Network twinning-based vehicle-road collaborative awareness sharing method and system

Abstract

The invention discloses a network twinning-based vehicle-road collaborative perception sharing method and system, and the method comprises the steps: proposing a road user intention correlation criterion in a dynamic scene, quantifying the traffic safety information value between road users according to the criterion, and enabling an infrastructure to selectively share the high-value road information with a vehicle; different sensing requirements of vehicles and real-time channel busy rates of infrastructures in a vehicle network are considered, a sensing information adaptive transmission frequency adjustment and dynamic bandwidth allocation algorithm under a multi-vehicle request scene is designed, and channel congestion and communication resource waste are avoided. According to the method, the optimization method is integrated, the time-varying characteristic of the infrastructure auxiliary vehicle system is considered, high-value safety information is shared and the transmission frequency of the infrastructure is dynamically adjusted according to the characteristics of low time delay, high flexibility and high reliability of network twinning, so that the collaborative awareness service quality is improved, the overall task completion time delay is reduced, and the system performance is improved. And the system information redundancy is reduced.

Description

Technical Field

[0001] The present invention relates to the field of edge computing, and in particular to a method and system for sharing vehicle perception based on edge cloud collaboration. Background Art

[0002] As the level of vehicle intelligence continues to improve, more and more vehicles have assisted driving functions. Vehicle assisted driving can reduce the driver's driving pressure, relieve driving fatigue, and improve travel efficiency. However, in recent years, the reliability and safety of assisted driving systems have been questioned. According to the latest data from the National Highway Traffic Safety Administration of the United States: As of August 2024, there have been as many as 386 vehicle accidents involving advanced driver assistance systems this year. Since single-vehicle intelligence does not have beyond-visual-range perception capabilities, the reaction time will be too long due to obstacles, bad weather, etc., affecting driving safety.

[0003] For this reason, vehicle-road collaboration integrates beyond-line-of-sight perception information and assists vehicle driving through low-latency vehicle-road data interaction to improve road safety and operating efficiency, which has become a research hotspot in the current intelligent vehicle field. Compared with the existing cooperative perception services based on adjacent vehicles, vehicle-road collaboration can provide beyond-line-of-sight perception information for intelligent vehicles and even non-intelligent vehicles, with high availability. However, since vehicles obtain a larger perception range through information shared by roadside infrastructure, their reliability will be affected by the quality of information shared by the infrastructure. For example, frequent data interaction can easily lead to channel congestion and reduce the timeliness of perception information; sharing too much data will lead to information redundancy and waste computing and communication resources, while sharing too little data will result in incomplete messages obtained by vehicles, affecting driving safety. Summary of the invention

[0004] This application proposes a vehicle-road collaborative perception sharing method and system based on network twins. First, the network twin receives global road information collected by roadside infrastructure, maps these data to a coupled road traffic and wireless network simulator after state correction, constructs a parallel virtualization of the vehicle-to-roadside infrastructure network, and forms a twin with autonomous identity and information interaction functions. Then, in order to maximize the utility of the twin, a road individual intention correlation criterion is formulated, and a high-value content selection learning algorithm based on deep deterministic policy gradient is proposed. Speed ​​entropy technology is introduced in the design of the reward function to encourage the system to maintain a certain degree of exploration to adapt to dynamic situations. Finally, considering the diversity of vehicle perception needs and changes in real-time channel busy rates, adaptive transmission frequency adjustment and dynamic bandwidth allocation algorithms are designed to maximize the timeliness and accuracy of information while avoiding channel congestion and resource waste.

[0005] To achieve the above objectives, this application provides the following solutions:

[0006] A vehicle-road cooperative perception sharing method based on network twins includes the following steps:

[0007] By receiving the global road information collected by the roadside infrastructure, a parallel virtual twin of the vehicle-to-roadside infrastructure network is constructed;

[0008] Obtain more accurate dynamic system states from a series of incomplete and noisy measurements by combining sensor inputs and model predictions;

[0009] Comprehensively consider factors such as the movement intention, speed and direction, and physical proximity of road participants to select users who pose a collision threat to the target vehicle;

[0010] According to the different perception requirements of target vehicles and the real-time channel busy rate, an algorithm for adaptive adjustment of the transmission frequency of collaborative perception information and dynamic bandwidth allocation in the vehicle network by the infrastructure is proposed.

