Intelligent network connection mixed traffic flow modeling method fusing perception risk

Abstract

The invention discloses an intelligent network connection mixed traffic flow modeling method fusing perception risks. The method comprises the steps that mixed traffic data are collected and preprocessed, a two-dimensional risk field is constructed with the mass center of a vehicle as the core, the real-time perception risk value of each vehicle is calculated, and risk sensitivity parameters of different types of vehicles are set; after a risk-behavior coupled lane changing decision model and a risk-decision coupled longitudinal behavior model are constructed based on a real-time perception risk value, the model and the differentiated risk perception parameters are integrated to a unified simulation platform, permeability combinations of different types of vehicles are set, dynamic simulation of mixed traffic flow is executed, and a simulation operation result is output. According to the method, operation mechanism analysis and management and control strategy evaluation under multi-type vehicle permeability combination can be realized, and the behavior authenticity of a hybrid simulation model is improved.

Description

Technical Field

[0001] This invention belongs to the field of intelligent transportation and traffic simulation technology, and in particular relates to an intelligent connected mixed traffic flow modeling method that integrates risk perception. Background Technology

[0002] With the large-scale application of autonomous driving technology, vehicle-to-everything (V2X) technology, and intelligent connected vehicles (CAVs), mixed traffic flows (i.e., mixed traffic flows) involving high-level adaptive cruise control (ACC) vehicles, autonomous driving (AV) vehicles, and traditional human-driven (HV) vehicles on highways, urban expressways, ramps, and other road sections are gradually becoming a reality. This type of mixed traffic flow is characterized by heterogeneous vehicle categories, differences in behavioral models, complex interactions, and uneven risk perception capabilities.

[0003] In existing technologies, models of vehicle longitudinal behavior (such as car-following) and lateral behavior (such as lane-changing) are relatively mature. For example, classic car-following models (such as the Intelligent Driver Model and the Wiedemann model) are used to describe the longitudinal motion of vehicles under conditions such as headway, speed difference, and acceleration; lane-changing models (such as the MOBIL model and the discrete choice model (Logit model)) are used to characterize vehicles' lane-changing intentions and gap selection decisions. At the same time, traffic simulation platforms (such as SUMO and VISSIM) are also widely used for traffic control strategy evaluation and traffic flow operation analysis.

[0004] However, existing technologies still have significant shortcomings in the following aspects: The behavioral differences and risk perception differences of various vehicle types (ACC, AV, HV) in the same traffic flow have not been systematically incorporated into the model, making it difficult to truly reflect the behavioral mechanisms of mixed traffic flows. Although some studies have considered vehicle risk perception or safety measurement, most of them remain at the macro level or local models, and there is a lack of unified modeling that uses risk value as an input variable for behavioral decisions (longitudinal following or lateral lane changing). In lane-changing decision-making models, most only generate intentions and select gaps based on traffic conditions (such as distance between vehicles, speed difference, and lane speed difference), and lack the inclusion of the vehicle's real-time perceived risk level into the two key sub-processes of intention generation and gap selection, thus failing to fully characterize the behavioral inhibition effect in high-risk scenarios. Although traffic simulation platforms have been used in mixed traffic scenarios involving multiple types of vehicles, most have not integrated differentiated risk perception-behavior coupling models into the simulation platform, nor have they further analyzed key indicators such as mixed traffic flow operation, lane-changing behavior, and risk exposure levels under different penetration rate combinations.

[0005] Therefore, there is an urgent need to propose a modeling method for intelligent connected mixed traffic flow that integrates risk perception. Summary of the Invention

[0006] The purpose of this invention is to provide an intelligent connected mixed traffic flow modeling method that integrates risk perception. This method systematically incorporates risk perception in a two-dimensional plane into the longitudinal following behavior model and the lateral lane-changing behavior model of vehicles in mixed traffic scenarios, targeting the behavioral mechanisms of high-level adaptive cruise control (ACC), autonomous driving (AV), and human-driven (HV) vehicles. Furthermore, it differentiates the behavioral parameters of different vehicle types and integrates them into a unified simulation platform to achieve accurate simulation and evaluation of the mixed traffic flow operation under different penetration rate combinations.

[0007] To achieve the above objectives, this invention provides a method for modeling intelligent connected mixed traffic flow that integrates risk perception, comprising: Collect trajectory and environmental data of autonomous vehicles and human-driven vehicles at different levels in mixed traffic scenarios, and preprocess the collected data. Based on the preprocessed data, a two-dimensional risk field is constructed with the vehicle's centroid as the core, the real-time perceived risk value of each vehicle is calculated, and risk sensitivity parameters for different types of vehicles are set to distinguish the differentiated risk perception results. The real-time perceived risk value is placed in front of the longitudinal following model and coupled with the basic longitudinal action function to construct a risk-decision coupled longitudinal behavior model. The real-time perceived risk value is introduced into the lane-changing motivation generation module and lane-changing gap selection module of the lateral lane-changing model to construct a risk-behavior coupled lane-changing decision model. The risk-decision coupled longitudinal behavior model, the risk-decision coupled lane-changing decision model, and the differentiated risk perception parameters are integrated into a unified simulation platform. Different penetration rate combinations of different types of vehicles are set, dynamic simulation of mixed traffic flow is performed, and the simulation results are output.

[0008] Preferably, the process of collecting trajectory and environmental data of autonomous vehicles at different levels and human-driven vehicles in mixed traffic scenarios, and preprocessing the collected data, includes: The system simultaneously collects trajectory and environmental data from high-level adaptive cruise vehicles, autonomous vehicles, and human-driven vehicles. The trajectory and environmental data includes vehicle position, speed, acceleration, lateral offset, lane number, distance to the vehicle in front and behind, and status of surrounding traffic participants. The collected vehicle position and speed signals are smoothed and filtered by Savitzky-Golay to eliminate noise, and the lateral offset of the vehicle and its relative motion with the preceding vehicle are calculated using a predetermined sliding window. Based on the continuous changes in the vehicle's lateral offset, lane number, and relative motion state with the vehicle in front, follow-up event segments and lane-changing event segments are identified and extracted. Segments with a duration lower than a preset threshold are removed to obtain a structured trajectory dataset.

