Risk prediction and control method for automatic driving vehicle confluence area

Abstract

The invention discloses a risk prediction and control method for an automatic driving vehicle confluence area. The method comprises the following steps: S1, constructing a safety field model; s2, on the basis of the safety field model in the S1, a prediction input sequence of the vehicle state is obtained by combining model prediction control MPC; s3, constructing a total cost function of the vehicle decision; s4, on the basis of the total cost function in the S3, constructing a game model of the vehicle and each surrounding vehicle, and performing optimization solution on the model; and S5, obtaining an optimal control input sequence of the vehicle based on the prediction control input sequence obtained in the step S2, the target function and the optimization decision in the step S4. The problems that the risk prediction and control strategy is not comprehensive enough, the flexibility is poor, the risk prediction is inaccurate, and training data dependence exists in the existing method are solved.

Description

Technical Field

[0001] The present invention belongs to the technical field of autonomous driving, and in particular relates to a risk prediction and control method for an autonomous driving vehicle merging area. Background Art

[0002] The merging and diverging of vehicles in the weaving area is a common traffic scene. Improper handling may lead to serious traffic congestion or even accidents. The merging vehicles have to slow down or even stop on the ramp to wait for the right opportunity to merge. When vehicles on the auxiliary road merge into and exit the main road, on the one hand, the front and rear vehicles may not be far enough apart when merging due to blind spots; on the other hand, different drivers have different driving abilities and driving styles. At the same time, drivers cannot synchronize observation, calculation and action well in a very short time, resulting in the failure of vehicles to adjust their speed in time when merging, causing frequent traffic accidents such as rear-end collisions and side scrapes.

[0003] In the context of my country's vigorous development of connected autonomous driving, the introduction of intelligent connected vehicles is seen as a potential solution. First, intelligent connected vehicles can accurately perceive the surrounding environment and exchange information with other vehicles in real time, such as speed, direction, and expected behavior, thereby improving the vehicle's perception ability in weaving areas. This enhanced perception ability helps reduce the risk of accidents caused by blind spots, especially in complex traffic scenarios. Secondly, through vehicle-to-vehicle and vehicle-to-infrastructure communications, intelligent connected vehicles can more effectively coordinate merging and diverging actions, optimize the vehicle's driving path, and reduce unnecessary deceleration and parking.

[0004] In order to analyze the longitudinal and lateral interactions of vehicles on the road, many scholars have studied the interactive behaviors between vehicles, considered the driving risks caused by various traffic factors, and evaluated the potential driving risks in real traffic scenarios. The current risk prediction and control methods for autonomous vehicles usually have the following shortcomings: in complex scenarios such as merging areas, the interactive effects between multiple vehicles are usually ignored, resulting in incomplete risk prediction and control strategies; the rules are single and inflexible, and cannot cover all the complex interactions in actual driving scenarios. The ability to handle complex scenarios is insufficient, resulting in inaccurate risk prediction; there is a problem of dependence on training data, and in practical applications, it may be challenging to obtain comprehensive and accurate data. Summary of the invention

[0005] The purpose of an embodiment of the present invention is to provide a risk prediction and control method for an autonomous driving vehicle merging area, so as to solve the problems of the existing methods, such as the risk prediction and control strategies are not comprehensive enough, have poor flexibility, inaccurate risk prediction, and are dependent on training data.

[0006] In order to solve the above technical problems, the technical solution adopted by the present invention is:

[0007] The risk prediction and control method of the merging area of ​​the autonomous driving vehicle includes the following steps:

[0008] S1: Construct a safety field model;

[0009] S2: Based on the safety field model of S1, the predicted input sequence of vehicle state is obtained in combination with model predictive control MPC;

[0010] S3: Construct the total cost function of vehicle decision;

[0011] S4: Based on the total cost function of S3, a game model between the vehicle and each surrounding vehicle is constructed, and the model is optimized and solved;

[0012] S5: Based on the predicted control input sequence obtained in S2, the objective function, and the optimization decision of S4, the optimal control input sequence of the vehicle is obtained.

[0013] Furthermore, S1 includes:

[0014] For the active vehicle, the safety field force generated by the surrounding vehicles on the active vehicle is calculated using the following formula:

[0015] P ov (X,Y)=a ov e η (1)

[0016] Where (X, Y) is the position of the active vehicle, P ov (X,Y) represents the safety field force generated by the surrounding vehicles at the position (X,Y) on the active vehicle, X represents the horizontal coordinate of the active vehicle position, and Y represents the vertical coordinate of the active vehicle position; a ov It represents the maximum field force generated by the surrounding vehicles; e is the base of the natural logarithm, and η is the safety field force spatial distribution function;

[0017]

[0018] Among them, (X ov ,Y ov ) represents the center of gravity of the surrounding vehicles; X ov Indicates the horizontal coordinate of the center of gravity of the surrounding vehicles, Y ov The vertical coordinate of the center of gravity of the vehicle; ρ x , ρ y Respectively represent the convergence coefficients in the horizontal and vertical directions; the vertical direction refers to the direction along which the vehicle moves, and the horizontal direction refers to the direction perpendicular to the direction in which the vehicle moves; represents the lateral speed of the surrounding vehicles; b represents the vehicle specification coefficient; ξ represents the relative speed of the surrounding vehicles to the active vehicle; c represents the weight factor;

[0019]

[0020] Among them, k ov Represents the safety field polarity function.

