Intelligent network connection vehicle social contact cooperative interaction control method and system in mixed traffic scene

Abstract

The invention relates to the technical field of vehicle cooperative interaction control, in particular to an intelligent network connection vehicle social cooperative interaction control method and system in a mixed traffic scene, and the method comprises the steps: obtaining a spatial feature vector and a time sequence feature vector; obtaining a comprehensive feature vector through the spatial feature vector and the time sequence feature vector; performing linear combination according to the comprehensive feature vectors to obtain a reward function; obtaining an optimal weight vector through inverse reinforcement learning according to the reward function; the SVO angle value of the HDV is estimated through the optimal weight vector, and the SVO angle estimated value of the CAV is obtained; sVO parameters are optimized and dynamically adjusted through the Stackelberg game according to the SVO angle estimation value of the CAV, and optimal response parameters of the HDV are obtained; and according to the IDM model and the optimal response parameter of the HDV, simulating social perception and response of a human driver so as to adjust the driving behavior. According to the invention, the accuracy of vehicle cooperative interaction control is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of vehicle collaborative interaction control, and in particular to a method and system for social collaborative interaction control of intelligent connected vehicles in mixed traffic scenarios. Background Art

[0002] In today's mixed traffic environments, where connected and autonomous vehicles (CAVs) and human-driven vehicles (HDVs) coexist, existing interactive control strategies have significant limitations. Traditional approaches employ a one-way adaptation model, where the CAV passively adapts to the HDV's behavior while ignoring the HDV's dynamic response to the CAV, resulting in insufficient synergy between the two. Furthermore, existing solutions often rely on fixed social value orientation (SVO) decisions, failing to dynamically adjust to traffic flow, penetration, or driving style, making it difficult to balance real-time safety, efficiency, and comfort requirements. Furthermore, existing HDV behavior modeling is overly simplified and fails to accurately capture the individual characteristics and social preferences of human drivers, leading to increased interaction uncertainty and traffic flow fluctuations. These issues collectively create a vicious cycle of one-way CAV adaptation, unmodeled HDV responses, and unoptimized fixed SVOs, ultimately limiting the overall performance of the mixed traffic system. This paper, through a two-layer closed-loop interactive control framework enabled by dynamic SVOs, achieves deep, two-way collaboration between CAVs and HDVs, thus overcoming existing technical bottlenecks. Summary of the Invention

[0003] The present invention provides a social collaborative interactive control method and system for intelligent connected vehicles in mixed traffic scenarios, which is used to solve the problems of one-way adaptation, inability to dynamically adjust, and neglect of driver's personalized characteristics in the interactive control strategy when existing intelligent connected vehicles coexist with manually driven vehicles.

[0004] The purpose of the present invention can be achieved through the following technical solutions: The first aspect of the present invention is to provide a method for controlling social collaborative interaction of intelligent connected vehicles in a mixed traffic scenario, comprising: Acquire vehicle distance data, speed data, and acceleration data during driving; fuse all types of data based on the vehicle distance data to obtain a spatial feature vector; obtain four types of time series features of the vehicle, namely, speed trend, acceleration pattern, lane change intention, and headway change, and obtain a time series feature vector based on the time series features; obtain a comprehensive feature vector based on the spatial feature vector and the time series feature vector; obtain a reward function by performing a linear combination based on the comprehensive feature vector; obtain an optimal weight vector based on the reward function through inverse reinforcement learning; estimate the SVO angle value of the HDV using the optimal weight vector to obtain an estimated SVO angle value of the CAV; The SVO angle estimation value of CAV is used to dynamically adjust the SVO parameters through Stackelberg game optimization to obtain the optimal response parameters of HDV; According to the IDM model and the optimal response parameters of HDV, the social perception and response of human drivers are simulated to adjust driving behavior.

[0005] Furthermore, the method of fusing all types of data based on vehicle distance data to obtain a spatial feature vector includes: The data standardization operation is completed by performing linear normalization on the data of each category, and each standardized data is combined into a standardized feature vector; that is, the vehicle distance standardized feature vector, the speed standardized feature vector and the acceleration standardized feature vector are obtained; Constructing a feature space by using the vehicle distance normalized feature vector, the speed normalized feature vector and the acceleration normalized feature vector; In the distance-based attention mechanism, the similarity between each data in each normalized feature vector and each data in the vehicle distance normalized feature vector is calculated. The similarity is specifically expressed as:

[0006] Where, represents the first vehicle distance normalized feature vector The data points corresponding to each data point in the feature space are: Indicates the The first of the normalized eigenvectors The data points corresponding to each data point in the feature space are: represents the Euclidean distance between two points, Indicates the similarity between two data; According to the similarity between each data in each standardized feature vector and all data in the vehicle distance standardized feature vector, The function calculates the weight of each data in each standardized feature vector relative to each data in the vehicle distance standardized feature vector; The process of obtaining each data in the spatial feature vector is specifically expressed by the formula:

[0007] Where, represents the first vehicle distance normalized feature vector data, represents the first vehicle distance normalized feature vector The data is relative to the first in the vehicle distance normalized feature vector The weight of the data; represents the first normalized eigenvector of acceleration data, represents the first normalized eigenvector of acceleration The data is relative to the first in the vehicle distance normalized feature vector The weight of the data, represents the first in the velocity normalized eigenvector data, represents the first in the velocity normalized eigenvector The data is relative to the first in the vehicle distance normalized feature vector The weight of the data; represents the number of data in the normalized feature vector of vehicle distance, represents the number of data in the acceleration normalized feature vector, represents the number of data in the velocity normalized feature vector, Represents the first feature vector in the space data.