[0009] Preferably, the process of constructing a two-dimensional risk field and calculating the real-time perceived risk value for each vehicle includes: A double-elliptical two-dimensional risk field is established with the center of mass of the main vehicle as the origin. The repulsive forces of the main vehicle on each potential conflict source around it are superimposed to obtain the overall risk value. Calculate the repulsive force of the main vehicle to each conflict source, the repulsive force including the influence of the lateral and longitudinal relative position, vehicle mass and risk perception judgment differential parameters; Different risk perception judgment differential parameter values ​​are set according to vehicle type, and the overall risk value is weighted to obtain the differential risk perception result, which is used as the input parameter for subsequent longitudinal and lateral behavior models.

[0010] Preferably, the process of constructing a risk-decision coupled longitudinal behavioral model includes: The basic longitudinal acceleration is calculated based on the relative distance and speed difference between the vehicle and the vehicle in front, as well as the speed of the vehicle itself. The basic longitudinal motion function is then constructed based on the basic longitudinal acceleration. A risk perception correction function is established, which takes the real-time perceived risk value as input and outputs a risk correction term. The basic longitudinal action function and the risk perception correction function are coupled to generate the final longitudinal acceleration command, and the longitudinal following behavior is dynamically adjusted through risk perception.

[0011] Preferably, the process of constructing a risk-behavior coupled lane-changing decision-making model includes: Based on the real-time perceived risk value and the traffic state variables of the current lane and the target lane, a discrete choice model is used to calculate the probability of a vehicle's lane-changing intention. When the probability of lane change intention reaches a preset threshold, the gap acceptance probability is calculated based on the gap status in front and behind the target lane, the speed difference, and the real-time perceived risk value. When the probability of accepting the gap exceeds the execution threshold, a lane-changing action is triggered. At the same time, the safety constraints are adaptively adjusted according to the real-time perceived risk value to suppress lane-changing behavior in high-risk scenarios.

[0012] Preferably, the process of performing dynamic simulation of mixed traffic flow includes: The longitudinal behavior model, lane-changing decision model, and differentiated risk perception parameters are loaded into the simulation platform, and the vehicle state, simulation time step, and total duration are initialized. By configuring the ratio of high-level adaptive cruise vehicles to autonomous vehicles through the penetration rate setting module, mixed traffic scenarios with different levels of automation can be formed. When the simulation is running, the real-time perceived risk value of all vehicles is updated sequentially at each time step, the longitudinal motion state is updated according to the longitudinal behavior model, and the lateral lane change behavior is judged and executed according to the lane change decision model. The process is iterated until the simulation ends. Meanwhile, key operational indicators, including vehicle speed, traffic flow, density, lane change frequency, and risk exposure level, are recorded and calculated during the simulation process.

[0013] Preferably, the simulation platform includes: The model loading module is used to import the longitudinal behavior model, lane change decision model, and risk perception parameters. The penetration rate setting module is used to set the ratio of high-level adaptive cruise vehicles to autonomous vehicles in traffic flow; The behavior parameter configuration module is used to configure risk avoidance sensitivity parameters for different types of vehicles. The risk-decision coupling module is used to input the vehicle's real-time perceived risk value into the longitudinal behavior model and lane-changing decision model at each simulation time step; The simulation results analysis module is used to collect simulation data and calculate key operating indicators.

[0014] Preferably, the penetration rate setting module is used to set the following scenarios: In scenarios where vehicles are driven entirely by humans, the proportion of high-level adaptive cruise control vehicles and autonomous driving vehicles is 0%. In a mixed penetration scenario, the ratio of high-level adaptive cruise control vehicles to autonomous driving vehicles is 10% to 50%. In scenarios with high automation penetration, the proportion of high-level adaptive cruise vehicles and autonomous vehicles should not be less than 50%.

[0015] Preferably, the process by which the behavior parameter configuration module configures risk avoidance sensitivity parameters for different types of vehicles includes: Differentiated risk avoidance sensitivity parameters were set for human-driven vehicles, high-level adaptive cruise control vehicles, and autonomous vehicles, so that the risk avoidance sensitivity parameter value of human-driven vehicles was the highest, the risk avoidance sensitivity parameter value of autonomous vehicles was the lowest, and the risk avoidance sensitivity parameter value of high-level adaptive cruise control vehicles was in between.

[0016] Preferably, the dynamic simulation runs by performing the following operations at fixed time steps: The two-dimensional risk field is invoked to calculate the real-time perceived risk value of each vehicle at the current moment; The real-time perceived risk value is transmitted to the longitudinal behavior model and lane-changing decision model of each vehicle; The vehicle position, speed, and lane number are updated based on the longitudinal acceleration output by the longitudinal behavior model and the lateral commands output by the lane change decision model. At the end of each time step, refresh the status information of all vehicles and determine whether the simulation termination condition is met.

[0017] Compared with the prior art, the present invention has the following advantages and technical effects: The method of this invention can systematically introduce two-dimensional planar risk perception, modify the target acceleration in the car-following model and the utility function of the lane-changing model at each time step by adjusting the risk-specific parameters of different vehicles, thereby forming a heterogeneous traffic flow evolution algorithm that integrates risk perception, and integrating it into the simulation platform to realize the analysis of the operation mechanism and the evaluation of control strategies under the combination of penetration rates of multiple types of vehicles.

[0018] The method of this invention realistically reproduces the differences in longitudinal following and lateral lane changing behaviors of ACC, AV and HV vehicles in mixed traffic flow, thereby improving the behavioral realism of the mixed traffic simulation model.

[0019] The method of this invention introduces two-dimensional planar risk perception and incorporates it into the vehicle behavior decision-making process, enabling the model to quantitatively characterize the trend of vehicle behavior suppression or acceleration of lane changing / speed change under different risk levels, thereby enhancing the model's sensitivity to safety / risk factors.

[0020] The method of this invention integrates the longitudinal and lateral sub-models of risk-decision coupling into a unified simulation platform, which is applicable to mixed traffic simulation under the condition of multiple vehicle penetration rates, thereby providing a systematic and operable tool for evaluating traffic control strategies.