[0021] Furthermore, the S1 further includes:

[0022] For active vehicles, the safety field force generated by road markings on the active vehicle is calculated using the following formula:

[0023]

[0024] Among them, P r (X,Y) represents the safety field force generated by the road marking at position (X,Y), a r represents the maximum field force of the road; d represents the minimum distance from the position (X, Y) to the road marking; d r represents the safety threshold; W represents the active vehicle width;

[0025] For active vehicles, the total safety field force on the active vehicle is calculated using the following formula:

[0026]

[0027] Among them, P c (X,Y) represents the total safety field force received by the active vehicle at position (X,Y); P i ov (X,Y) represents the safety field force generated by the i-th surrounding vehicle on the position (X,Y) of the active vehicle; m is the total number of surrounding vehicles; represents the impact of the jth road marking on the safety field at the active vehicle's position (X, Y); n is the total number of road markings.

[0028] Further, the S2 includes:

[0029] S21: Determine the vehicle dynamics model;

[0030]

[0031] Where: The lateral acceleration a represents the rate of change of the active vehicle's velocity in the lateral direction. x,δ ; ν x,δ represents the lateral velocity of the active vehicle; ν y,δ represents the velocity of the active vehicle in the longitudinal direction; The longitudinal acceleration a represents the rate of change of the active vehicle's velocity in the longitudinal direction. y,δ ; Indicates the direction angle of the active vehicle; X δ , Y δRespectively represent the position of the active vehicle in the X and Y directions, represents the position X of the active vehicle δ , Y δ The rate of change over time is the velocity in the horizontal and vertical directions ν x,δ , ν y,δ ; ν δ represents the speed of the active vehicle, δ represents the lane marking;

[0032] S22: The control equation of the active vehicle is expressed as:

[0033]

[0034] Where: T is the transpose, x represents the state vector of the active vehicle; u represents the control input vector of the active vehicle; f[x,u] represents the function based on the vehicle state mapping to vehicle control.

[0035] Furthermore, the S2 further includes:

[0036] Based on the model predictive control MPC, equation (8) is expressed as follows:

[0037] x(k+1)=x(k)+ΔT·f[x(k),u(k)](9)

[0038] Where: x(k) represents the state vector of the active vehicle at the kth moment; x(k+1) represents the state vector of the active vehicle at k+1, i.e., the predicted state vector; u(k) represents the control input vector of the active vehicle at the kth moment; f[x(k),u(k)] represents the change of the state of the active vehicle under the state vector x(k) and the control input vector u(k); ΔT represents the sampling interval;

[0039] y(k) represents the output vector of the active vehicle at the kth moment. Based on equation (6), the output vector y(k) is expressed as a parameter based on the safety field force:

[0040]

[0041] In the formula, g represents the output function; represents the lateral velocity at time k; represents the vertical velocity at time k, It represents the total safety field force on the vehicle at time k;

[0042] Based on model predictive control MPC, the vehicle control input vector is predicted:

[0043]

[0044] In the formula, y(k+1|k) represents the output vector predicted at time k+1; x(k+1|k) represents the state vector predicted at time k+1, u(k|k) represents the control input vector at the current time k, and N p 、N c Respectively represent the horizontal prediction steps and the vertical prediction steps, N p ≥N c ;

[0045] The output sequence y(k) and the control input sequence u(k) are:

[0046] y(k)=[y(k+1|k),y(k+2|k),...y(k+N p |k)] T (12)

[0047] u(k)=[u(k|k),u(k+1|k),... u(k+N p -1|k)] T (13).

[0048] Further, the S3 includes:

[0049] Calculate the cost function of driving safety ds , the calculation formula is as follows:

[0050] Υ ds =‖α|-1|Υ ds-log +|α|Υ ds-lat (14)

[0051] Where: ds-log , Υ ds-lat They represent the longitudinal and lateral safety costs of the active vehicle respectively; α represents the lane-changing action function of the vehicle, α∈{-1,0,1}={left lane change, no lane change, right lane change};

[0052] Υ ds-log =κ ν-log λ δ / [(Δν y,δ ) 2 +υ]+κ s-log / [(Δs y,δ ) 2 +υ](15)

[0053] Υ ds-lat =κ ν-lat λ δ+α / [(Δν x,δ+α ) 2 +υ]+κ s-lat / [(Δs x,δ+α ) 2 +υ](16)

[0054] In the formula, κ ν-log , κ ν-lat represents the vertical and horizontal risk weighting coefficients; δ , δ+α Indicates the weight factor related to control in longitudinal and lateral control; Δν y,δ , Δν x,δ+α It represents the speed error in longitudinal and lateral control, that is, the difference between the expected speed and the actual speed; υ represents the smoothing factor; κ s-log , κ s-lat Indicates the risk weighting coefficient related to the vertical and horizontal positions in vertical and horizontal control; Δs y,δ , Δs x,δ+α It represents the position error in longitudinal and lateral control, that is, the difference between the desired position and the actual position;