[0011] Preferably, the method for achieving accurate dynamic system state includes:

[0012] Analyze user motion models to enhance the universality of the model in different scenarios;

[0013] Consider the changes in speed and angular velocity to reflect the dynamic behavior of the vehicle in different environments;

[0014] Combine motion models and nonlinear prediction algorithms to accurately track vehicle status, thereby better understanding and predicting target behavior in the driving environment;

[0015] Based on experience and analysis of actual data, the statistical characteristics of sensor errors and variances are determined to reduce measurement errors.

[0016] Preferably, the method for obtaining high security value information includes:

[0017] Analyze the target vehicle's movement intention according to its moving speed and deflection angle;

[0018] According to the target vehicle's intention, calculate the difference between its current Euclidean distance to surrounding users and the Euclidean distance under the driver's maximum reaction time;

[0019] Based on the distance difference, the moving speed of the target vehicle and its surrounding users is compared to quantify the safety value between the two.

[0020] According to the number of perceptions contained in the perception container, the size of the perception task is calculated;

[0021] According to the moving speed of the vehicle and the position of the edge node, the transmission rate between the vehicle and the edge node is calculated;

[0022] The latency required for uploading and distributing data is calculated based on the transmission rate between the vehicle and the edge node, the size of the sensing task, and the computing power of the edge node.

[0023] The perceived service quality of the target vehicle is calculated based on the transmission delay of the collaborative perception information and the value of the content.

[0024] Preferably, the high-value content selection learning algorithm based on deep deterministic policy gradient includes:

[0025] Determine whether the transmitted information is useful to the target vehicle based on the quantified safety value;

[0026] Calculate the information entropy based on historical data to determine the user's behavior characteristics and dynamic changes on the road;

[0027] Based on the perception task, determine whether the transmitted information is within the field of view of the target vehicle.

[0028] Preferably, the communication resource allocation and transmission frequency adjustment algorithm includes:

[0029] According to the communication monitoring module of the infrastructure, obtain the real-time channel busy rate;

[0030] According to the collaborative perception information required by the vehicle, the correlation between the target vehicle and its surrounding users is normalized, and the safety value priority scores of road users and the target vehicle are calculated;

[0031] According to the current network load and the number of vehicles, the weighted sum of the safety value priority score and the real-time channel busy rate score of the road users and target vehicles is calculated to ensure that vehicles with high urgency get more bandwidth;

[0032] According to the real-time channel busy rate, the maximum channel busy threshold, the minimum channel busy threshold and the transmission frequency obtained by the last execution of the algorithm, the current transmission frequency is updated using an adaptive step size mechanism.

[0033] The present application also discloses a vehicle-road cooperative perception sharing system based on network twins, including a motion state correction module, a high-value content learning selection module, and a dynamic communication resource allocation module based on different perception requirements of target vehicles and real-time channel busy rate:

[0034] The motion state correction module is used to capture the error caused by sensor measurement based on sensor input, determine the motion model of the road user, and obtain an accurate user state vector;

[0035] The high-value content learning selection module is used to evaluate the magnitude of the correlation between road users, comprehensively consider factors such as the movement intention, speed and direction, and physical proximity of road participants, and send user information that poses a collision threat to the target vehicle;

[0036] The dynamic communication resource allocation module realizes adaptive transmission frequency adjustment and dynamic bandwidth allocation of perception information in multi-vehicle request scenarios by considering the diversity of vehicle perception needs and changes in real-time channel busy rate.

[0037] The beneficial effects of this application are:

[0038] The present application discloses a method and system for vehicle-road collaborative perception sharing based on network twins. By receiving collaborative perception information shared by roadside infrastructure, vehicles can obtain richer real-time information such as road conditions and traffic flow, thereby improving driving safety and traffic efficiency. However, existing sharing strategies face many challenges, including channel congestion caused by frequent data interaction and information redundancy caused by indiscriminate data sharing. To this end, a vehicle-road collaborative perception sharing scheme based on network twins is proposed. This scheme relies on the perception information collected by the twins. By correcting the motion state and quantifying the correlation between road user intentions, the infrastructure selectively shares road information with vehicles, effectively improving the accuracy of information. In addition, in view of the diversity of vehicle perception needs and the changes in real-time channel busy rates, an adaptive transmission frequency adjustment and dynamic bandwidth allocation algorithm for perception information in multi-vehicle request scenarios are designed to avoid channel congestion and resource waste. Compared with traditional sharing strategies, the vehicle perception service quality is improved by 2.45 times, and information redundancy is reduced by 75.4%. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] In order to more clearly illustrate the technical solution of the present application, the following briefly introduces the drawings required for use in the embodiments. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.