[0008] Furthermore, the four types of time series features of the vehicle's speed trend, acceleration pattern, lane change intention, and headway change are obtained, and a time series feature vector is obtained through the time series features, including: A preset length is selected and the time series observation window is determined by the preset length. The speed trend is obtained by linear regression of the speed sequence within the time series observation window. The specific formula is:

[0009] Where, Indicates the first A moment of time, Represents the mean value of the time of all moments in the time series observation window, Indicates the first The speed of a moment, represents the mean value of the speed at all moments in the time series observation window, Represents the number of all moments in the time series observation window, Indicates speed trend; The acceleration pattern is calculated by dividing the speed difference between the two endpoints within the time series observation window by the time difference. Lane change intention is determined by vehicle trajectory and steering angle changes. Headway changes are extracted by calculating the time interval between the front and rear vehicles and their changes. The time series feature vector is specifically expressed as:

[0010] Where, Indicates the speed trend, Indicates the acceleration mode, Indicates intention to change lanes. Indicates the change in headway distance, represents the time series feature vector.

[0011] Furthermore, obtaining a comprehensive feature vector through the spatial feature vector and the temporal feature vector includes:

[0012] Where, represents the spatial eigenvector, represents the time series feature vector, represents the comprehensive feature vector, represents the weight of the spatial eigenvector; where, Acquisition was determined through cross-validation.

[0013] Furthermore, performing linear combination according to the comprehensive feature vector to obtain a reward function; obtaining an optimal weight vector according to the reward function through inverse reinforcement learning; estimating the SVO angle value of the HDV by using the optimal weight vector to obtain the SVO angle estimation value of the CAV, includes: Determine the state space and action space, generate trajectories through the state space and action space, and each trajectory ; in, Indicates the first state of each trajectory and the first action performed in the first state, Indicates the first state and The first state executed Action, is a trajectory consisting of states and actions; In inverse reinforcement learning, the reward function is defined as the comprehensive feature vector The linear combination of ; specifically expressed as:

[0014] Where, represents the comprehensive feature vector, represents the reward function value, represents the weight vector to be learned, Indicates the transposition of the weight vector to be learned; Among them, the specific learning process of the weight vector to be learned in the linear combination is: The probability distribution of each trajectory can be expressed as:

[0015] Where, Indicates the first The comprehensive feature vector of the corresponding state and action at step time; represents the weight vector to be learned, Indicates the transposition of the weight vector to be learned, Indicates the total step length corresponding to each trajectory, represents an exponential function with a natural constant as the base, represents the partition function, Represents each trajectory The probability value of Indicates the first The reward value of the step; The partition function is specifically expressed as:

[0016] Where, Indicates the Track No. The comprehensive feature vector of the corresponding state and action at step time, Indicates generating the first Tracks probability; represents the number of all trajectories, Indicates the The first track The reward value of the step; The relative entropy projection method is used to perform convergence analysis on the weight vector to be learned and determine the final weight vector; specifically, it is expressed as:

[0017] Where, Represents each trajectory The probability value of Represents the empirical distribution probability value of the observed trajectory; express Relative to The relative entropy of represents the weight vector in the iterative process, represents the optimal weight vector; Indicates that the weight vector corresponding to the minimum relative entropy is found as the optimal weight vector; The SVO angle value of the HDV is estimated by using the optimal weight vector to obtain an SVO angle estimation value.

[0018] Furthermore, the method of dynamically adjusting SVO parameters by using the estimated SVO angle of the CAV through Stackelberg game optimization to obtain the optimal response parameters of the HDV includes:

[0019] Where, Indicates the The system status at a moment, Indicates that CAV SVO angle estimation value at the moment; Indicates that HDV The acceleration corresponding to the optimal response after observing the choice of CAV at each moment, represents the utility function of CAV; Indicates the SVO angle estimation value corresponding to the maximum utility function, Indicates that CAV The optimal SVO angle after the game at each moment; Among them, the utility function Build access through the unity of security, efficiency, and interactivity.

[0020] Furthermore, the method of simulating the social perception and response of a human driver based on the IDM model and the optimal response parameters of the HDV to adjust driving behavior includes: The optimal SVO angle is used as the target SVO angle value calculated for the current scenario in the SVO dynamic adjustment formula. The current SVO angle value is then updated using the SVO dynamic adjustment formula to obtain an updated current SVO angle value. The IDM parameters are updated using the SVO parameter mapping relationship based on the updated current SVO angle value to obtain updated IDM parameters. The expected acceleration calculation formula is used to calculate the updated IDM parameters to obtain the expected acceleration. The final output acceleration is obtained through smoothing. At the same time, the system makes lane change decisions using a lane change utility function, calculates the lane change utility, and compares it with a threshold to determine whether to execute the lane change operation. Finally, the system performs a safety check and triggers the emergency braking mechanism when the safety conditions are not met. Finally, the output acceleration is fed back to the CAV decision-making process. Among them, the lane-changing utility function is determined by safety and necessity; Among them, the SVO dynamic adjustment formula is specifically expressed as:

[0021] Where, Indicates the current time The urgency of the situation, Indicates the target SVO angle value calculated based on the current scene, represents the adjustment gain coefficient, represents the rate of change of SVO, Indicates the current time SVO angle; The urgency of the scene is obtained by weighted average of traffic density, safety risk and time pressure; The parameter mapping relationship of SVO is specifically expressed as follows:

[0022]

[0023]

[0024] Where, represents the baseline comfort acceleration, represents the benchmark expected time interval, Indicates the minimum safe distance between vehicles. represents the sine function, represents the SVO angle, Indicates the comfortable acceleration after SVO angle mapping, represents the expected time after SVO angle mapping, Indicates the minimum safe distance after SVO angle mapping; Among them, comfortable acceleration, expected time interval and minimum safety distance are the core parameters in IDM; Expected acceleration calculation formula:

[0025]

[0026] Where, represents the desired acceleration of the vehicle, represents the maximum acceleration, Indicates the current speed. Indicates the desired safety distance, Indicates the actual vehicle distance, represents the updated expected time interval in the IDM parameters, Indicates the safety time interval, Indicates the speed difference between the current vehicle and the preceding vehicle, represents the maximum acceleration of the vehicle, Indicates the maximum deceleration, denotes an absolute value symbol, denotes a maximum speed of the vehicle; wherein the corresponding formula in the smoothing process is specifically denoted as:

[0027] denotes a smoothed acceleration, denotes an expected acceleration of the vehicle, denotes a preset smoothing coefficient, denotes a control period, denotes a current vehicle acceleration.