[0055] Cost function of driving comfort The calculation formula is as follows:

[0056]

[0057] Where: a x,δ , a y,δ They represent the lateral acceleration and longitudinal acceleration of the active vehicle respectively; Respectively represent the weight coefficients of lateral acceleration and longitudinal acceleration;

[0058] Cost function of driving efficiency pe The calculation formula is as follows:

[0059]

[0060] Where: ν y,δ represents the actual longitudinal speed on lane δ; Indicates the target speed; represents the speed limit of lane δ;

[0061] The total cost function Υ calculation formula for vehicle decision-making is as follows:

[0062]

[0063] in, They represent the weighting coefficients of the cost function for driving safety, the cost function for driving comfort, and the cost function for driving efficiency respectively.

[0064] Further, the S4 includes:

[0065] A game model between the active vehicle and surrounding vehicles is established, and the calculation formula is as follows:

[0066]

[0067] In formula (21), represents the longitudinal optimization acceleration of the active vehicle; α * represents the optimized lane-changing decision of the active vehicle; represents the maximum total cost function when the active vehicle plays the game; argmin represents the optimal control of the active vehicle;

[0068] In formula (22), a y,δ ,α represents the acceleration and lane turning action of the active vehicle; represents the acceleration decision after the active vehicle interacts with the i-th surrounding vehicle; i represents different surrounding vehicles interacting with the active vehicle, lov represents the vehicle in front of the active vehicle, and rov represents the vehicle behind the active vehicle; γ 2 (a y,δ ,α) represents the optimal action of the surrounding vehicles based on the behavior of the active vehicle; ζ represents the decision parameter in the interactive game between the active vehicle and the surrounding vehicles, Φ 2 represents the action space; i (a y,δ ,α,ζ) is the cost function of any action taken by the surrounding vehicles based on the active vehicle behavior, is the cost of the actions taken by the surrounding vehicles based on the acceleration decisions of the active vehicle;

[0069] During the game, the active vehicle should satisfy the following equation:

[0070]

[0071] in, represents the maximum longitudinal speed of the active vehicle; represents the minimum longitudinal acceleration and maximum longitudinal acceleration of the active vehicle, ν y,δ represents the longitudinal velocity of the active vehicle.

[0072] Further, the S5 includes:

[0073] The decision-making process of active vehicles should meet the following requirements:

[0074]

[0075] In formula (24), J p (k) is the objective function at time k, Δy(k) represents the output error, i.e., the difference between the expected output and the actual output; ΔY(k) represents the difference between the expected lateral position and the actual lateral position; Indicates the rate of change of the control input; Q 1 , Q 2, R represent the weight matrices of output error, lateral position error, and control input change rate, respectively; Δy T , ΔY T , They are Δy(k), ΔY(k), The transposed matrix of

[0076] In formula (25), u * (k) represents the optimal control input sequence, argmin represents the optimal control input sequence by minimizing J p (k) optimizes the vehicle control input u'(k); u'(k) represents the optimized control input sequence obtained by the model predictive control MPC and the game process, which is obtained by the simultaneous equations (13), (21), and (23);

[0077] A k represents the active vehicle state matrix; B k represents the active vehicle control matrix;

[0078] Finally, the optimal control input sequence of the vehicle is obtained.

[0079] The beneficial effects of the present invention are

[0080] 1. Accurate risk prediction and multi-vehicle interaction modeling. The present invention establishes a model of the safety field force on active vehicles, taking into account the impact of surrounding vehicles and road infrastructure. This model effectively quantifies the safety field force in multi-vehicle interaction scenarios, and can predict the interaction risk between active vehicles and surrounding vehicles in real time, and more accurately predict the risk of vehicles in merging areas. This multi-vehicle interaction modeling method overcomes the shortcomings of traditional methods that ignore the mutual influence between vehicles, making risk prediction more comprehensive and accurate.

[0081] 2. Dynamic decision-making and control, real-time adjustment of strategies. The present invention combines the model predictive control (MPC) method with the real-time status of the vehicle and environmental perception information to obtain an integrated vehicle decision-control method to achieve a close integration of vehicle decision-making and control, and can adjust the decision-making strategy in real time to adapt to changes in the traffic environment. It improves the flexibility of decision-making and control and overcomes the shortcomings of fixed rules and models in traditional methods that cannot adapt to dynamic environments. At the same time, the method of the present invention takes into account the safety, efficiency and comfort of the vehicle during driving, ensuring the safety and efficiency of the vehicle in merging and diverging operations.

[0082] 3. Based on the safety field model and the real-time status of vehicles, the present invention can predict the interaction risk between vehicles in real time, and make dynamic and intelligent real-time risk prediction and driving decisions by building a game process between vehicles. It can optimize the operation strategy of vehicles in the merging area, reasonably allocate vehicle resources, and achieve more efficient and safe merging and diversion. The application of this game theory effectively improves the shortcomings of the existing technology in handling complex interaction situations and improves the smoothness of traffic flow. BRIEF DESCRIPTION OF THE DRAWINGS

[0083] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. 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 creative work.