[0021] The method of this invention can be used for road construction, smart highway operation, and control decision support in vehicle-road cooperative scenarios, enriching the technical means of mixed traffic flow analysis and simulation. Attached Figure Description

[0022] The accompanying drawings, which form part of this application, are used to provide a further understanding of this application. The illustrative embodiments and descriptions of this application are used to explain this application and do not constitute an undue limitation of this application. In the drawings: Figure 1 This is a schematic diagram of the method flow according to an embodiment of the present invention. Detailed Implementation

[0023] It should be noted that, unless otherwise specified, the embodiments and features described in this application can be combined with each other. This application will now be described in detail with reference to the accompanying drawings and embodiments.

[0024] It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions, and although a logical order is shown in the flowchart, in some cases the steps shown or described may be executed in a different order than that shown here.

[0025] like Figure 1 As shown, this embodiment provides a method for modeling intelligent connected mixed traffic flow that integrates risk perception, including: Collect trajectory and environmental data of autonomous vehicles and human-driven vehicles at different levels in mixed traffic scenarios, and preprocess the collected data. Based on the preprocessed data, a two-dimensional risk field is constructed with the vehicle's centroid as the core, the real-time perceived risk value of each vehicle is calculated, and risk sensitivity parameters for different types of vehicles are set to distinguish the differentiated risk perception results. By placing the real-time perceived risk value in the longitudinal following model and coupling it with the basic longitudinal action function, a risk-decision coupled longitudinal behavior model is constructed. By incorporating real-time perceived risk values ​​into the lane-changing motivation generation module and lane-changing gap selection module of the lateral lane-changing model, a risk-behavior coupled lane-changing decision model is constructed. The risk-decision coupled longitudinal behavior model, the risk-decision coupled lane-changing decision model, and the differentiated risk perception parameters are integrated into a unified simulation platform. Different penetration rate combinations of different types of vehicles are set, dynamic simulation of mixed traffic flow is performed, and the simulation results are output.

[0026] Furthermore, the process of collecting trajectory and environmental data of different levels of automated vehicles and human-driven vehicles in mixed traffic scenarios, and preprocessing the collected data, includes: Simultaneously collect trajectory and environmental data of high-level adaptive cruise vehicles, autonomous vehicles, and human-driven vehicles. The trajectory and environmental data include vehicle position, speed, acceleration, lateral offset, lane number, distance to the vehicle in front and behind, and status of surrounding traffic participants. The collected vehicle position and speed signals are smoothed and filtered by Savitzky-Golay to eliminate noise, and the lateral offset of the vehicle and its relative motion with the preceding vehicle are calculated using a predetermined sliding window. Based on the continuous changes in vehicle lateral deviation, lane number, and relative motion state with the vehicle in front, follow-up event segments and lane-changing event segments are identified and extracted. Segments with a duration of less than a preset threshold are removed to obtain a structured trajectory dataset.

[0027] Furthermore, this embodiment simultaneously collects trajectory and environmental data of different levels of autonomous vehicles (high-level adaptive cruise control (ACC) vehicles and autonomous driving vehicles (AV) vehicles) and human-driven vehicles (HV vehicles) in mixed traffic scenarios, and completes missing data completion, anomaly removal and spatiotemporal alignment. Specifically, by simultaneously collecting trajectory and environmental data of different levels of autonomous vehicles (high-level adaptive cruise control (ACC) vehicles and autonomous driving (AV) vehicles) and human-driven vehicles (HV vehicles) in mixed-traffic scenarios, information such as the longitudinal position, speed, acceleration, lane number, and status of vehicles in front and behind is obtained within a typical road cross-section. This data can originate from onboard GNSS / IMU terminals, roadside video detection equipment, or millimeter-wave radar systems, and after unified time synchronization and coordinate transformation, it forms trajectory samples that can be directly used for modeling.

[0028] First, Savitzky-Golay smoothing filters were used to eliminate position and velocity noise from the original position and velocity signals. Then, a 0.1s sliding window was used to calculate the lateral offset of the vehicle and its relative motion with the vehicle in front. When the vehicle remained in the same lane, the absolute value of the lateral offset with the same vehicle in front was less than 0.3m, and the relative distance was less than 100m within a window of at least 3s, it was identified as a car-following event segment. When the cumulative lateral displacement of the vehicle exceeded half the width of the lane and the lane number changed, and the peak lateral velocity was greater than 0.3m / s, it was classified as a lane-changing event segment. Segments of less than 5s were discarded to ensure the stability and representativeness of the subsequent modeling samples.

[0029] To facilitate model calibration and comparison for different vehicle types, the observation trajectories for different vehicles are specified as follows: , , , Unified representation as , for Observe vehicle kinematic parameters at all times, such as front-to-back distance. ,speed Longitudinal acceleration lateral acceleration Lateral offset wait.

[0030] The above methods can be used to obtain structured and noise-suppressed multi-source fusion vehicle trajectory data, providing a reliable input basis for risk field calculation, longitudinal car-following modeling, and lateral lane-changing modeling.

[0031] Specifically, the data collection in this embodiment includes vehicle position, speed, acceleration, lateral offset, lane number, distance between vehicles, environmental conditions of surrounding traffic participants (including motor vehicles, non-motor vehicles, and pedestrians), as well as road geometry, traffic signs, traffic flow composition, weather and lighting conditions.

[0032] Furthermore, the process of constructing a two-dimensional risk field and calculating the real-time perceived risk value for each vehicle includes: A double-elliptical two-dimensional risk field is established with the center of mass of the main vehicle as the origin. The repulsive forces of the main vehicle on each potential conflict source around it are superimposed to obtain the overall risk value. Calculate the repulsive force of the main vehicle to each conflict source. The repulsive force includes the influence of the relative position in the lateral and longitudinal directions, vehicle mass, and risk perception judgment differential parameters. Different risk perception judgment parameters are set according to vehicle type. The overall risk value is weighted to obtain the differentiated risk perception result, which is used as the input parameter for subsequent longitudinal and lateral behavior models.