[0028] The second aspect of the present application is to provide a social interaction control system for intelligent networked vehicles in a mixed traffic scenario, comprising: an HDV feature recognition module: used to obtain vehicle distance data, speed data and acceleration data in the driving process; all types of data are fused based on the vehicle distance data to obtain a spatial feature vector; four types of time sequence features of the vehicle speed trend, acceleration mode, lane changing intention and headway change are obtained, and a time sequence feature vector is obtained through the time sequence features; a comprehensive feature vector is obtained through the spatial feature vector and the time sequence feature vector; a reward function is obtained through linear combination according to the comprehensive feature vector; the best weight vector is obtained through inverse reinforcement learning according to the reward function; the SVO angle value of the HDV is estimated through the best weight vector, and the SVO angle estimation value of the CAV is obtained; an SVO optimization decision module: used to optimize and dynamically adjust the SVO parameters through Stackelberg game based on the SVO angle estimation value of the CAV, and obtain the optimal response parameters of the HDV; a socially aware HDV semi-dynamic response module: used to simulate the social perception and response of human drivers according to the IDM model and the optimal response parameters of the HDV, so as to adjust the driving behavior.

[0029] The third aspect of the present application is to provide an electronic device, comprising a memory, a processor and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to realize the social interaction control method for intelligent networked vehicles in a mixed traffic scenario.

[0030] The fourth aspect of the present application is to provide a computer readable storage medium, which stores a computer program, wherein the computer program is executed by a processor to realize the social interaction control method for intelligent networked vehicles in a mixed traffic scenario.

[0031] ​Compared with the prior art, the beneficial effects of the present invention are: obtaining spatial feature vectors and temporal feature vectors; obtaining a comprehensive feature vector through the spatial feature vectors and the temporal feature vectors; obtaining a reward function based on linear combination of the comprehensive feature vectors; obtaining an optimal weight vector through inverse reinforcement learning based on the reward function; estimating the SVO angle value of the HDV through the optimal weight vector to obtain the SVO angle estimation value of the CAV; solving the unidirectionality problem of the interactive control strategy; dynamically adjusting the SVO parameters through Stackelberg game optimization through the SVO angle estimation value of the CAV to obtain the optimal response parameters of the HDV; improving the accuracy of the analysis of the driver's personalized driving style; simulating the social perception and response of the human driver based on the IDM model and the optimal response parameters of the HDV to adjust the driving behavior, solving the problem of coexistence of existing intelligent connected vehicles and manually driven vehicles, and improving the accuracy of vehicle collaborative interactive control. BRIEF DESCRIPTION OF THE DRAWINGS

[0032] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0033] Figure 1 The present invention provides a flowchart of the steps of a social collaborative interaction control method for intelligent connected vehicles in a mixed traffic scenario; Figure 2 The present invention provides a module flow diagram of a social collaborative interactive control system for intelligent connected vehicles in a mixed traffic scenario. DETAILED DESCRIPTION

[0034] In order to enable those skilled in the art to better understand the solutions of the present invention, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the embodiments described are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts should fall within the scope of protection of the present invention.

[0035] It should be noted that the terms "first", "second", etc. in the description and claims of the present invention and the above-mentioned drawings are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that the numbers used in this way can be interchanged where appropriate, so that the embodiments of the present invention described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions. For example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0036] In response to the problems existing in the background technology, the research and design of a social collaborative interaction control method and system for intelligent connected vehicles in mixed traffic scenarios have important practical significance.

[0037] like Figure 1 As shown, the first aspect of the present invention is to provide a social collaborative interaction control method for intelligent connected vehicles in a mixed traffic scenario, comprising the following steps: Step S001: Obtain a spatial feature vector and a temporal feature vector; obtain a comprehensive feature vector through the spatial feature vector and the temporal feature vector; obtain a reward function by performing linear combination based on the comprehensive feature vector; obtain an optimal weight vector through inverse reinforcement learning based on the reward function; estimate the SVO angle value of the HDV through the optimal weight vector to obtain the SVO angle estimation value of the CAV.

[0038] To address these technical challenges, a two-layer closed-loop interactive control decision-making framework based on SVO features was proposed. This framework comprises three core submodules: an SVO feature recognition module based on online inverse reinforcement learning, a dynamic decision-making module based on Stackelberg game theory, and an HDV response module. By online identifying HDV SVO features, the CAV's decision-making behavior is optimized and real-time adjustments are made based on the HDV's response characteristics, achieving safe and efficient mixed traffic control.

[0039] The SVO feature recognition module based on online inverse reinforcement learning adopts a multi-layered architecture design, including a data acquisition layer, a feature processing layer, an online learning layer, and a prediction and evaluation layer. In terms of functional design, the module includes data preprocessing, feature extraction, feature fusion, online learning, and prediction and evaluation.

[0040] Specifically, the vehicle distance data, speed data and acceleration data are collected during the driving process; the data standardization operation is completed by linear normalization of each category of data, and each standardized data is combined into a standardized feature vector; that is, the vehicle distance standardized feature vector, the speed standardized feature vector and the acceleration standardized feature vector are obtained.

[0041] Constructing a feature space by using the vehicle distance normalized feature vector, the speed normalized feature vector and the acceleration normalized feature vector; In the distance-based attention mechanism, the similarity between each data in each normalized feature vector and each data in the vehicle distance normalized feature vector is calculated; the similarity is specifically expressed as:

[0042] Where, represents the first vehicle distance normalized feature vector The data points corresponding to each data point in the feature space are: Indicates the The first of the normalized eigenvectors The data points corresponding to each data point in the feature space are: represents the Euclidean distance between two points, Indicates the similarity between two data.

[0043] According to the similarity between each data in each standardized feature vector and all data in the vehicle distance standardized feature vector, The function calculates the weight of each data in each standardized feature vector relative to each data in the vehicle distance standardized feature vector; Functions are well-known technologies and will not be described in detail here.