[0084] Figure 1 It is a flow chart of the method of the present invention.

[0085] Figure 2 This is a schematic diagram of the scenario where an autonomous vehicle enters a merging area.

[0086] Figure 3 This is a comparison chart of the average speed of autonomous driving vehicles entering the merging area during the training process of the method of the present invention and other methods.

[0087] Figure 4 It is a comparison chart of the collision probability of an autonomous driving vehicle entering a merging area during the training process between the inventive method and other methods. DETAILED DESCRIPTION

[0088] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments 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 creative work are within the scope of protection of the present invention.

[0089] Driving Safety Field (DSF) is a "physical field" that characterizes the degree of influence of various elements of people, vehicles and roads in the vehicle driving environment on driving risks. It consists of the "kinetic energy field" related to the vehicle, the "potential energy field" related to the road, and the "behavior field" related to people. The main idea is to treat traffic elements as charged charges, use field strength to characterize the influence of different attribute elements on driving risks, and use field force to explain the interaction between elements. Using the driving safety field model, the risk posed by various traffic elements on the road to a certain driving vehicle can be calculated.

[0090] CAV driving characteristics: Connected and Autonomous Vehicle (CAV) can integrate communication and network technologies based on sensors, controllers, actuators and other devices, and have functions such as environmental perception, intelligent decision-making, and collaborative control to achieve safe, efficient, comfortable, and energy-saving intelligent driving, including intelligence and networking.

[0091] The method of the present invention considers the interaction between the active vehicle on the ramp of the expressway merge area and the surrounding vehicles on the main line. The experimental scenario of the embodiment is as follows: Figure 2 As shown, both CAV1 and CAV2 have the motivation to change lanes. If CAV2 is used as the active vehicle, the main line vehicles within a certain range of it can be used as surrounding vehicles. The impact of these surrounding vehicles on the active vehicle is considered in the subsequent safety field model and game process. In addition, the number and type of surrounding vehicles (CAV or HDV) are random in this embodiment, considering the vehicle decision control effect of the active vehicle under different road conditions.

[0092] This embodiment provides a risk prediction and control method for an autonomous driving vehicle merging area. Figure 1 , the complete steps are as follows:

[0093] S1: Construct a safety field model.

[0094] Potential field is an effective method to simulate vehicle dynamics and describe its interaction with surrounding obstacles. The safety field of road users can be regarded as a repulsive field generated by a moving vehicle, which maintains a safe buffer with surrounding objects. The active vehicle is at position (X, Y), and the safety field force generated by surrounding vehicles can be expressed as:

[0095] P ov (X,Y)=a ov e η (1)

[0096]

[0097] Where: P ov (X,Y) represents the safety field force generated by the surrounding vehicles at position (X,Y); a ov Represents the maximum field force of the surrounding vehicles, which is a constant here; e η represents the spatial distribution of the potential field, expressed in natural logarithms, η is the spatial distribution function, and the natural logarithm is used to ensure the non-negativity of the spatial distribution; X represents the abscissa of the active vehicle position, and Y represents the ordinate of the active vehicle position; (X ov ,Y ov ) represents the center of gravity of the surrounding vehicles; X ovIndicates the horizontal coordinate of the center of gravity of the surrounding vehicles, Y ov The vertical coordinate of the center of gravity of the vehicle; ρ x , ρ y Represent the convergence coefficients in the horizontal and vertical directions respectively; represents the lateral speed of the surrounding vehicles; b represents the specification coefficient of the active vehicle; ξ represents the relative speed of the vehicles around the active vehicle, reflecting the influence of the speed of the surrounding vehicles on the active vehicle; c represents the weight factor, which is used to adjust the influence of ξ on η; k ov is the potential field polarity function, when (XX ov )<0, indicating that the potential field strength increases in the horizontal direction, XX ov ≥0 means that the potential field strength decays or remains unchanged in the transverse direction.

[0098] The safety field of an active vehicle affected by road markings can be expressed as:

[0099]

[0100] Where: P r (X,Y) represents the safety field force generated by the road factor at the position (X,Y), a r represents the maximum field force of the road, which is a constant here; d represents the minimum distance from the position (X, Y) to the road marking; d r represents the safety threshold; W represents the active vehicle width.

[0101] The safety field model considering surrounding vehicles and road infrastructure can be expressed as:

[0102]

[0103] Where: P c (X,Y) represents the total safety field force received by the active vehicle at position (X,Y); P i ov (X,Y) represents the safety field force generated by the i-th surrounding vehicle on the active vehicle at its position (X,Y); m is the number of surrounding vehicles; represents the impact of the jth road marking on the safety field at the active vehicle's position (X, Y), and n is the total number of road markings.

[0104] S2: Based on the safety field force of vehicles in multi-vehicle interaction scenarios, combined with the MPC model, the vehicle status in multi-vehicle scenarios is predicted and controlled.