[0033] Furthermore, this embodiment proposes differentiated modeling of longitudinal models and risk perception for different vehicles in mixed traffic flow: it introduces a risk function with the driver's two-dimensional risk perception field as the core, and uses independent risk perception parameters as driving behavior parameters for different types of vehicles. Specifically, this embodiment models the risk perception characteristics of different types of vehicles in mixed traffic scenarios. By establishing a two-dimensional psycho-physical risk field with double ellipses based on the vehicle's centroid, the real-time risk perception level of the vehicle is calculated, and this risk level is used as the input parameter for subsequent longitudinal and lateral behavior models.

[0034] The basic construction process of the risk field includes: based on the unified modeling concept of driving psychology and physical kinematics, a double-elliptical risk field is established on a two-dimensional plane with the vehicle's center of mass as the reference point. This risk field simultaneously considers physical risk factors (i.e., spatial distance and relative speed between vehicles) and psychological risk factors (i.e., the subjective risk sensitivity and reaction differences of the driver or automatic control system).

[0035] For any subject vehicle At any moment The level of risk is determined by all potential sources of conflict around it. The combined risk impact (including vehicles ahead, vehicles in adjacent lanes, and static obstacles) is represented as: in: For the main vehicle at time The overall risk level; The weights for different conflict sources are used to characterize the relative importance of the risk sources; The risk-repelling force of the main vehicle against each source of conflict; The number of potential conflict sources identified for the main vehicle.

[0036] Coupled modeling of physical and psychological risks includes: At the physical level, risk repulsion force This is related to physical parameters such as the relative distance, relative speed, and mass between the vehicle and the risk source; at the psychological level, different driving subjects (such as ACC, AV, HV) have different risk judgments, manifested as different perceived intensities of the same physical risk. Therefore, the repulsive force of the vehicle towards each risk source is defined as: in, Parameters that characterize the differences in risk perception judgment among ACC, AV, and HV vehicles. As a source of risk quality The main vehicle (this vehicle) quality As a source of risk Received from this vehicle The repulsive field strength.

[0037] In this embodiment, parameters Different values ​​can be taken depending on the vehicle type, for example: human-driven vehicles (HV): This indicates a relatively direct risk assessment; Adaptive Cruise Control (ACC) vehicles: This indicates relatively stable perception but a slightly delayed response; Autonomous Vehicles (AVs): This indicates a smoother decision-making process but lower risk sensitivity.

[0038] Furthermore, the psychophysical field strength experienced by the risk source This can be further broken down into the sum of physical risk and psychological weight: The first item is the distance attenuation item, which reflects the impact of physical spatial distance on risk perception; the second item is the time to collision (the ratio of distance to relative vehicle speed), which reflects the urgency of potential conflict; and the third item is the psychological orientation item, which reflects the differences in driver perception of targets at different azimuth angles. Let be the weighting coefficient, satisfying ; These are the distance, time, and azimuth attenuation constants, respectively. This refers to the azimuth angle of the risk source relative to the direction of travel of the main vehicle. When the risk source is outside the driver's field of vision (e.g., ... The psychological perception item is set to 0, and only the physical risk effect is retained.

[0039] Risk field output and differentiated results: Through the above calculations, the risk level of the main vehicle at any time can be obtained. The results were then weighted according to vehicle type to obtain differentiated perception results: Risk level This serves as the risk perception input parameter for the vehicle in subsequent longitudinal following model and lateral lane-changing model.

[0040] Among them, high-risk values ​​correspond to an increased perceived danger situation for the driver, which manifests as a stronger tendency to decelerate in the longitudinal model and a reduced willingness to change lanes or a delayed execution of lane changes in the lateral model.

[0041] The psychophysical risk field model constructed in this step can calculate the risk level of each vehicle in real time with a time resolution of 0.1 s, comprehensively reflecting the driver's spatial safety distance, potential conflict time, and psychological sensitivity. This model retains the determinism of physical constraints while reflecting the subjective differences in driving behavior, providing a unified quantitative indicator for subsequent longitudinal and lateral behavior modeling based on risk perception.

[0042] Furthermore, the process of constructing a risk-decision coupled longitudinal behavioral model includes: Calculate the basic longitudinal acceleration based on the relative distance and speed difference between the vehicle and the vehicle in front, as well as the vehicle's speed; and construct the basic longitudinal motion function based on the basic longitudinal acceleration. Establish a risk perception correction function, which takes the real-time perceived risk value as input and outputs a risk correction term. The basic longitudinal action function and the risk perception correction function are coupled to generate the final longitudinal acceleration command, and the longitudinal following behavior is dynamically adjusted through risk perception.

[0043] Furthermore, the process of constructing the longitudinal car-following model for mixed traffic flow in this embodiment includes: For ACC vehicles, AV vehicles, and HV vehicles, a risk perception-decision coupling longitudinal model is adopted. The risk perception of different vehicles is placed in advance and combined with the basic decision model to jointly construct the longitudinal behavior output of the vehicle. Specifically, the risk perception parameters are placed in advance in the behavior decision, driving the longitudinal acceleration and deceleration action decision, thereby completing the risk-decision coupling longitudinal behavior output.

[0044] Specifically, longitudinal following models are constructed for ACC, AV, and HV vehicles, respectively, in which risk perception parameters are incorporated into behavioral decision-making. The model uses a basic longitudinal action function. Risk perception-decision mapping function coupling.

[0045] This embodiment uses an independent risk perception parameter. This describes the perception process of different vehicle types, coupled with the action execution process described by the conventional vehicle car-following model. It can be represented as: in, For vehicles in longitudinal acceleration at time t, For the basic vertical action function, Characterize the impact of risk perception on the final action execution. For the risk-decision mapping effect parameters, To characterize the coupling relationship between basic longitudinal actions and the impact of risk perception.

[0046] After the model is built, it is necessary to use the extracted ACC, AV, and HV vehicle trajectories. , , It can contain time series of multiple observation indicators. The risk perception-decision coupling longitudinal model of ACC, AV and HV vehicles is calibrated. The longitudinal behavior and lateral behavior are calibrated separately. The longitudinal and lateral behaviors are calibrated using a single objective, and the longitudinal and lateral behavior parameter sets are obtained.