[0044] The spatial feature vector is obtained based on the weight of each data in each standardized feature vector relative to each data in the vehicle distance standardized feature vector and all standardized feature vectors. The process of obtaining each data in the spatial feature vector is specifically expressed by the formula:

[0045] Where, represents the first vehicle distance normalized feature vector data, represents the first vehicle distance normalized feature vector The data is relative to the first in the vehicle distance normalized feature vector The weight of the data; represents the first normalized eigenvector of acceleration data, represents the first normalized eigenvector of acceleration The data is relative to the first in the vehicle distance normalized feature vector The weight of the data, represents the first in the velocity normalized eigenvector data, represents the first in the velocity normalized eigenvector The data is relative to the first in the vehicle distance normalized feature vector The weight of the data; represents the number of data in the normalized feature vector of vehicle distance, represents the number of data in the acceleration normalized feature vector, represents the number of data in the velocity normalized feature vector, Represents the first feature vector in the space data.

[0046] So far, the spatial feature vector is obtained through the above method.

[0047] Four types of time series features are extracted: vehicle speed trend, acceleration pattern, lane change intention, and headway change. A preset length is selected to determine the time series observation window. The speed trend is obtained by linear regression of the speed sequence within the time series observation window. The specific formula is:

[0048] Where, Indicates the first A moment of time, Represents the mean value of the time of all moments in the time series observation window, Indicates the first The speed of a moment, represents the mean value of the speed at all moments in the time series observation window, Represents the number of all moments in the time series observation window, Indicates speed trend.

[0049] Among them, the acceleration pattern is calculated by dividing the speed difference between the two end points in the time series observation window by the time difference; the lane change intention is determined by the vehicle trajectory and steering angle changes; and the headway change is extracted by calculating the time interval between the front and rear vehicles and its changes.

[0050] The time series feature vector is obtained through the vehicle's speed trend, acceleration pattern, lane change intention and headway change; the time series feature vector is specifically expressed as:

[0051] Where, Indicates the speed trend, Indicates the acceleration mode, Indicates intention to change lanes. Indicates the change in headway distance, represents the time series feature vector.

[0052] The spatial feature vector and the temporal feature vector are fused to obtain a comprehensive feature vector. The comprehensive feature vector is specifically expressed as:

[0053] Where, represents the spatial eigenvector, represents the time series feature vector, represents the comprehensive feature vector, represents the weight of the spatial eigenvector; where, Acquisition was determined through cross-validation.

[0054] At this point, the comprehensive feature vector is obtained through the above method.

[0055] An online inverse reinforcement learning algorithm based on maximum entropy analyzes HDV driving trajectory data in real time to learn the inherent reward function weights and infer SVO characteristics. The algorithm takes the HDV's state-action sequence as input and outputs a feature weight vector that represents the driver's behavioral preferences. This paper employs an incremental learning strategy that continuously integrates new observations through a sliding time window mechanism, enabling the model to capture dynamic changes in driver behavior characteristics.

[0056] Determine the state space and action space, generate trajectories through the state space and action space, and each trajectory .in, Indicates the first state of each trajectory and the first action performed in the first state, Indicates the first state and The first state executed Action, is a trajectory consisting of states and actions.

[0057] Among them, the reward function in inverse reinforcement learning is defined as the comprehensive feature vector A linear combination of

[0058] Where, represents the comprehensive feature vector, represents the reward function value, represents the weight vector to be learned, Represents the transposition of the weight vector to be learned.

[0059] The specific learning process of the weight vector to be learned in the above linear combination is: Based on the maximum entropy principle, the probability distribution of each trajectory can be expressed as:

[0060] Where, Indicates the first The comprehensive feature vector of the corresponding state and action at step time; represents the weight vector to be learned, Indicates the transposition of the weight vector to be learned, Indicates the total step length corresponding to each trajectory, represents an exponential function with a natural constant as the base, represents the partition function, Represents each trajectory The probability value of Indicates the first The reward value of the step.

[0061] Among them, the partition function It is defined as the sum of the exponential rewards of all trajectories; specifically expressed as:

[0062] Where, Indicates the Track No. The comprehensive feature vector of the corresponding state and action at step time, Indicates generating the first Tracks probability; represents the number of all trajectories, Indicates the The first track The reward value of the step.

[0063] The relative entropy projection method is used to perform convergence analysis on the weight vector to be learned and determine the final weight vector; specifically, it is expressed as:

[0064] Where, Represents each trajectory The probability value of Represents the empirical distribution probability value of the observed trajectory; express Relative to The relative entropy can also be understood as Relative to The difference, represents the weight vector in the iterative process, represents the optimal weight vector; It means finding the weight vector corresponding to the minimum relative entropy as the optimal weight vector.

[0065] The SVO angle value of the HDV is estimated by using the optimal weight vector to obtain an estimated SVO angle value. The process of estimating the SVO angle is a well-known technology and will not be described in detail here.

[0066] Step S002: dynamically adjusting the SVO parameters through Stackelberg game optimization based on the SVO angle estimation value of the CAV to obtain the optimal response parameters of the HDV.

[0067] It should be noted that CAVs optimize their SVO orientation through a Stackelberg game. Taking into account safety utility, efficiency utility, and interaction utility, they find the optimal SVO value that both protects their own interests and promotes good interaction. The Stackelberg game optimization process is well known and will not be detailed here.

[0068] Specifically, CAV selects the optimal SVO as the leader, and HDV responds as the follower. In this game framework, the interactive decision-making process can be described as follows: CAV first selects the optimal SVO based on the current state. Select the optimal SVO angle , HDV observes the optimal SVO angle selection of CAV and makes the optimal response parameters The process can be formalized as:

[0069] Where, Indicates the The system status at a moment, Indicates that CAV SVO angle estimation value at the moment; Indicates that HDV The acceleration corresponding to the optimal response after observing the choice of CAV at each moment, represents the utility function of CAV; Indicates the SVO angle estimation value corresponding to the maximum utility function, Indicates that CAV The optimal SVO angle after the game at each moment.