[0105] First, determine the dynamics models of various types of vehicles. Based on existing research, this embodiment uses the MOBIL model to represent the driving characteristics of CAV vehicles. In the MOBIL model, the lane-changing behavior of a vehicle not only considers its own acceleration changes, but also the acceleration changes of other vehicles in adjacent lanes to avoid traffic instability caused by lane changes. The IDM model is used to represent the driving characteristics of HDV vehicles, simulating the following behavior of human-driven HDV vehicles, taking into account the distance and speed difference between vehicles, so as to calculate the appropriate acceleration to ensure safe following. Predict and control the vehicle status in multi-vehicle scenarios.

[0106] S21: First, determine the vehicle dynamics model. This formula describes the vehicle dynamics model, including the calculation of speed, acceleration, heading angle (head direction) and position, simulating the movement behavior of the vehicle on a plane.

[0107]

[0108] Where: The lateral acceleration a represents the rate of change of the active vehicle's velocity in the lateral direction. x,δ ; ν x,δ represents the velocity in the horizontal direction; ν y,δ Indicates the speed in the longitudinal direction; Indicates the direction angle of the active vehicle, that is, the angle between the vehicle's current driving direction and geographic north; Describes the position X of the active vehicle in the horizontal and vertical directions respectively δ , Y δ The rate of change of the horizontal and vertical speed is given by the vehicle speed ν δ and direction angle; δ represents the lane marking.

[0109] S22: Based on the vehicle dynamics model, the input and output states of the vehicle can be determined. By making decisions on the vehicle output state, the vehicle is controlled in combination with MPC. Therefore, the input state also takes into account the prediction of the vehicle state. The vehicle control model can be expressed as:

[0110]

[0111] Where: The state vector representing the vehicle includes: the vehicle's position, speed, direction angle and other information; u represents the vehicle's control vector, that is, all input signals applied to the system by the control system, such as acceleration, steering wheel angle, etc.; f[x,u] represents the function based on the vehicle state mapped to the vehicle control;

[0112] Based on the discreteness of the MPC control system, the above formula can be expressed as follows:

[0113] x(k+1)=x(k)+ΔT·f[x(k),u(k)](9)

[0114] Where: x(k) represents the state vector of the system at the kth moment, which contains all the state variables of the system at that moment; x(k+1) represents the state vector of the system at k+1, which is the predicted system state; f[x(k),u(k)] describes the change of the system state under the conditions of state x(k) and input u(k), which is usually determined by the vehicle's dynamic model and describes the evolution law of the system under given state and control input; ΔT represents the sampling interval.

[0115] Based on the MPC prediction results, the output vector y(k) is expressed as a parameter based on the safety field force value to achieve real-time prediction and control based on the safety field force:

[0116]

[0117] Where y(k) represents the output vector of the vehicle system at time k; g represents the output function; represents the lateral velocity at time k; represents the vertical velocity at time k, It represents the total safety field force (or potential field strength) that the vehicle is subject to at time k.

[0118] Based on MPC, multi-step prediction is performed:

[0119]

[0120] In the formula, y(k+1|k) represents the system output vector at the first step of prediction at time k+1; x(k+1|k) represents the predicted state at time k+1; u(k|k) represents the input at the current time k, and so on. N p , N c Respectively represent the horizontal prediction steps and the vertical prediction steps, and N p ≥N c .

[0121] The output sequence and control input sequence are:

[0122] y(k)=[y(k+1|k),y(k+2|k),...y(k+N p |k)] T (12)

[0123] u(k)=[u(k|k),u(k+1|k),...u(k+N p -1|k)] T (13)

[0124] S3: Construct the cost function of vehicle decision-making. Consider three factors: driving safety, riding comfort and driving efficiency to evaluate the decision cost of the vehicle. For active vehicles, driving safety mainly includes the influence of lateral movement and longitudinal movement. The cost function of driving safety is ds Specifically as follows:

[0125] Υ ds =‖α-1|Υ ds-log +|αΥ ds-lat (14)

[0126] Where: ds-log , Υ ds-lat They represent the longitudinal and lateral safety costs of the active vehicle respectively. The longitudinal direction refers to the movement along the vehicle's forward direction, while the lateral direction refers to the movement perpendicular to the vehicle's forward direction. α represents the lane-changing action function of the vehicle, α∈{-1,0,1}:={left lane change, no lane change, right lane change}.

[0127] The horizontal and vertical security costs can be expressed as follows:

[0128] Υ ds-log =κ ν-log λ δ / [(Δν y,δ ) 2 +υ]+κ s-log / [(Δs y,δ ) 2 +υ](15)

[0129] Υ ds-lat =κ ν-lat λ δ+α / [(Δν x,δ+α ) 2 +υ]+κ s-lat / [(Δs x,δ+α ) 2 +υ](16)

[0130] In the formula, κ ν-log , κ ν-lat represents the vertical and horizontal risk weighting coefficients, which are used to adjust the speed-related risk costs; δ , δ+α Represents the weight factor related to the control input (such as steering angle, acceleration) in longitudinal and lateral control; Δν y,δ , Δν x,δ+α It represents the speed error in longitudinal and lateral control, usually the difference between the expected speed and the actual speed; υ represents the smoothing factor, which avoids the denominator being zero and increases the stability of the calculation; κ s-log , κ s-lat Indicates the risk weighting coefficient related to the vertical and horizontal positions in vertical and horizontal control; Δsy,δ , Δs x,δ+α It indicates the position error in longitudinal and lateral control, usually referring to the difference between the desired position and the actual position.

