demand) occurring simultaneously under extreme weather conditions such as heavy rain.

[0132] System implementation: The implementation area is the road network surrounding the city's transportation hub "South Station". The system is in full-function operation.

[0133] Implementation steps: Data perception and processing: During heavy rain, sensors detected a surge in road surface smoothness entropy caused by water accumulation in tunnels; video analysis showed that drivers' anxiety led to an increase in risky behaviors; at the same time, the subway faced instantaneous passenger flow pressure due to train delays.

[0134] Risk Coupling Analysis: The coupling analysis layer assesses three types of risks in parallel and identifies that in the "tunnel exit" area, the three risks highly overlap in time and space, which is very likely to induce serious accidents and regional congestion, and is judged as the highest level of compound congestion risk.

[0135] Tiered and coordinated control: The system implements an enhanced three-tiered control mode. In addition to conventional traffic management, several measures have been strengthened: intelligent lane lights are activated in tunnels to guide traffic flow; rainstorm and flood safety warnings are sent to all vehicles; and subway and taxi dispatching is coordinated to implement phased passenger flow control and vehicle dispatching.

[0136] Feedback and Optimization: After the rainstorm, the system initiated specialized learning based on the entire process data. Analysis revealed that risk thresholds are more sensitive under severe weather conditions. Therefore, the trigger thresholds for hidden bottlenecks were dynamically lowered, and the parameters related to weather impact in the behavioral game model were updated, making the system more intelligent in responding to similar extreme scenarios in the future.

[0137] Implementation Results: In complex and harsh environments, the system demonstrated its robustness and decision-making advantages. By integrating multi-source risk information and activating enhanced collaborative commands, it effectively prevented traffic paralysis that could have been caused by a chain reaction of multiple factors, showcasing the system's core value in emergency management.

[0138] Case 5 Model self-optimization process based on real-world data: Objective: To demonstrate how the system feedback optimization layer utilizes actual operational data and machine learning to continuously improve the accuracy of the internal analysis model, thereby achieving self-evolution of system performance.

[0139] System Implementation: This case study focuses on the system's software learning module, using long-term operational data from multiple bottleneck road sections across the city as the training basis.

[0140] Implementation steps: Data collection: The feedback optimization layer continuously collects real following behavior data (distance, speed, etc.) of different conservative drivers on bottleneck sections and corresponding traffic scenario data.

[0141] Model training: A new driver behavior simulation agent is trained using reinforcement learning algorithms with "safety" and "efficiency" as the comprehensive reward objectives.

[0142] Model Validation and Update: The new agent was tested in the simulation environment, and its behavior reduced traffic delays by 15% compared to the old model, and was closer to real-world data. Subsequently, the new model was smoothly updated into the behavioral game simulation unit of the production system.

[0143] Effect closed loop: The updated model makes the system's prediction of traffic flow more accurate, thus enabling the generation of better control strategies. The implementation effect of the new strategy is then collected and used as input for the next round of optimization, forming a closed loop of continuous improvement.

[0144] Implementation Results: The system has overcome the limitations of relying entirely on preset parameters and fixed models, establishing a closed loop of "practice-learning-optimization." This self-optimization capability ensures that the system can adapt to the long-term evolution of traffic flow characteristics, such as the widespread adoption of new vehicles and changes in driving habits, maintaining long-term effectiveness.

[0145] Comparison Case 1 Limitations of traditional single-point signal optimization systems: To highlight the technical advantages of the system of this invention, a comparison is made with traditional adaptive signal control systems. Traditional systems rely solely on intersection loop detectors to detect traffic flow, using the optimization of individual or arterial signal timing as the only means.