[0040] Figure 1 The architecture diagram of the network twin-assisted vehicle-infrastructure perception sharing system;

[0041] Figure 2 This is a system flow chart of Embodiment 1 of the present application;

[0042] Figure 3 This is a schematic diagram of a user motion state vector of Embodiment 1 of the present application;

[0043] Figure 4 This is a schematic diagram of scene classification based on different yaw angular velocities according to the first embodiment of the present application;

[0044] Figure 5 This is a schematic diagram of the Euclidean distance difference based on motion intention in the first embodiment of the present application.

[0045] Figure 6This is a schematic diagram of collaborative perception information value assessment according to Embodiment 1 of the present application; DETAILED DESCRIPTION

[0046] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0047] In order to make the above-mentioned objects, features and advantages of the present application more obvious and easy to understand, the present application is further described in detail below in conjunction with the accompanying drawings and specific implementation methods.

[0048] Embodiment 1

[0049] like Figure 2 As shown, it is a system flow chart of the first embodiment of the present application, which includes the following steps:

[0050] Road User Motion State Estimation: To better adapt to the motion patterns of different road users, the CTRV motion model is introduced to address these limitations. The model more accurately reflects the dynamic behavior of the vehicle in different environments by considering the changes in speed and angular velocity. In order to obtain a more accurate dynamic system state from a series of incomplete and noisy measurements, the Extended Kalman Filter algorithm is used, which handles nonlinear systems by combining sensor inputs and model predictions.

[0051] Specifically, Figure 3 As shown, set As the road user's previous state vector, As the current state vector of the road user, x and y represent the coordinates of the X-axis and Y-axis in the coordinate system respectively; v is the speed scalar; θ is the yaw angle; ω is the yaw angular velocity. Then, the change rate of the user's motion state is

[0052]

[0053] After a time step Δt, its state transfer function can be obtained

[0054]

[0055] In order to enhance the adaptability of the model in different scenarios, the following Figure 4 There are two scenarios: straight and turning: when |ω|>|ω0|, it means the user is turning; when |ω|≤|ω0|, it means the user is moving in an arc. Based on this, the Jacobian matrix of the state transfer function with respect to the state variables under different motion states is derived.

[0056]

[0057] Among them, A is equal to sin(Δtω+θ), B is equal to cos(Δtω+θ), C is equal to sinθ, and D is equal to cosθ. ω0 is the threshold for switching between the two modes, which can be obtained by collecting the yaw rate data of the vehicle in straight and turning motions and analyzing the statistical average value.

[0058] For the design of EKF, let X = [x, y, v, θ, ω] be the system state vector, Q be the covariance matrix of process noise, and P be the system state covariance matrix. Then the state estimation model can be defined as

[0059]

[0060] Q k The model error that depends on the motion comes from the linear acceleration and yaw acceleration noise. The linear acceleration and yaw acceleration satisfy the mean of 0 and the variances are Gaussian distribution, for process noise with mean 0, the control input u in the state transfer equation is 0, and u will not affect the change of state, so the process noise matrix Q is mainly considered k Then, the change of the state vector caused by noise during Δt can be described as

[0061] ΔQ k =G·μ

[0062] According to the definition of covariance and error:

[0063]

[0064] Combining the Kalman gain and the actual measured value to update the state, the process is expressed as

[0065]

[0066] Among them, K k is the calculated Kalman gain; h() represents the nonlinear measurement function of the measurement matrix H that maps the estimation result to the measurement space; J H is the measurement function h() with respect to the state variable X k The Jacobian matrix of k is the measurement value of the road environment perception from the RSU; R is the noise in the measurement process, which depends on the characteristics of the sensor and can be obtained from the actual data.