[0131] Cost function of driving comfort rc It can be expressed as:

[0132]

[0133] Where: a x,δ , a y,δ They represent the lateral acceleration and longitudinal acceleration of the active vehicle respectively; Represent the weight coefficients of lateral acceleration and longitudinal acceleration respectively.

[0134] Driving efficiency is mainly related to the longitudinal motion of the vehicle, and the cost function of driving efficiency can be expressed as:

[0135]

[0136] Where: pe The cost function representing driving efficiency, which represents the difference between the actual speed of the vehicle and the target speed This measures whether the vehicle's speed is at or close to the ideal value, thereby evaluating driving efficiency; ν y,δ represents the actual longitudinal speed on lane δ; Indicates the speed limit of lane δ, reflecting that when making vehicle decisions, in order to ensure safety and driving efficiency, the target speed of the vehicle cannot exceed the speed limit of the lane However, in the cost function of driving efficiency, by comparing the actual speed with the lane speed limit, it is ensured that the vehicle's speed does not exceed the safety limit and that the driving efficiency is not affected due to excessive deceleration.

[0137] Finally, the total cost function Υ when the vehicle makes a decision hν It is expressed as:

[0138]

[0139] in, They represent the weighted coefficients of the cost function for driving safety, the cost function coefficient for driving comfort, and the cost function coefficient for driving efficiency, respectively, and are used to adjust the impact of each cost item on the decision, thereby obtaining a comprehensive cost evaluation.

[0140] S4: Based on the vehicle decision cost function, construct the Stackelberg game process between the active vehicle and each surrounding vehicle, and perform a two-way optimization solution. Establish the game model:

[0141]

[0142] In formula (21), represents the longitudinal optimization acceleration of the active vehicle; α * represents the lane-changing decision optimization of the active vehicle; represents the maximum total cost function when the active vehicle plays the game;

[0143] In formula (22), a y,δ ,α represents the acceleration and lane turning action of the active vehicle; represents the acceleration decision after the active vehicle interacts with the i-th surrounding vehicle; i represents different surrounding vehicles interacting with the active vehicle, lov represents the vehicle in front of the active vehicle, and rov represents the vehicle behind the active vehicle; γ 2 (a y,δ ,α) represents the optimal action of the surrounding vehicles based on the behavior of the active vehicle; ζ represents the decision parameter in the interactive game between the active vehicle and the surrounding vehicles. This decision is specifically manifested as the adjustment of acceleration, Φ 2 Represents the action space, i.e., the decision range of each vehicle’s acceleration. In the vehicle game model, it ensures that the vehicle’s control input complies with physical limitations and remains reasonable in the game formed between all vehicles; i (a y,δ ,α,ζ) is the cost function of any action made by the surrounding vehicles based on the active vehicle behavior. This function is a global strategy cost evaluation function, which is used to analyze the interaction costs of surrounding vehicles under various possible strategies in a specific game scenario. It is the cost of the actions taken by the surrounding vehicles based on the acceleration decision of the active vehicle. It evaluates the actual reaction cost of the surrounding vehicles when responding to a specific action of the active vehicle (such as acceleration adjustment).

[0144] In addition, the speed and acceleration constraints of the vehicle in the actual scenario should also be considered during the game:

[0145]

[0146] in, represents the maximum longitudinal speed of the active vehicle; represents the minimum longitudinal acceleration and maximum longitudinal acceleration of the active vehicle, ν y,δ represents the longitudinal velocity of the active vehicle.

[0147] S5: Combining the cost function of S3 with the game model of the vehicle interaction process in S4, the decision-making control of the autonomous driving vehicle can be transformed into a constrained optimization problem:

[0148]

[0149] Where: u min 、umax represents the minimum control vector and the maximum control vector of the active vehicle; J p (k) is the control performance index, i.e., the objective function at time k, which is specifically for the vehicle's control performance. The optimization goal is to improve the efficiency of the control input and the accuracy of the vehicle's trajectory. Δy(k) represents the output error, which is the difference between the expected output and the actual output, which reflects the deviation of the vehicle's current state from the expected state (such as position, speed); ΔY(k) represents the difference between the expected lateral position and the actual lateral position, which reflects the deviation from the predetermined trajectory in the lateral position. For autonomous vehicles, staying in the center of the lane or avoiding obstacles are important lateral position control goals; It indicates the rate of change of control, which reflects the adjustment strength of the control strategy; frequent or excessive input changes will lead to system instability; Q 1 , Q 2 , R represent the output error, lateral position error, and control change rate weight matrix respectively; represents the optimal control of the active vehicle by minimizing J p (k) Optimize vehicle control. ΔY T , They are Δy(k), ΔY(k), The transposed matrix of

[0150] A k represents the active vehicle state matrix; B k represents the active vehicle control matrix, obtained by equation (7) and equation (8);

[0151] In formula (25), u * (k) represents the optimal control input sequence, argmin represents the optimal control input sequence by minimizing J p (k) optimizes the vehicle control input u'(k); u'(k) represents the optimized control input sequence obtained by MPC and game theory, which is obtained by combining equations (13), (21), and (23);

[0152] The optimized optimal control sequence can be expressed as:

[0153] u * (k)=[u * (k|k),u * (k+1|k),...,u * (k+N p -1|k)] T , which can be used to update vehicle status and perform real-time control.