[0146] In the same evening rush hour scenario on the "Middle Section of Financial Avenue," traditional systems only perceive high traffic volume and low speed. Their analytical logic attributes this to "excessive demand," and their decision is simply to extend the green light time upstream, attempting to expedite vehicle passage. However, due to the failure to eliminate the physical interference of hidden bottlenecks and the micro-level disturbances of driver interaction, more vehicles are rapidly injected into the problematic section, causing congestion to accumulate rapidly behind the bottleneck point and even overflow to upstream intersections, leading to even wider delays. Traditional systems lack the ability to identify the root causes of risk, intervene in driver behavior, and coordinate across systems. When facing complex coupled risks, their single control measures often have limited effectiveness or even backfire.

[0147] Compared to Cases 1 to 5 and Comparative Case 1, this invention, through a comprehensive comparison of Cases 1 to 5 and the comparative traditional system, reveals the fundamental differences between the two in terms of core concepts, technical capabilities, and practical effectiveness. The limitations of the traditional system (comparative example) lie in its single perception dimension, linear analytical logic, and isolated control methods. It can only perceive the macroscopic phenomenon of traffic flow through detectors such as coils, simply attributing congestion to "demand exceeding supply." Its only decision output is adjusting intersection signal timing, which is a passive, single-point "open-loop control." This model is not only powerless to resolve congestion caused by the coupling of complex factors such as hidden road defects, driver interaction, and overflowing passenger flow from rail transit, but may even exacerbate congestion by blindly injecting more vehicles into bottleneck sections, producing a counterproductive effect. Conversely, the system of this invention constructs an intelligent closed loop of multi-source perception, coupled analysis, hierarchical collaboration, and dynamic optimization. It acts like a traffic commander with "eagle eyes" and a "super brain": it can not only "see" traffic flow, but also "see" the hidden physical bottlenecks that cause traffic flow disorder (Case 1), the micro-behavioral games between drivers (Case 2), and the macro-supply and demand contradictions between subway passenger flow and road network capacity (Case 3). Its coupling analysis layer can clarify the interactions and superposition effects among these heterogeneous risks, thereby achieving a precise diagnosis of the root causes of congestion, rather than merely treating the symptoms.

[0148] Based on this in-depth diagnostics, the system's control response exhibits a high degree of systematicity, coordination, and adaptability. It no longer relies on signal timing as a "single-soldier weapon," but instead constructs a hierarchical collaborative operational system encompassing "spatial allocation (lane management), behavioral guidance (human-machine interaction), and capacity scheduling (railway linkage)" (Case 4). From guiding vehicles around a manhole cover to dissuading dangerous lane-cutting, and dispatching a batch of buses to evacuate subway passengers, the system can automatically match and coordinate the execution of a series of measures from micro to macro levels based on risk levels. More importantly, its built-in feedback optimization layer (Case 5) enables the system to evolve by "learning from practice," dynamically calibrating models and thresholds based on historical control effects, thereby continuously improving its decision-making intelligence in dealing with future complex scenarios (such as extreme weather). In summary, traditional systems are like ordinary traffic lights that can only direct traffic based on the single indicator of "vehicle queue length." In contrast, the system of this invention is a smart city traffic hub that integrates multi-dimensional information such as "road health status, driver psychological state, and public transportation load," and coordinates the efforts of traffic police, navigation platforms, subway companies, and other stakeholders to conduct precise traffic management and proactive deployment. This comparison represents a leap from "single-point automation" to "system intelligence," resulting in improved traffic efficiency, reduced safety hazards, and enhanced urban resilience, demonstrating significant technological advancement and practical value.

[0149] In summary, compared with the prior art, the present invention has the following technical effects: 1. This invention integrates multi-source heterogeneous data, including road infrastructure status, driver micro-interaction behavior, and rail transit macro-passenger flow, and constructs a coupled analysis model. The system effectively reveals three types of heterogeneous risks and their cumulative effects: hidden physical bottlenecks, traffic flow disturbances caused by driver game theory, and imbalances in rail passenger flow supply and demand. This ability to move from single-phenomenon monitoring to multi-dimensional coupled diagnosis transforms traffic management from passively responding to "existing congestion" to proactively intervening in "potential risks," significantly improving the accuracy and timeliness of risk warnings and laying a core decision-making foundation for precise traffic management.