[0070] In this embodiment, the dynamic Stackelberg game framework is represented as a five-tuple .in, , which represents the CAV and human-driven vehicles participating in the game, is the system state space, is the action space, is the state transition function, is the utility function.

[0071] At the moment , system status , which can be expressed as:

[0072] in, and Respectively represent CAV in the The position and velocity vectors at each moment, Indicates that CAV The acceleration at a moment, Indicates that CAV The heading angle at a moment; and Respectively represent HDV in The position and velocity vectors at each moment, Indicates that HDV The acceleration at a moment, Indicates that HDV The heading angle at a moment.

[0073] Action Space A set of selectable vehicle actions is defined, where the action of the CAV is the selection of the SVO angle and the action of the HDV is the acceleration response.

[0074] Among them, the utility function It is built by taking security, efficiency and interactivity into consideration.

[0075] Generate control instructions based on the optimal SVO angle, execute the control instructions and update the status; calculate the collision risk based on the updated status, determine whether to trigger emergency braking based on the collision risk, and update the status again.

[0076] Step S003: According to the IDM model and the optimal response parameters of the HDV, the social perception and response of the human driver are simulated to adjust the driving behavior.

[0077] It should be noted that for HDVs, the semi-dynamic SVO-driven IDM car-following model provides a more accurate description of their behavior simulation. This model introduces SVO into the traditional IDM framework, reflecting the driver's inherent style through a base SVO value while allowing SVO to dynamically adjust within a limited range to adapt to environmental changes. For example, in high-density or emergency interaction scenarios, the model will appropriately increase the SVO value to enhance cooperative behavior; while in low-density scenarios, the SVO value may be reduced to pursue higher traffic efficiency. This design not only preserves the continuity and consistency of driving style but also enhances the model's adaptability to complex traffic environments. Furthermore, the SVO value is mapped to IDM parameters via trigonometric functions, influencing the weight distribution of the free acceleration term and the interactive deceleration term, thereby regulating the vehicle's longitudinal following behavior.

[0078] It's also worth noting that while traditional IDM can describe basic car-following behavior, it struggles to capture the rich individual characteristics and social preferences of human drivers. To overcome this limitation, this embodiment proposes an improved IDM model that incorporates SVO. This model incorporates SVO parameters to quantify the driver's trade-off between their own interests and the interests of others during decision-making, enabling an accurate description of human driving behavior.

[0079] Specifically, the response process of the social perception HDV semi-dynamic response module algorithm is as follows: The optimal SVO angle obtained above is used as the target SVO angle for the current scenario in the SVO dynamic adjustment formula. The current SVO angle is then updated using the SVO dynamic adjustment formula to obtain an updated current SVO angle. The IDM parameters are updated using the SVO parameter mapping based on the updated current SVO angle to obtain updated IDM parameters. The expected acceleration calculation formula is then applied to the updated IDM parameters to obtain the expected acceleration. The final output acceleration is obtained through smoothing. The system also uses a lane change utility function to make lane change decisions. The lane change utility is calculated and compared with a threshold to determine whether to execute the lane change. Finally, the system performs a safety check and triggers the emergency braking mechanism if safety conditions are not met. The output acceleration is then fed back to the CAV decision-making process. The lane change utility function is determined based on safety and necessity.

[0080] The parameter mapping relationship based on SVO is defined as follows:

[0081]

[0082]

[0083] Where, represents the baseline comfort acceleration, represents the benchmark expected time interval, Indicates the minimum safe distance between vehicles. represents the sine function, represents the SVO angle, Indicates the comfortable acceleration after SVO angle mapping, represents the expected time after SVO angle mapping, Represents the minimum safe distance after SVO angle mapping. The coefficients 0.5, 0.3, and 0.2 selected in the formula are empirical results obtained through parameter sensitivity analysis and driving comfort experiments.

[0084] Among them, comfortable acceleration, expected time interval and minimum safety distance are the core parameters in IDM.

[0085] To achieve driving style stability, a base SVO angle is set for each HDV to represent its inherent driving characteristics. Based on this inherent driving characteristic, an adjustment range is given to adapt to different scenarios. The adjustment range of the SVO angle is specifically expressed as:

[0086] Where, represents the basic SVO angle, Indicates the maximum allowable deviation of the SVO angle; wherein in this embodiment, the maximum allowable deviation of the SVO angle for , where the maximum allowable deviation of the SVO angle in this embodiment is . It should be noted that the difference in the SVO angle between adjacent steps cannot be greater than .

[0087] During dynamic adjustment, the rate of change of SVO is related to the urgency of the current scenario. Related; Among them, the SVO dynamic adjustment formula is specifically expressed as:

[0088] Where, Indicates the current time The urgency of the situation, Indicates the target SVO angle value calculated based on the current scene, represents the adjustment gain coefficient, represents the rate of change of SVO, Indicates the current time SVO angle.

[0089] The urgency of the scene is obtained by weighted averaging of traffic density, safety risk, and time pressure. Traffic density, safety risk, and time pressure in the scene are all quantified in the known art and will not be described in detail here.

[0090] The expected acceleration calculation formula is specifically expressed as:

[0091]

[0092] Where, represents the desired acceleration of the vehicle, represents the maximum acceleration, Indicates the current speed. Indicates the desired safety distance, Indicates the actual vehicle distance, represents the updated expected time interval in the IDM parameters, represents the safety time interval (i.e., the time interval between the current vehicle and the preceding vehicle, i.e., the expected time interval after updating in the IDM parameter). Indicates the speed difference between the current vehicle and the preceding vehicle, represents the maximum acceleration of the vehicle, Indicates the maximum deceleration, Indicates the absolute value symbol, Indicates the maximum speed of the vehicle.

[0093] Among them, the corresponding formula in the smoothing process is specifically expressed as:

[0094] Where, represents the smoothed acceleration, represents the desired acceleration of the vehicle, Indicates the preset smoothing coefficient, represents the control period, Indicates the current vehicle acceleration.