[0154] Figure 3This is a comparison chart of the average speeds of autonomous driving vehicles entering the merging area during the training process. It can be seen that with the increase in the number of training rounds, the average speed of the method of the present invention is significantly higher than that of the existing rule method and potential field method, showing its superiority.

[0155] Figure 4 The method of the present invention has significant advantages in improving safety and can effectively reduce the collision rate of vehicles in the merging area. Compared with the traditional rule method and potential field method, the effect of the present invention is more significant, showing better learning ability and adaptability, so that the vehicle can plan and drive the path more safely after training.

[0156] Each embodiment in this specification is described in a related manner, and the same or similar parts between the embodiments can be referred to each other, and each embodiment focuses on the differences from other embodiments. In particular, for the system embodiment, since it is basically similar to the method embodiment, the description is relatively simple, and the relevant parts can be referred to the partial description of the method embodiment.

[0157] The above description is only a preferred embodiment of the present invention and is not intended to limit the protection scope of the present invention. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention are included in the protection scope of the present invention.

Claims

1. A risk prediction and control method for an autonomous vehicle merging area, characterized in that: The following steps are involved: S1: Construct a safety field model; S2: Based on the safety field model of S1, the predicted input sequence of vehicle state is obtained in combination with model predictive control (MPC); S3: Construct the total cost function of vehicle decision; S4: Based on the total cost function of S3, a game model between the active vehicle and surrounding vehicles is constructed, and the model is optimized and solved; S5: Based on the predicted control input sequence obtained in S2, the objective function, and the optimization decision of S4, the optimal control input sequence of the vehicle is obtained.

2. The risk prediction and control method for the merging area of ​​an autonomous driving vehicle according to claim 1, characterized in that: S1 includes: For the active vehicle, the safety field force generated by the surrounding vehicles on the active vehicle is calculated using the following formula: P ov (X,Y)=a ov e η (1) Where (X, Y) is the position of the active vehicle, P ov (X,Y) represents the safety field force generated by the surrounding vehicles at the position (X,Y) on the active vehicle, X represents the horizontal coordinate of the active vehicle position, and Y represents the vertical coordinate of the active vehicle position; a ov It represents the maximum field force generated by the surrounding vehicles; e is the base of the natural logarithm, and η is the safety field force spatial distribution function; Among them, (X ov ,Y ov ) represents the center of gravity of the surrounding vehicles; X ov Indicates the horizontal coordinate of the center of gravity of the surrounding vehicles, Y ov The vertical coordinate of the center of gravity of the vehicle; ρ x , ρ y Respectively represent the convergence coefficients in the horizontal and vertical directions; the vertical direction refers to the direction along which the vehicle moves, and the horizontal direction refers to the direction perpendicular to the direction in which the vehicle moves; represents the lateral speed of the surrounding vehicles; b represents the vehicle specification coefficient; ξ represents the relative speed of the surrounding vehicles to the active vehicle; c represents the weight factor; Among them, k ov Represents the safety field polarity function.

3. The risk prediction and control method for the merging area of ​​an autonomous driving vehicle according to claim 2, characterized in that: The S1 further comprises: For active vehicles, the safety field force generated by road markings on the active vehicle is calculated using the following formula: Among them, P r (X,Y) represents the safety field force generated by the road marking at position (X,Y), a r represents the maximum field force of the road; d represents the minimum distance from the position (X, Y) to the road marking; d r represents the safety threshold; W represents the active vehicle width; For active vehicles, the total safety field force on the active vehicle is calculated using the following formula: Among them, P c (X,Y) represents the total safety field force received by the active vehicle at position (X,Y); P i ov (X,Y) represents the safety field force generated by the i-th surrounding vehicle on the position (X,Y) of the active vehicle; m is the total number of surrounding vehicles; represents the impact of the jth road marking on the safety field at the active vehicle's position (X, Y); n is the total number of road markings.

4. The risk prediction and control method for the merging area of ​​an autonomous driving vehicle according to claim 1, characterized in that: The S2 includes: S21: Determine the vehicle dynamics model; Where: The lateral acceleration a represents the rate of change of the active vehicle's velocity in the lateral direction. x,δ ; ν x,δ represents the lateral velocity of the active vehicle; ν y,δ represents the velocity of the active vehicle in the longitudinal direction; The longitudinal acceleration a represents the rate of change of the active vehicle's velocity in the longitudinal direction. y,δ ; Indicates the direction angle of the active vehicle; X δ , Y δ Respectively represent the position of the active vehicle in the X and Y directions, represents the position X of the active vehicle δ , Y δ The rate of change over time is the velocity in the horizontal and vertical directions ν x,δ , ν y,δ ; ν δ represents the speed of the active vehicle, δ represents the lane marking; S22: The control equation of the active vehicle is expressed as: Where: T is the transpose, x represents the state vector of the active vehicle; u represents the control input vector of the active vehicle; f[x,u] represents the function based on the vehicle state mapping to vehicle control.