[0150] 2. Based on different risk levels determined by coupling analysis, this invention automatically matches and triggers an integrated hierarchical response instruction set, ranging from basic lane management to driver behavior guidance and even cross-transit resource scheduling. This "risk-driven, hierarchical response" model ensures that the intensity of regulation is precisely matched with the severity of risk, avoiding resource waste or insufficient intervention. Simultaneously, the instructions are executed collaboratively across roadside equipment, onboard terminals, and the subway operation system, forming a closed-loop regulatory force of "people-vehicle-road-rail," thereby systematically resolving congestion from multiple levels, including spatial, behavioral, and capacity aspects, significantly improving traffic flow efficiency and the overall resilience of the urban transportation system.

[0151] 3. This invention can dynamically calibrate key parameters and decision thresholds in the risk assessment model using actual effect data after the implementation of control measures. This allows the system to continuously accumulate experience and adapt to the long-term evolution of traffic flow characteristics (such as changes in driving habits and the emergence of new bottlenecks) and different external conditions (such as severe weather), making the analysis model increasingly accurate and the control strategy continuously optimized. Ultimately, this forms a virtuous cycle of evolution that becomes smarter with use, overcoming the fundamental defects of traditional systems where parameters are fixed and difficult to adapt to dynamic and complex environments.

[0152] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the claims of this invention and their equivalents, this invention also intends to include these modifications and variations.

Claims

1. A smart traffic congestion relieving method for urban road network, characterized in that, Includes the following steps: Collect spatiotemporal characteristic data of hidden bottlenecks in the target road network, driver interaction behavior data, and rail passenger flow coupling correlation data, and perform preprocessing. Based on the preprocessed spatiotemporal characteristic data of hidden bottlenecks, driver interaction behavior data, and rail passenger flow coupling correlation data, a coupling analysis model is constructed to determine the comprehensive risk level. Based on the comprehensive risk classification value and the preset risk assessment parameters, the classification risk types caused and superimposed by the spatiotemporal characteristics of hidden road bottlenecks, driver interaction behavior data and passenger flow coupling correlation data of rail network are assessed. Based on the assessed risk level, a matching graded collaborative diversion instruction set is generated, and each diversion instruction in the graded collaborative diversion instruction set is distributed to the corresponding traffic control system execution terminal to alleviate traffic congestion. The method for constructing the coupling analysis model includes: Based on the preprocessed spatiotemporal characteristic data of hidden bottlenecks, driver interaction behavior data, and rail passenger flow coupling correlation data, the hidden bottleneck interference degree index, traffic flow disorder coefficient, and rail supply-demand imbalance ratio of the target section at the evaluation time are obtained respectively. Based on the degree of influence corresponding to the hidden bottleneck interference degree index, traffic flow disorder coefficient, and rail supply-demand imbalance ratio, their corresponding dynamic weight coefficients are determined and weighted fusion calculations are performed to obtain the comprehensive graded risk value. 2.The smart traffic congestion relieving method for urban road network according to claim 1, wherein, It also includes a feedback optimization step: collecting multi-dimensional traffic status feedback data after the execution of the traffic diversion instructions, and dynamically calibrating the risk assessment parameters of the coupled analysis model and the instruction triggering threshold of the hierarchical collaborative traffic diversion instruction set based on the multi-dimensional traffic status feedback data through an online learning mechanism.