[0047] More specifically, the basic vertical action function is established: The basic longitudinal motion function adopts the IDM form to describe the vehicle's motion based on the relative distance. Speed ​​difference and its own speed Longitudinal acceleration behavior: in: This is the maximum acceleration; For the desired speed; The speed index (usually taken as 4); The desired safe distance; Minimum following distance; To anticipate the time difference; Reduce speed for comfort.

[0048] Risk perception correction function established: To demonstrate the impact of risk perception on longitudinal decision-making, this invention outputs the risk field. A car-following model is introduced, and acceleration is dynamically adjusted through a risk correction function. It is defined as follows: In the formula, The sensitivity coefficient for the IDM term. and These represent the sensitivity coefficients when the driver's risk level is below and above the expected risk level, respectively. and The acceleration coefficients are those for when the driver's risk level is below and above the expected risk level. and This is the risk impact coefficient. This represents the driver's expected risk value. and The separate representation is for better explanation of the model mechanism; during calibration, it can be regarded as a whole. Perform calibration.

[0049] Longitudinal car-following model parameter calibration: Using the results obtained in step S1 , , Trajectory data is used to calibrate longitudinal risk-decision coupling models for various vehicle types. The calibration objective is to minimize the model output margin. Measured distance The mean square error between them is expressed as: In the formula, The optimal set of parameters determined by calibration. For the number of observation points, The relative mean square percentage error of the spacing is used. A genetic algorithm is selected for calibration.

[0050] Furthermore, the process of constructing a risk-behavior coupled lane-changing decision-making model includes: Based on real-time perceived risk values ​​and traffic state variables of the current lane and the target lane, a discrete choice model is used to calculate the probability of a vehicle's willingness to change lanes. When the probability of lane change intention reaches a preset threshold, the gap acceptance probability is calculated based on the gap status in front and behind the target lane, the speed difference, and the real-time perceived risk value. When the probability of accepting the gap exceeds the execution threshold, a lane-changing action is triggered. At the same time, the safety constraints are adaptively adjusted based on the real-time perceived risk value to suppress lane-changing behavior in high-risk scenarios.

[0051] Furthermore, the process of constructing the lateral lane-changing model for mixed traffic flow vehicles in this embodiment includes: Considering the impact of risk perception on vehicle lane-changing decisions, risks in the two-dimensional plane are incorporated into the lane-changing decision models of different vehicles. The decision-making behavior parameters of different vehicles are different to characterize the differences in lane-changing behavior. This embodiment constructs a lateral lane-changing model. In the two key sub-processes of lane-changing intention generation and lane-changing gap selection, the two-dimensional plane risk value is used as one of the decision inputs. Different vehicles exhibit heterogeneous lane-changing behaviors based on their risk perception sensitivity, lane-changing cost and other behavioral parameters.

[0052] More specifically, with independent risk perception parameters The perception process of different types of vehicles is described and coupled with the action execution process described by the conventional vehicle car-following model. The lateral lane-changing behavior decision sub-model further includes the following steps: Based on the calculations for each vehicle At any moment The risk value in the two-dimensional plane is used as one of the input factors in the lane change motivation generation stage and the lane change gap selection stage, so that the vehicle's lane change decision depends not only on traditional traffic state variables, but also on the perceived safety / risk characteristics. The motivation generation stage for free lane changing uses a discrete choice model (Logit model) to calculate vehicle... willingness to change lanes The utility function of the model includes risk variables. and vehicle behavior parameters Each vehicle can have different risk avoidance sensitivities, thus modeling heterogeneous lane-changing behavior; After the motivation generation stage, if Once the trigger threshold is reached, the lane change clearance selection phase begins. In this phase, the clearance status in front of and behind the target lane, speed difference, and risk value are considered. and behavioral parameters Common input gap acceptance model, calculate gap acceptance probability The formula explicitly shows the impact of risk variables on the probability of accepting the gap, thereby realizing a risk-behavior coupled lane-changing decision-making process; when When the set execution threshold is exceeded, a lane-changing action is triggered. At the same time, a safety lower limit that adapts to risk (such as minimum TTC / maximum deceleration constraint) is set to ensure that lane-changing is suppressed in high-risk scenarios.

[0053] More specifically, this embodiment constructs a lateral lane-changing behavior model under mixed traffic flow conditions, comprehensively considering the vehicle's traffic state, risk perception level, and behavioral differences, by introducing a risk perception value. The system comprises two key sub-modules: lane-changing motivation generation and gap selection, enabling a two-layer coupled decision-making process between risk and behavior. Specifically, it includes the following sub-steps: Motivation generation (Logit structure): For lane changing behavior, vehicles At any moment The lane-changing intention is described using a Logit model based on discrete choice theory: in, For vehicles At any moment The utility function is defined as: in, The coefficients to be calibrated in the model. The speed difference between the vehicle in front and the vehicle in front in the current lane. The duration for which this vehicle remains at a low speed (lower than the speed of the vehicle in front) in the current lane. Let be the average speed of vehicles in adjacent lanes within the target lane. Therefore... This reflects the potential for increased speed when changing lanes. For vehicles The risk level of the current lane. If changing lanes to the target lane, the risk value (or estimated risk) is calculated based on the traffic conditions in the target lane. These are vehicle behavior characteristic parameters (e.g., sensitivity to lane-changing intentions, conservative / aggressive type).

[0054] Gap selection: Define "total clearance" This refers to the available headroom (or distance) between vehicles in front and behind the target lane. Define the "critical gap". For vehicles The minimum acceptable gap can be expressed as: in, , The speed difference between this vehicle and the vehicle in front / behind in the target lane This represents the current risk value of the vehicle (or the estimated risk for the target lane). For the random error term (usually assumed to follow a normal or log-normal distribution), when If the vehicle accepts the gap, it will change lanes; otherwise, it will refuse or wait for the next gap.