[0095] Step S004: Performance evaluation and verification.

[0096] It should be noted that this embodiment conducts a comprehensive simulation verification of the proposed two-layer closed-loop interactive control framework, and therefore the following indicators are designed to evaluate and verify the interactive performance.

[0097] Specifically, interaction quality assessment, prediction assessment, and system stability assessment are conducted; Among them, the interaction quality evaluation indicators are: In order to evaluate the interaction quality between CAV and HDV, an interaction matching index is designed:

[0098] Where, Indicates the The speed of the CAV at a given moment, Indicates the The speed of HDV at this moment, Indicates the maximum speed of the vehicle, Indicates the absolute value symbol, represents an exponential function with a natural constant as the base, Indicates the duration of the interaction, Indicates the interaction quality between CAV and HDV. A higher interaction quality indicates better interaction synchronization.

[0099] In order to quantitatively evaluate the interaction effect, the speed matching index is defined:

[0100] Where, Indicates the speed matching between CAV and HDV.

[0101] Among them, the prediction evaluation indicators are:

[0102] Where, Indicates the The predicted SVO angle of samples, Indicates the The true SVO angle of the samples, represents the number of test samples, represents the permissible error threshold (in this embodiment, ); Represents an indicator function, the value is 1 when the condition is met, otherwise it is 0; Denotes the predicted difference factor.

[0103] In order to quantitatively evaluate the prediction performance, the comprehensive prediction error index is defined:

[0104] Where, represents the predicted SVO angle, represents the true SVO angle, Indicates the The predicted position at the moment, Indicates the The actual position at a moment, Indicates the maximum vehicle distance. represents the Euclidean distance between two locations, represents the difference between two angles, represents the maximum SVO angle, represents the comprehensive prediction error, represents the angle weight coefficient, Represents the distance weight coefficient; wherein the angle weight coefficient and the distance weight coefficient are both preset values. In this embodiment , In this embodiment, and There is no specific limitation and implementers can decide based on specific circumstances.

[0105] Among them, system stability assessment includes: In order to evaluate the stability of the system under different working conditions, the traffic flow fluctuation index is defined:

[0106] Where, represents the standard deviation of all sample velocities, Indicates the degree of traffic flow fluctuation. This metric reflects the speed distribution characteristics of traffic flow, with smaller values ​​indicating more stable traffic flow. This metric is particularly important for evaluating the overall performance of the two-layer closed-loop interactive control framework.

[0107] To quantitatively evaluate car-following performance, a comprehensive car-following index is defined:

[0108] Where, Indicates the actual vehicle distance. Indicates the expected vehicle distance, Indicates the speed of the following vehicle, Indicates the speed of the vehicle ahead. It represents the comprehensive following index. represents the vehicle distance weight coefficient, Represents the speed weight coefficient; among them, the vehicle distance weight coefficient and speed weight coefficient are both preset values.

[0109] The above-mentioned metrics were used to analyze the effectiveness of the dynamic SVO strategy. Compared to the fixed SVO baseline, the dynamic SVO strategy achieved an average overall performance improvement of approximately 28.73%, with the most significant improvement in safety performance reaching up to 60%. In particular, the HDV feature recognition module based on online inverse reinforcement learning performed exceptionally well, achieving a mean absolute error of only 1.56°, a 60.3% improvement over the traditional IRL method, providing a reliable foundation for dynamic CAV decision-making. The SVO optimization mechanism based on the Stackelberg game demonstrated strong adaptability in diverse mixed traffic scenarios.

[0110] Secondly, the socially aware HDV semi-dynamic response module significantly improves the accuracy of modeling human driving behavior. The improved IDM model outperforms the traditional IDM in two key metrics: distance maintenance and speed coordination. Distance fluctuations are reduced by approximately 57% and speed following errors are reduced by approximately 32%. This improvement not only enhances the stability of mixed traffic flows but also strengthens the system's adaptability to complex mixed traffic scenarios. In particular, the system maintains an interaction matching degree exceeding 80%, demonstrating excellent collaborative performance.

[0111] At this point, this embodiment is completed.

[0112] like Figure 2 As shown, the second aspect of the present invention is to provide a social collaborative interaction control system for intelligent connected vehicles in mixed traffic scenarios, including: HDV feature recognition module 101: used to obtain vehicle distance data, speed data, and acceleration data during vehicle travel; fuse all types of data based on the vehicle distance data to obtain a spatial feature vector; obtain four types of time series features of the vehicle, namely, speed trend, acceleration pattern, lane change intention, and headway change, and obtain a time series feature vector based on the time series features; obtain a comprehensive feature vector based on the spatial feature vector and the time series feature vector; obtain a reward function by linearly combining the comprehensive feature vector; obtain an optimal weight vector based on the reward function through inverse reinforcement learning; estimate the SVO angle value of the HDV using the optimal weight vector to obtain an estimated SVO angle value of the CAV; SVO optimization decision module 102: for dynamically adjusting SVO parameters through Stackelberg game optimization based on the SVO angle estimation value of the CAV to obtain the optimal response parameters of the HDV; The social perception HDV semi-dynamic response module 103 is used to simulate the social perception and response of the human driver according to the IDM model and the optimal response parameters of the HDV, so as to adjust the driving behavior.

[0113] The third aspect of the present invention is to provide an electronic device, including a memory, a processor, and a computer program stored in the memory and runnable on the processor. When the processor executes the computer program, it implements a social collaborative interaction control method for intelligent connected vehicles in a mixed traffic scenario.

[0114] The fourth aspect of the present invention is to provide a computer-readable storage medium, which stores a computer program. When the computer program is executed by a processor, it implements a social collaborative interaction control method for intelligent connected vehicles in a mixed traffic scenario.

[0115] Those skilled in the art will appreciate that embodiments of the present invention may be provided as methods, systems, or computer program products. Thus, the present invention may take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to magnetic disk storage, optical storage, etc.) containing computer-usable program code.