3. The intelligent traffic congestion mitigation method for urban road networks according to claim 1, characterized in that, The steps for assessing risk stratification include: Based on driver interaction behavior data, the interaction decision-making process between drivers is simulated, and the traffic flow disorder coefficient is output to quantitatively characterize the traffic flow stability risk caused by driver interaction behavior. The driver interaction behavior data includes the frequency of cutting in, following distance, and number of lane changes. Driving style labels are generated by clustering historical driving data. The driving style labels include at least aggressive, conservative, and neutral types. Based on the coupled data of rail passenger flow, a time-series prediction model is used to estimate the distribution of exiting passenger flow at subway stations within a target time period. This is then combined with a matching analysis of the dynamic capacity of the associated road network to quantify and determine the risk of congestion caused by the superposition of rail passenger flow. The matching analysis includes converting the estimated exiting passenger flow at subway stations into an equivalent number of vehicles and comparing this equivalent number of vehicles with the dynamic capacity to obtain a supply-demand imbalance ratio that characterizes the supply-demand relationship of the rail system. The traffic flow disorder coefficient and the supply-demand imbalance ratio are used as core quantitative indicators. They are input together with the hidden bottleneck interference degree index into the weighted fusion algorithm of the coupled analysis model to calculate the comprehensive graded risk value and complete the graded risk assessment including "hidden congestion and behavioral game superposition risk" and "railway superposition congestion risk".

4. The intelligent traffic congestion mitigation method for urban road networks according to claim 3, characterized in that, The comprehensive risk level is determined according to the following formula: ， ， ， ， In the formula, This is a comprehensive risk classification value; Index of the degree of interference from hidden bottlenecks; The traffic flow turbulence coefficient represents the degree of impact of driver interaction. To meet the basic traffic needs of the road section; This refers to the equivalent number of vehicles generated from the passenger flow exiting the subway station. This item is for the dynamic traffic capacity of the road section. The overall characterization of the risks associated with supply and demand imbalances under track coupling; , , Each risk item is in Dynamic weighting coefficients at any given time; For the preprocessed first The original quantization parameters of the spatiotemporal characteristics of a hidden bottleneck; This represents the total number of latent bottleneck spatiotemporal characteristics; For the first The weights of the spatiotemporal features of the hidden bottleneck; This represents the total number of driver behavior status categories. For each state category The probability of its occurrence; To predict passenger flow exiting subway stations; The percentage of passengers choosing motor vehicle travel when exiting the station; The average number of occupants in a motor vehicle; This is a correction factor.

5. The intelligent traffic congestion mitigation method for urban road networks according to claim 1, characterized in that, The hierarchical and coordinated guidance instruction set includes: Level 1 control instructions, targeting a single hidden congestion risk, include at least using signal control to create dynamic lane allocation to avoid bottleneck areas; The secondary control order addresses the combined risks of hidden congestion and behavioral game theory. Based on the primary control order, it adds targeted intervention for high-risk driving behaviors and fine-tuning of local traffic signal parameters. The Level 3 control order addresses the risk of congestion on rail lines by initiating cross-transportation coordinated scheduling based on the Level 2 control order. This includes coordinated control of subway passenger flow, surface connecting transport capacity, and traffic flow on related road networks.

6. The intelligent traffic congestion mitigation method for urban road networks according to claim 2, characterized in that, The steps for dynamically calibrating parameters in an online learning mechanism include: Collect multi-dimensional traffic status feedback data after the execution of the traffic diversion command, and quantify the collected data into measurable system status indicators. The indicators include at least the changes in average vehicle speed on road sections, changes in queue length at intersections, and actual traffic flow disorder coefficients generated after the execution of the hierarchical coordinated traffic diversion command. The dynamic weighting coefficients that directly affect the calculation results of the comprehensive risk level in the coupled analysis model, and the risk level thresholds on which the triggering conditions of different levels of instructions in the hierarchical collaborative guidance instruction set depend, are collectively defined as the action parameters to be optimized and adjusted in the learning framework. Based on quantified system state indicators, a reward function is designed and a reward value is obtained; the value of the reward function is positively correlated with the increase in average vehicle speed, the reduction in queue length, and the decrease in traffic flow disorder coefficient in the next cycle. Based on the reward value, the reinforcement learning framework is driven to periodically iteratively optimize the defined action parameters; the coupled analysis model and the hierarchical collaborative guidance instruction set are updated based on the optimized parameters.