[0067] like Figure 5 As shown in the figure, after the system motion state is corrected, the Euclidean distance between the target vehicle at time t and any road participant j in the set to be processed is further calculated. Can be described as

[0068]

[0069] Considering the movement intentions of both parties, calculate the human body reaction time T rea The x-coordinate and y-coordinate trajectory points are as follows. When |ω|≤|ω0|, the longitudinal and lateral displacements can be described as:

[0070]

[0071] When |ω|>|ω0|, it is expressed as:

[0072]

[0073] Then, at t+T rea The Euclidean distance between the two parties at the moment Can be recorded as

[0074]

[0075] Finally, we can get and Different Δd t , Δd t Written as

[0076]

[0077] If Δd t is positive, indicating that the distance between the two is approaching, and there may be a risk of collision in the future. It is believed that individual j has safety relevance to vehicle i, Δd t The closer The higher the safety correlation between the two, the higher the safety correlation between the two. If it is non-positive, it is considered that individual j has little or no safety correlation with vehicle i at time t. Therefore, the intention correlation criterion is quantitatively expressed as

[0078]

[0079] Figure 6 The results of perceived information relevance value using EGO as an example are shown.

[0080] For the high-value content selection learning algorithm, the perception sharing model is defined as an MDP, where the twin checks the state st and derives the optimal action that maximizes the reward at rt, where the reward rt is used to determine whether the perceived object is useful for vehicle driving. Therefore, the MDP contains the following three parts.

[0081] s i (t): s i(t) represents the traffic environment information at time t, mainly including cav i The edge server has the location and status information of the surrounding road entities, computing resources and the current channel communication capacity. Therefore, it can be defined as

[0082]

[0083] a i (t)={0,1}:At each moment t, The information of each road participant needs to be considered. i When (t) = 0, it means that the individual information needs to be transmitted to cav i , i.e. it is contained in when a i When (t) = 1, it means that the individual information does not need to be transmitted to cav i .

[0084] r(s i (t),a i (t)): Agent behavior is based on rewards, that is, each behavior is for the purpose of obtaining higher rewards, so appropriate rewards play a vital role in the proposed algorithm. The reward function comprehensively considers three important factors in the vehicle interaction process to more comprehensively evaluate the impact of individual behavior on the traffic system. Speed ​​entropy is a measure of its own movement pattern, reflecting the behavioral characteristics and dynamic changes of individuals on the road; whether individual j is positive to cav i Visible. Compared with cav i The agent pays more attention to the dangerous individuals in the visible field of view, and the agent pays more attention to the non-line-of-sight position threats. Therefore, the reward function is expressed as

[0085]

[0086] Among them, EV j It represents the speed entropy of individual j, which can be calculated from the historical data collected by the twin. Insight is a hyperparameter, considering j in cav i It will have greater value outside the field of vision, and its value will decrease accordingly within the field of vision, so set it to Insight = {0.8, 1}.

[0087] For the communication resource allocation algorithm: According to the different perception requirements of the target vehicles and the real-time channel busy rate (CBR), the infrastructure can adaptively adjust the transmission frequency of perception information (CPM) in the vehicle network and dynamically allocate bandwidth. CBR is defined as the ratio of the time the channel is occupied to the total time in a certain period of time, reflecting the communication environment between RSU and vehicles in the I2V network.

[0088] Contains the currently available communication bandwidth Btotal , the target vehicle set V and the collaborative perception information they need Share poi , real-time channel busy rate CBR, CBR maximum threshold CBR max and CBR minimum threshold CBR min , maximum CPM transmission frequency TF max and the CPM transmission frequency TF determined in the last execution of the algorithm last The acquisition of real-time CBR is achieved through real-time monitoring by the communication monitoring module of RSU. When CBR is lower than the minimum threshold, it means that the network communication conditions are good. On the contrary, when CBR exceeds the maximum threshold, it means that the network communication conditions are poor. It is worth noting that the hardware conditions of the RSU communication module play a decisive role in the maximum transmission frequency of CPM.

[0089] The transmission frequency is set based on monitoring the real-time CBR. If CBR=1, it means that the current channel is in a completely blocked state, and the CPM transmission frequency is set to 0 because no data transmission can be performed at this time.