[0116] The present invention is described with reference to flowcharts and / or block diagrams of methods, systems, and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowcharts and / or block diagrams, as well as combinations of processes and / or blocks in the flowcharts and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowcharts and / or block diagrams. Figure 1 a process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0117] These computer program instructions may also be stored in a computer readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 a process or multiple processes and / or boxes Figure 1 The function specified in one or more boxes.

[0118] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operational steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing the instructions executed on the computer or other programmable device for implementing the process. Figure 1 a process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.

[0119] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to the above embodiments, ordinary technicians in the field should understand that the specific implementation methods of the present invention can still be modified or replaced by equivalents. Any modification or equivalent replacement that does not depart from the spirit and scope of the present invention should be covered by the scope of protection of the present invention.

Claims

1. A social collaborative interactive control method for intelligent connected vehicles in mixed traffic scenarios, characterized by: include: Obtain vehicle distance data, speed data, and acceleration data during driving; All types of data are fused based on vehicle distance data to obtain spatial feature vectors; Four types of time series features are obtained, including the vehicle's speed trend, acceleration pattern, lane change intention, and headway change. A time series feature vector is obtained from the time series features. A comprehensive feature vector is obtained from the spatial feature vector and the time series feature vector. A reward function is obtained by linearly combining the comprehensive feature vector. An optimal weight vector is obtained based on the reward function through inverse reinforcement learning. The SVO angle value of the HDV is estimated using the optimal weight vector to obtain an estimated SVO angle value of the CAV. The SVO angle estimation value of CAV is used to dynamically adjust the SVO parameters through Stackelberg game optimization to obtain the optimal response parameters of HDV; According to the IDM model and the optimal response parameters of HDV, the social perception and response of human drivers are simulated to adjust driving behavior.

2. The social collaborative interaction control method for intelligent connected vehicles in mixed traffic scenarios according to claim 1 is characterized in that: The method of fusing all types of data based on vehicle distance data to obtain a spatial feature vector includes: The data standardization operation is completed by performing linear normalization on the data of each category, and each standardized data is combined into a standardized feature vector; that is, the vehicle distance standardized feature vector, the speed standardized feature vector and the acceleration standardized feature vector are obtained; Constructing a feature space by using the vehicle distance normalized feature vector, the speed normalized feature vector and the acceleration normalized feature vector; In the distance-based attention mechanism, the similarity between each data in each normalized feature vector and each data in the vehicle distance normalized feature vector is calculated. The similarity is specifically expressed as: Where, represents the first vehicle distance normalized feature vector The data points corresponding to each data point in the feature space are: Indicates the The first of the normalized eigenvectors The data points corresponding to each data point in the feature space are: represents the Euclidean distance between two points, Indicates the similarity between two data; According to the similarity between each data in each standardized feature vector and all data in the vehicle distance standardized feature vector, The function calculates the weight of each data in each standardized feature vector relative to each data in the vehicle distance standardized feature vector; The process of obtaining each data in the spatial feature vector is specifically expressed by the formula: Where, represents the first vehicle distance normalized feature vector data, represents the first vehicle distance normalized feature vector The data is relative to the first in the vehicle distance normalized feature vector The weight of the data; represents the first normalized eigenvector of acceleration data, represents the first normalized eigenvector of acceleration The data is relative to the first in the vehicle distance normalized feature vector The weight of the data, represents the first in the velocity normalized eigenvector data, represents the first in the velocity normalized eigenvector The data is relative to the first in the vehicle distance normalized feature vector The weight of the data; represents the number of data in the normalized feature vector of vehicle distance, represents the number of data in the acceleration normalized feature vector, represents the number of data in the velocity normalized feature vector, Represents the first feature vector in the space data.

3. The social collaborative interaction control method for intelligent connected vehicles in mixed traffic scenarios according to claim 1 is characterized in that: The four types of time series features of the vehicle's speed trend, acceleration pattern, lane change intention, and headway change are obtained, and a time series feature vector is obtained through the time series features, including: A preset length is selected and the time series observation window is determined by the preset length. The speed trend is obtained by linear regression of the speed sequence within the time series observation window. The specific formula is: Where, Indicates the first A moment of time, Represents the mean value of the time of all moments in the time series observation window, Indicates the first The speed of a moment, represents the mean value of the speed at all moments in the time series observation window, Represents the number of all moments in the time series observation window, Indicates speed trend; The acceleration pattern is calculated by dividing the speed difference between the two endpoints within the time series observation window by the time difference. Lane change intention is determined by vehicle trajectory and steering angle changes. Headway changes are extracted by calculating the time interval between the front and rear vehicles and their changes. The time series feature vector is specifically expressed as: Where, Indicates the speed trend, Indicates the acceleration mode, Indicates intention to change lanes. Indicates the change in headway distance, represents the time series feature vector.

4. The social collaborative interaction control method for intelligent connected vehicles in mixed traffic scenarios according to claim 1 is characterized in that: The method of obtaining a comprehensive feature vector through the spatial feature vector and the temporal feature vector includes: Where, represents the spatial eigenvector, represents the time series feature vector, represents the comprehensive feature vector, represents the weight of the spatial eigenvector; where, Acquisition was determined through cross-validation.