7. The intelligent traffic congestion mitigation method for urban road networks according to claim 1, characterized in that, It also includes adaptive risk response steps in the event of extreme weather or emergencies, specifically including: Real-time monitoring of meteorological data or emergency alarm information; when extreme weather conditions or emergencies of a specific level are detected, the parameter adaptive adjustment mode is automatically triggered. Based on the currently monitored scene characteristics, historical traffic data and optimized parameter sets for similar scenarios are matched from the historical database; the parameter set includes at least the risk assessment parameters of the coupled analysis model calibrated for similar scenarios and the instruction triggering thresholds of the hierarchical collaborative traffic management instruction set. In the parameter adaptive adjustment mode, the parameter set is used as the initial strategy of the online learning mechanism. During the duration of this scenario, the online learning mechanism continues to be executed, and the risk assessment parameters and instruction trigger thresholds are quickly fine-tuned using real-time collected multi-dimensional traffic status feedback data to dynamically respond to the superimposed risks of hidden bottleneck deterioration, driving behavior distortion and traffic demand fluctuation caused by extreme weather or sudden events.

8. A smart traffic congestion mitigation system for urban road networks, used to implement the method described in any one of claims 2-7, characterized in that, It includes a multi-source data sensing layer, a coupling analysis layer, a hierarchical control layer, and a feedback optimization layer connected in sequence; The multi-source data perception layer is used to collect multi-source heterogeneous data of the target road network and perform preprocessing. The multi-source heterogeneous data includes latent bottleneck spatiotemporal feature data, driver interaction behavior data, and rail passenger flow coupling correlation data. The coupling analysis layer is used to construct a coupling analysis model based on the preprocessed multi-source heterogeneous data, obtain a comprehensive graded risk value, and, in combination with preset risk assessment parameters, assess the graded risks caused and superimposed by the spatiotemporal characteristic data of road hidden bottlenecks, driver interaction behavior data, and rail network passenger flow coupling correlation data, respectively. The graded risk types include at least single hidden congestion risk, hidden congestion and behavioral game superposition risk, and rail superposition congestion risk. The hierarchical control layer is used to generate a hierarchical collaborative diversion instruction set that matches the hierarchical risk type output by the coupling analysis layer, and distribute each diversion instruction in the diversion instruction set to the corresponding traffic control system execution terminal. The feedback optimization layer is used to collect multi-dimensional traffic status feedback data after the execution of traffic diversion instructions. Based on the multi-dimensional traffic status feedback data, the risk assessment parameters of the coupled analysis model and the instruction triggering threshold of the hierarchical collaborative traffic diversion instruction set are dynamically calibrated through an online learning mechanism.

9. A smart traffic congestion mitigation system for urban road networks according to claim 8, characterized in that, The multi-source data perception layer includes lidar, stress sensors, and video detectors deployed on road infrastructure for collecting multi-source heterogeneous data, as well as an edge computing node network for processing multi-source heterogeneous data; the traffic control system execution terminal includes a traffic signal controller, variable message signs, vehicle guidance terminals, and a public transportation operation scheduling platform, which are used to receive and execute hierarchical collaborative traffic management instruction sets.

10. A smart traffic congestion mitigation system for urban road networks according to claim 8, characterized in that, The coupling analysis layer includes a hidden bottleneck assessment unit, a behavioral game simulation unit, and a track coupling assessment unit. The hidden bottleneck assessment unit uses a multi-index fusion method based on information entropy to calculate the hidden bottleneck entropy value, which is used to assess the risk of a single hidden congestion. The behavioral game simulation unit simulates driver interaction decisions based on game rules, which is used to assess the risk of the superposition of hidden congestion and behavioral game. The track coupling evaluation unit uses a time-series prediction model to perform passenger flow and road network matching analysis to assess the risk of track congestion.

Images