[0090] According to the current network load and the number of vehicles, the available bandwidth is dynamically calculated, and a priority-based and real-time CBR weighted formula is designed to ensure that vehicles with high urgency get more bandwidth, where the priority is shared with the POI set Shared with the target vehicle poi Related, Share poi The higher the relevance of the included road individuals, the more urgent the vehicle is. The priority formula can be expressed as

[0091]

[0092] Among them, S cbr =1-CBR,S pri =max(Relevance poi )

[0093] When CBR <CBR min When the adaptive step size is set to

[0094]

[0095] Send frequency update to

[0096] TF=min(TF max , TF last +L)

[0097] When CBR>CBR max When the adaptive step size is set to

[0098]

[0099] Send frequency update to

[0100] TF=max(1,TF last -L)

[0101] When CBR min ≤CBR≤CBR max There is no need to adjust the sending frequency.

[0102] The embodiments described above are only descriptions of the preferred methods of the present application, and are not intended to limit the scope of the present application. Without departing from the design spirit of the present application, various modifications and improvements made to the technical solutions of the present application by ordinary technicians in this field should fall within the scope of protection determined by the claims of the present application.

Claims

1. A vehicle-road cooperative perception sharing method based on network twins, characterized in that: The method comprises the following steps: Based on the modified motion intention, the traffic safety information value between road users is quantified, and the infrastructure selectively shares high-value information with target vehicles; Based on deep deterministic policy gradient and velocity entropy, a twin high-value content selection learning algorithm is proposed; Adjust the information transmission frequency based on the different perception needs of vehicles and the real-time channel busy rate of the infrastructure in the vehicle network; Integrating the above three optimization methods into one system can improve the vehicle perception service quality and reduce the overall redundancy of the system while maintaining information accuracy.

2. The vehicle-road cooperative perception sharing method based on network twinning according to claim 1 is characterized in that: Methods for quantifying safety value include: Based on the global road information collected by roadside infrastructure, combined with sensor characteristics, the data interference caused by noise is analyzed to calculate the road user state vector at the current moment; Analyze the target vehicle's movement intention according to its moving speed and deflection angle; According to the target vehicle's intention, calculate the difference between its current Euclidean distance to surrounding users and the Euclidean distance under the driver's maximum reaction time; Based on the distance difference, the movement speed of the target vehicle and its surrounding users is compared to quantify the safety value between the two.

3. The vehicle-road cooperative perception sharing method based on network twinning according to claim 1 is characterized in that: Information sharing decisions include: The high-value content selection strategy of the twin is trained based on deep deterministic policy gradient. At the same time, information entropy technology is introduced to encourage the twin to explore the action space more extensively during training and improve its adaptability to unknown environments. Determine whether the transmitted information is useful to the target vehicle based on the quantified safety value; Calculate the information entropy based on historical data to determine the user's behavior characteristics and dynamic changes on the road; Based on the perception task, determine whether the transmitted information is within the field of view of the target vehicle.

4. The vehicle-road cooperative perception sharing method based on network twinning according to claim 1 is characterized in that: Communication resource allocation and transmission frequency adjustment: According to the communication monitoring module of the infrastructure, obtain the real-time channel busy rate; According to the collaborative perception information required by the vehicle, the correlation between the target vehicle and its surrounding users is normalized, and the safety value priority scores of road users and the target vehicle are calculated; According to the current network load and the number of vehicles, the weighted sum of the safety value priority score and the real-time channel busy rate score of the road users and target vehicles is calculated to ensure that vehicles with high urgency get more bandwidth; According to the real-time channel busy rate, the maximum channel threshold, the minimum channel threshold and the transmission frequency obtained by the last execution of the algorithm, the current transmission frequency is adjusted using an adaptive step size mechanism.

5. A vehicle-road cooperative perception sharing system based on network twins, characterized in that: The twin receives global road information collected by roadside infrastructure to build a parallel virtual environment of the vehicle-to-roadside infrastructure network, and integrates user intention correction, high safety value selection, and dynamic transmission frequency adjustment methods under multi-vehicle requests into the system. The various methods work together and interact with each other to jointly complete the perception decision of the current vehicle in infrastructure sharing.

Images