5. The social collaborative interaction control method for intelligent connected vehicles in mixed traffic scenarios according to claim 1 is characterized in that: The method of performing linear combination according to the comprehensive feature vector to obtain a reward function; obtaining an optimal weight vector according to the reward function through inverse reinforcement learning; estimating the SVO angle value of the HDV by using the optimal weight vector to obtain the SVO angle estimation value of the CAV includes: Determine the state space and action space, generate trajectories through the state space and action space, and each trajectory ; in, Indicates the first state of each trajectory and the first action performed in the first state, Indicates the first state and The first state executed Action, is a trajectory consisting of states and actions; In inverse reinforcement learning, the reward function is defined as the comprehensive feature vector The linear combination of ; specifically expressed as: Where, represents the comprehensive feature vector, represents the reward function value, represents the weight vector to be learned, Indicates the transposition of the weight vector to be learned; Among them, the specific learning process of the weight vector to be learned in the linear combination is: The probability distribution of each trajectory can be expressed as: Where, Indicates the first The comprehensive feature vector of the corresponding state and action at step time; represents the weight vector to be learned, Indicates the transposition of the weight vector to be learned, Indicates the total step length corresponding to each trajectory, represents an exponential function with a natural constant as the base, represents the partition function, Represents each trajectory The probability value of Indicates the first The reward value of the step; The partition function is specifically expressed as: Where, Indicates the Track No. The comprehensive feature vector of the corresponding state and action at step time, Indicates generating the first Tracks probability; represents the number of all trajectories, Indicates the The first track The reward value of the step; The relative entropy projection method is used to perform convergence analysis on the weight vector to be learned and determine the final weight vector; specifically, it is expressed as: Where, Represents each trajectory The probability value of Represents the empirical distribution probability value of the observed trajectory; express Relative to The relative entropy of represents the weight vector in the iterative process, represents the optimal weight vector; Indicates that the weight vector corresponding to the minimum relative entropy is found as the optimal weight vector; The SVO angle value of the HDV is estimated by using the optimal weight vector to obtain an SVO angle estimation value.

6. The social collaborative interaction control method for intelligent connected vehicles in mixed traffic scenarios according to claim 1 is characterized in that: The method of dynamically adjusting SVO parameters by using the estimated SVO angle of the CAV through Stackelberg game optimization to obtain the optimal response parameters of the HDV includes: Where, Indicates the The system status at a moment, Indicates that CAV SVO angle estimation value at the moment; Indicates that HDV The acceleration corresponding to the optimal response after observing the choice of CAV at each moment, represents the utility function of CAV; Indicates the SVO angle estimation value corresponding to the maximum utility function, Indicates that CAV The optimal SVO angle after the game at each moment; Among them, the utility function Build access through the unity of security, efficiency, and interactivity.

7. The social collaborative interaction control method for intelligent connected vehicles in mixed traffic scenarios according to claim 6 is characterized in that: The method simulates the social perception and response of human drivers based on the IDM model and the optimal response parameters of the HDV to adjust driving behavior, including: The optimal SVO angle is used as the target SVO angle value calculated for the current scenario in the SVO dynamic adjustment formula. The current SVO angle value is then updated using the SVO dynamic adjustment formula to obtain an updated current SVO angle value. The IDM parameters are updated using the SVO parameter mapping relationship based on the updated current SVO angle value to obtain updated IDM parameters. The expected acceleration calculation formula is used to calculate the updated IDM parameters to obtain the expected acceleration. The final output acceleration is obtained through smoothing. At the same time, the system makes lane change decisions using a lane change utility function, calculates the lane change utility, and compares it with a threshold to determine whether to execute the lane change operation. Finally, the system performs a safety check and triggers the emergency braking mechanism when the safety conditions are not met. Finally, the output acceleration is fed back to the CAV decision-making process. Among them, the lane-changing utility function is determined by safety and necessity; Among them, the SVO dynamic adjustment formula is specifically expressed as: Where, Indicates the current time The urgency of the situation, Indicates the target SVO angle value calculated based on the current scene, represents the adjustment gain coefficient, represents the rate of change of SVO, Indicates the current time SVO angle; The urgency of the scene is obtained by weighted average of traffic density, safety risk and time pressure; The parameter mapping relationship of SVO is specifically expressed as follows: Where, represents the baseline comfort acceleration, represents the benchmark expected time interval, Indicates the minimum safe distance between vehicles. represents the sine function, represents the SVO angle, Indicates the comfortable acceleration after SVO angle mapping, represents the expected time interval after SVO angle mapping, Indicates the minimum safe distance after SVO angle mapping; Among them, comfortable acceleration, expected time interval and minimum safety distance are the core parameters in IDM; Expected acceleration calculation formula: Where, represents the desired acceleration of the vehicle, represents the maximum acceleration, Indicates the current speed. Indicates the desired safety distance, Indicates the actual vehicle distance, Indicates the updated expected time interval in the IDM parameters, Indicates the safety time interval, Indicates the speed difference between the current vehicle and the preceding vehicle, represents the maximum acceleration of the vehicle, Indicates the maximum deceleration, Indicates the absolute value symbol, Indicates the maximum speed of the vehicle; Among them, the corresponding formula in the smoothing process is specifically expressed as: Where, represents the smoothed acceleration, represents the desired acceleration of the vehicle, Indicates the preset smoothing coefficient, represents the control period, Indicates the current vehicle acceleration.

8. A social collaborative interactive control system for intelligent connected vehicles in mixed traffic scenarios, characterized by: include: HDV feature recognition module: used to obtain vehicle distance data, speed data and acceleration data during driving; All types of data are fused based on vehicle distance data to obtain spatial feature vectors; Four types of time series features are obtained, including the vehicle's speed trend, acceleration pattern, lane change intention, and headway change. A time series feature vector is obtained from the time series features. A comprehensive feature vector is obtained from the spatial feature vector and the time series feature vector. A reward function is obtained by linearly combining the comprehensive feature vector. An optimal weight vector is obtained based on the reward function through inverse reinforcement learning. The SVO angle value of the HDV is estimated using the optimal weight vector to obtain an estimated SVO angle value of the CAV. SVO optimization decision module: used to dynamically adjust SVO parameters through Stackelberg game optimization based on the SVO angle estimation value of CAV to obtain the optimal response parameters of HDV; Socially Aware HDV Semi-Dynamic Response Module: This module simulates the social perception and response of human drivers based on the IDM model and HDV's optimal response parameters, thereby adjusting driving behavior.

9. An electronic device, characterized in that: It includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the social collaborative interaction control method for intelligent connected vehicles in a mixed traffic scenario as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, it implements the social collaborative interaction control method of an intelligent connected vehicle in a mixed traffic scenario according to any one of claims 1 to 7.

Images