[0055] Risk constraints: To ensure driving safety in high-risk scenarios, this invention introduces a risk-adaptive safety constraint mechanism into the lane-changing execution logic. When At higher levels (e.g.) The model automatically increases the minimum safe clearance or time required for lane changing to the collision lower limit. The specific constraints are as follows: in The basic safety threshold (typically 2.0s) is used. This is the risk amplification factor (taken as 1 to 2).

[0056] Model calibration and output: Based on the extracted lane-change event segments, the parameters of the above-mentioned motivation generation model and gap acceptance model are calibrated respectively. The maximum likelihood estimation (MLE) method is used, with the actual lane-change occurrence as a binary response variable (0–1), to fit the parameter set: The calibration results are stored in the horizontal behavior database of the simulation platform.

[0057] Furthermore, the process of performing dynamic simulation of mixed traffic flow includes: The longitudinal behavior model, lane-changing decision model, and differentiated risk perception parameters are loaded into the simulation platform, and the vehicle state, simulation time step, and total duration are initialized. By configuring the ratio of high-level adaptive cruise vehicles to autonomous vehicles through the penetration rate setting module, mixed traffic scenarios with different levels of automation can be formed. When the simulation is running, the real-time perceived risk value of all vehicles is updated sequentially at each time step, the longitudinal motion state is updated according to the longitudinal behavior model, and the lateral lane change behavior is judged and executed according to the lane change decision model. The process is iterated until the simulation ends. Meanwhile, key operational indicators, including vehicle speed, traffic flow, density, lane change frequency, and risk exposure level, are recorded and calculated during the simulation process.

[0058] Furthermore, this embodiment integrates the risk perception model, longitudinal car-following model, and lateral lane-changing model into a unified simulation platform to achieve operational simulation and performance evaluation of different types of vehicles in mixed traffic flow. Specifically, it includes the following sub-steps: Model loading and initialization: Import the obtained vehicle models of various types into the simulation platform, and establish vehicle attribute tables, behavior parameter tables and risk perception interfaces through the model loading module.

[0059] During the simulation initialization phase, the platform reads the longitudinal parameters. Horizontal parameters and risk sensitivity coefficient Each vehicle is assigned a unique identifier (ID), initial position, speed, lane number, and other basic state variables. A simulation timeline is also established, with step sizes set. Total simulation duration This provides a unified time reference for subsequent operations.

[0060] Penetration rate setting: The penetration rate setting module defines the proportion of each type of vehicle in a mixed-traffic scenario to create different simulation combinations. The module can set at least three penetration rate combinations: Pure HV vehicle scenario: ACC / AV ratio = 0%, all vehicles are traditionally driven by humans; Mixed penetration scenario: ACC / AV ratio = 10%-50%, the rest are HV vehicles; High automation penetration scenario: ACC / AV ratio ≥ 50%, representing traffic flow dominated by automated vehicles. The simulation platform randomly assigns vehicle types according to a set ratio while maintaining consistent total road traffic flow to ensure comparability of results across different penetration scenarios.

[0061] Behavioral parameter configuration: Before the simulation, the risk avoidance sensitivity parameters for each type of vehicle were configured using the behavior parameter configuration module. , Configure differentiated settings. Among them: Set a higher risk aversion factor for HV vehicles (such as...) (in order to reflect its conservative driving characteristics); Set moderate risk avoidance parameters for ACC vehicles (such as...) ); Set a low risk sensitivity coefficient for AV vehicles (e.g.) This reflects its stable control strategy.

[0062] After configuration, the platform writes the parameters into the behavioral attribute table of the vehicle model, enabling different types of vehicles to exhibit heterogeneous risk responses and driving behaviors during simulation.

[0063] Simulation Execution and Status Update: The simulation platform operates according to a fixed time step. Iterative updates are performed every 0.1 s or 0.5 s (e.g., every 0.1 s or 0.5 s), specifically including the following process: Risk value update: Call the psychological-physical risk field module to calculate the risk value of each vehicle at time 10:00. Two-dimensional risk value And pass it as input parameters to the longitudinal and lateral behavior models.

[0064] Vertical Behavior Judgment: Based on a Risk-Decision Coupling Vertical Model Calculate longitudinal acceleration It also updates vehicle position and speed.

[0065] Lateral behavior judgment: Calculate the probability of lane-changing motivation for each vehicle. and gap acceptance probability If the lane-changing conditions are met, then lane switching will be performed.

[0066] Status refresh: After each time step, the platform simultaneously updates the position, speed, acceleration, risk value and lane number of all vehicles to achieve longitudinal and lateral linkage control.

[0067] Indicator output and performance evaluation: During the simulation, the platform records vehicle status and event information in real time and calculates key operational indicators, including: Lane change trigger rate: Total number of lane changes / Total simulation time, used to measure the frequency of lateral behavior; Percentage of vehicles exposed to high risk: Number of vehicles in a high-risk state / Total number of vehicles, where a high-risk state is defined as... ; Average vehicle delay : The average delay for all vehicle journeys; Vehicle-specific traffic improvement rate: It is used to measure the improvement effect of mixed traffic flow on traffic capacity.

[0068] The above indicators serve as a comprehensive basis for evaluating traffic flow efficiency, safety, and stability, and an evaluation report is generated after the simulation is completed.

[0069] Simulation and sensitivity analysis of control strategies: The simulation platform can perform sensitivity analysis by overlaying different traffic control strategies on the basis of model integration, so as to verify the applicability and stability of the risk perception model.

[0070] The main strategies include the following three categories: Variable Speed ​​Limit (VSL) strategy: Set up dynamic speed limit modules at road entrances and middle sections to simulate the impact of speed limits on traffic flow, risk exposure, and volatility; Lane closure strategy: By closing specific lanes or adjusting right-of-way, the traffic flow redistribution and risk redistribution process is simulated in scenarios such as accidents and construction. CAV Cooperative Control Strategy: Simulate vehicle-to-everything (V2X) cooperative deceleration and merging control under conditions of high automation penetration, and evaluate the impact of communication delay and cooperative ratio on traffic stability.

[0071] The system automatically generates sensitivity analysis reports for each strategy scenario, and outputs indicators including average speed change rate, risk exposure reduction rate, and stability improvement ratio, which are used to evaluate the control effect of mixed traffic flow.

[0072] This embodiment demonstrates that the method can effectively improve simulation accuracy and stability by using the risk field output as a dynamic correction term for the vehicle behavior model parameters and verifying it through trajectory data calibration.

[0073] Furthermore, the simulation platform integrated in this embodiment enables unified modeling of the risk perception and behavioral decision-making processes of various types of vehicles, supporting dynamic simulation of mixed traffic flow under multiple scenarios and strategies.

[0074] Furthermore, the simulation platform includes: The model loading module is used to import longitudinal behavior models, lane change decision models, and risk perception parameters; The penetration rate setting module is used to set the ratio of high-level adaptive cruise vehicles to autonomous vehicles in traffic flow; The behavior parameter configuration module is used to configure risk avoidance sensitivity parameters for different types of vehicles. The risk-decision coupling module is used to input the vehicle's real-time perceived risk value into the longitudinal behavior model and lane-changing decision model at each simulation time step; The simulation results analysis module is used to collect simulation data and calculate key operating indicators.

[0075] Furthermore, in this embodiment, the various modules of the simulation platform are used to import the constructed vehicle models of different types, set the penetration rate ratio of different types of vehicles, initialize or adjust the longitudinal and lateral behavior parameters of each type of vehicle, input the two-dimensional planar risk value of each vehicle into its decision model in real time, and output key operating indicators including vehicle speed, flow rate, density, lane change frequency, and risk exposure level.

[0076] Furthermore, the penetration rate setting module is used to set the following scenarios: In scenarios where vehicles are driven entirely by humans, the proportion of high-level adaptive cruise control vehicles and autonomous driving vehicles is 0%. In a mixed penetration scenario, the ratio of high-level adaptive cruise control vehicles to autonomous driving vehicles is 10% to 50%. In scenarios with high automation penetration, the proportion of high-level adaptive cruise vehicles and autonomous vehicles should not be less than 50%.

[0077] Furthermore, the process by which the behavioral parameter configuration module configures risk aversion sensitivity parameters for different types of vehicles includes: Differentiated risk avoidance sensitivity parameters were set for human-driven vehicles, high-level adaptive cruise control vehicles, and autonomous vehicles, so that the risk avoidance sensitivity parameter value of human-driven vehicles was the highest, the risk avoidance sensitivity parameter value of autonomous vehicles was the lowest, and the risk avoidance sensitivity parameter value of high-level adaptive cruise control vehicles was in between.

[0078] Furthermore, the behavior parameter configuration module in this embodiment allows for the configuration of risk aversion sensitivity parameters for each type of vehicle before simulation. , Differentiated settings are implemented to achieve heterogeneous behavior modeling.

[0079] Furthermore, during dynamic simulation runtime, the following operations are performed at fixed time steps: The two-dimensional risk field is invoked to calculate the real-time perceived risk value of each vehicle at the current moment; Real-time perceived risk values ​​are transmitted to the longitudinal behavior model and lane-changing decision model of each vehicle; The vehicle position, speed, and lane number are updated based on the longitudinal acceleration output by the longitudinal behavior model and the lateral commands output by the lane change decision model. At the end of each time step, refresh the status information of all vehicles and determine whether the simulation termination condition is met.

[0080] Furthermore, the simulation platform in this embodiment operates at fixed time steps during the simulation process. Update the position, speed status, and perceived two-dimensional plane risk values ​​of all vehicles every 0.1s or 0.5s (e.g., 0.1s or 0.5s). At each time step, it determines whether each vehicle has triggered a lane-changing or following behavior.

[0081] Furthermore, key operational metrics also include lane change trigger rate (total lane change count / simulation time), percentage of high-risk exposed vehicles (number of high-risk vehicles / total number of vehicles), and average vehicle delay. and the rate of increase in traffic due to vehicle model heterogeneity ( ).

[0082] Furthermore, the simulation platform in this embodiment can also generate sensitivity analysis reports based on different traffic control strategies (including variable speed limits, lane closures, and CAV collaborative control) to evaluate the control effect of the mixed traffic flow.

[0083] This embodiment discloses an intelligent connected mixed traffic flow modeling method that integrates risk perception. It collects trajectory and environmental data from high-level adaptive cruise control (ACC), automated driving (AV), and human-driven (HV) vehicles in mixed traffic scenarios. After preprocessing, it extracts car-following and lane-changing event segments. A risk field is constructed based on the vehicle centroid to obtain the real-time perceived risk value for each vehicle. Risk sensitivity parameters for different vehicle types are set to correct car-following and lane-changing behaviors, thus forming a heterogeneous traffic flow evolution algorithm that integrates risk perception. The risk perception results are pre-loaded into the longitudinal car-following model and coupled with the basic action function to form a risk-decision coupled model. In the lateral lane-changing model, risk is introduced into two sub-modules: motivation generation and gap acceptance, constructing a risk-behavior coupled lane-changing decision process. Finally, the longitudinal and lateral models are integrated into a simulation platform to simulate and evaluate mixed traffic flow under different penetration rate combinations. This embodiment can achieve accurate modeling of heterogeneous driving behaviors, improving the realism of mixed traffic simulation and traffic safety evaluation capabilities.

[0084] The above are merely preferred embodiments of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A method for modeling intelligent connected mixed traffic flow with perception risk awareness, characterized in that, The method comprises the steps of: Collecting trajectory and environmental data of different levels of automatic driving vehicles and human driving vehicles in mixed traffic scenes, and preprocessing the collected data; Based on the preprocessed data, a two-dimensional risk field is constructed with the vehicle centroid as the core, the real-time perception risk value of each vehicle is calculated, and the risk sensitivity parameters of different types of vehicles are set to distinguish the differentiated risk perception results; The real-time perception risk value is preposed to the longitudinal following model, coupled with the basic longitudinal action function, and a risk-decision coupled longitudinal behavior model is constructed; The real-time perception risk value is introduced into the lane changing motive generation module and the lane changing gap selection module of the lateral lane changing model, and a risk-behavior coupled lane changing decision model is constructed; The risk-decision coupled longitudinal behavior model, the risk-behavior coupled lane changing decision model and the differentiated risk perception parameter set are integrated into a unified simulation platform, the penetration rate combination of different types of vehicles is set, the dynamic simulation of mixed traffic flow is performed, and the simulation running results are output.

2. The method of claim 1, wherein the process of collecting trajectory and environmental data of different levels of automatic driving vehicles and human driving vehicles in mixed traffic scenes, and preprocessing the collected data comprises: Synchronously collecting trajectory and environmental data of high-level adaptive cruise vehicles, automatic driving vehicles and human driving vehicles, the trajectory and environmental data including vehicle position, speed, acceleration, lateral offset, lane number, front and rear vehicle distance and surrounding traffic participant state; Applying Savitzky-Golay smoothing filter to the collected vehicle position and speed signals to eliminate noise, and calculating vehicle lateral offset and relative motion state with the front vehicle in a predetermined sliding window; According to the continuous change characteristics of the vehicle lateral offset, lane number and relative motion state with the front vehicle, identifying and extracting the following event section and the lane changing event section, eliminating the segments with a length lower than a preset threshold, and obtaining a structured trajectory data set.

3. The method of claim 1, wherein the process of constructing a two-dimensional risk field and calculating the real-time perception risk value of each vehicle comprises: Establishing a double-ellipse two-dimensional risk field with the centroid of the subject vehicle as the origin, superimposing the repulsive force of the subject vehicle on each potential conflict source to obtain the overall risk value; Calculating the repulsive force of the subject vehicle on each conflict source, the repulsive force including lateral and longitudinal relative position, vehicle mass and risk perception judgment differentiation parameters; According to the vehicle type, setting different risk perception judgment differentiation parameter values, weighting the overall risk value to obtain differentiated risk perception results as input parameters of the subsequent longitudinal and lateral behavior models.

4. The method of claim 1, wherein the process of constructing a risk-decision coupled longitudinal behavior model comprises: Calculating the basic longitudinal acceleration according to the relative distance, speed difference between the host vehicle and the front vehicle, and the speed of the host vehicle, and constructing a basic longitudinal action function according to the basic longitudinal acceleration; Establishing a risk perception correction function, the risk perception correction function taking the real-time perception risk value as input and outputting a risk correction term; ​ ​ ​ The base longitudinal action function is coupled with a risk perception correction function to generate a final longitudinal acceleration instruction, and the longitudinal following behavior is dynamically adjusted by risk perception.

5. The method of claim 1, wherein, The process of constructing the risk-behavior coupled lane-changing decision model includes: Based on the real-time perceived risk value and the traffic state variables of the current lane and the target lane, the lane-changing willingness probability of the vehicle is calculated using a discrete choice model; When the lane-changing willingness probability reaches a preset threshold, the gap acceptance probability is calculated according to the front and rear gap states, the speed difference, and the real-time perceived risk value of the target lane; When the gap acceptance probability exceeds the execution threshold, the lane-changing action is triggered, and the safety constraint condition is adaptively adjusted according to the real-time perceived risk value to suppress the lane-changing behavior in high-risk scenarios.

6. The method of claim 1, wherein, The process of performing dynamic simulation of mixed traffic flow includes: loading the longitudinal behavior model, the lane-changing decision model, and the differentiated risk perception parameters into a simulation platform, and initializing the vehicle state, the simulation time step, and the total time length; configuring the proportion of high-level adaptive cruise vehicles and autonomous vehicles through a penetration rate setting module to form mixed traffic scenarios with different automation levels; When the simulation is running, the real-time perceived risk value of all vehicles is updated at each time step, the longitudinal motion state is updated according to the longitudinal behavior model, and the lateral lane-changing behavior is judged and executed according to the lane-changing decision model, and the loop iteration is repeated until the simulation is completed; At the same time, key operational indicators including vehicle speed, flow, density, lane-changing frequency, and risk exposure level are recorded and calculated during the simulation process.

7. The method of claim 1, wherein, The simulation platform includes: a model loading module for importing the longitudinal behavior model, the lane-changing decision model, and the risk perception parameters; a penetration rate setting module for setting the proportion of high-level adaptive cruise vehicles and autonomous vehicles in the traffic flow; a behavior parameter configuration module for configuring risk aversion sensitivity parameters for different types of vehicles; a risk-decision coupling module for inputting the real-time perceived risk value of the vehicle to the longitudinal behavior model and the lane-changing decision model at each simulation time step; a simulation result analysis module for collecting simulation data and calculating key operational indicators.

8. The method of claim 7, wherein, The penetration rate setting module is used to set the following scenarios: pure human-driven vehicle scenario, where the proportion of high-level adaptive cruise vehicles and autonomous vehicles is 0%; mixed penetration rate scenario, where the proportion of high-level adaptive cruise vehicles and autonomous vehicles is 10% to 50%; high automation penetration rate scenario, where the proportion of high-level adaptive cruise vehicles and autonomous vehicles is not less than 50%.

9. The method of claim 7, wherein, The process of configuring risk aversion sensitivity parameters for different types of vehicles by the behavior parameter configuration module includes: The risk avoidance sensitivity parameters are respectively set for human-driven vehicles, high-level adaptive cruise vehicles and autonomous vehicles, so that the risk avoidance sensitivity parameter value of the human-driven vehicles is the highest, the risk avoidance sensitivity parameter value of the autonomous vehicles is the lowest, and the risk avoidance sensitivity parameter value of the high-level adaptive cruise vehicles is between the two.

10. The method of claim 1, wherein, The dynamic simulation is executed at fixed time steps, and the following operations are performed at each time step: The two-dimensional risk field is called to calculate the real-time perceived risk value of each vehicle at the current time; The real-time perceived risk value is passed to the longitudinal behavior model and the lane-changing decision model of each vehicle; The vehicle position, speed and lane number are updated according to the longitudinal acceleration output by the longitudinal behavior model and the lateral instruction output by the lane-changing decision model; The state information of all vehicles is refreshed at the end of each time step, and it is judged whether the simulation termination condition is met.

Images