5. The risk prediction and control method for the merging area of ​​an autonomous driving vehicle according to claim 4, characterized in that: The S2 further includes: Based on model predictive control, equation (8) is expressed as follows: x(k+1)=x(k)+ΔT·f[x(k),u(k)] (9) Where: x(k) represents the state vector of the active vehicle at the kth moment; x(k+1) represents the state vector of the active vehicle at k+1, i.e., the predicted state vector; u(k) represents the control input vector of the active vehicle at the kth moment; f[x(k),u(k)] represents the change of the state of the active vehicle under the state vector x(k) and the control input vector u(k); ΔT represents the sampling interval; y(k) represents the output vector of the active vehicle at the kth moment. Based on equation (6), the output vector y(k) is expressed as a parameter based on the safety field force: In the formula, g represents the output function; represents the lateral velocity at time k; represents the velocity in the longitudinal direction at time k, It represents the total safety field force on the vehicle at time k; Based on model predictive control MPC, the vehicle control input vector is predicted: In the formula, y(k+1|k) represents the output vector predicted at time k+1; x(k+1|k) represents the state vector predicted at time k+1, u(k|k) represents the control input vector at the current time k, and N p 、N c Respectively represent the horizontal prediction steps and the vertical prediction steps, N p ≥N c ; The output sequence y(k) and the control input sequence u(k) are: y(k)=[y(k+1|k),y(k+2|k),...y(k+N p |k)] T (12) u(k)=[u(k|k),u(k+1|k),...u(k+N p -1|k)] T (13)。 6. The risk prediction and control method for the merging area of ​​an autonomous driving vehicle according to claim 1 or 4, characterized in that: The S3 includes: Calculate the cost function Y for driving safety ds , the calculation formula is as follows: AND ds =||α|-1|And ds-log +|α|Y ds-lat (14) Where: Y ds-log , Y ds-lat They represent the longitudinal and lateral safety costs of the active vehicle respectively; α represents the lane-changing action function of the vehicle, α∈{-1,0,1}={left lane change, no lane change, right lane change}; Y ds-log =k ν-log l δ / [(Dn y,δ ) 2 +u]+k s-log / [(Δs y,δ ) 2 +u] (15) Y ds-lat =k ν-lat l δ+α / [(Dn x,δ+α ) 2 +u]+k s-lat / [(Δs x,δ+α ) 2 +u] (16) In the formula, κ ν-log , κ ν-lat represents the vertical and horizontal risk weighting coefficients; δ , δ+α Indicates the weight factor related to control in longitudinal and lateral control; Δν y,δ , Δν x,δ+α It represents the speed error in longitudinal and lateral control, that is, the difference between the expected speed and the actual speed; υ represents the smoothing factor; κ s-log , κ s-lat Indicates the risk weighting coefficient related to the vertical and horizontal positions in vertical and horizontal control; Δs y,δ , Δs x,δ+α It represents the position error in longitudinal and lateral control, that is, the difference between the desired position and the actual position; Cost function of driving comfort The calculation formula is as follows: Where: a x,δ , a y,δ They represent the lateral acceleration and longitudinal acceleration of the active vehicle respectively; Respectively represent the weight coefficients of lateral acceleration and longitudinal acceleration; Cost function Y of driving efficiency pe The calculation formula is as follows: Where: y,δ represents the actual longitudinal speed on lane δ; Indicates the target speed; represents the speed limit of lane δ; The total cost function Y calculation formula for vehicle decision-making is as follows: in, They represent the weighting coefficients of the cost function for driving safety, the cost function for driving comfort, and the cost function for driving efficiency respectively.

7. The risk prediction and control method for the merging area of ​​an autonomous driving vehicle according to claim 6, characterized in that: The S4 includes: A game model between the active vehicle and surrounding vehicles is established, and the calculation formula is as follows: In formula (21), represents the longitudinal optimization acceleration of the active vehicle; α * represents the optimized lane-changing decision of the active vehicle; represents the maximum total cost function when the active vehicle plays the game; argmin represents the optimal control of the active vehicle; In formula (22), a y,δ ,α represents the acceleration and lane turning action of the active vehicle; represents the acceleration decision after the active vehicle interacts with the i-th surrounding vehicle; i represents different surrounding vehicles interacting with the active vehicle, lov represents the vehicle in front of the active vehicle, and rov represents the vehicle behind the active vehicle; γ 2 (a y,δ ,α) represents the optimal action of the surrounding vehicles based on the behavior of the active vehicle; ζ represents the decision parameter in the interactive game between the active vehicle and the surrounding vehicles, Φ 2 represents the action space; Y i (a y,δ ,α,ζ) is the cost function of any action taken by the surrounding vehicles based on the active vehicle behavior, is the cost of the actions taken by the surrounding vehicles based on the acceleration decisions of the active vehicle; During the game, the active vehicle should satisfy the following equation: in, represents the maximum longitudinal speed of the active vehicle; represents the minimum longitudinal acceleration and maximum longitudinal acceleration of the active vehicle, ν y,δ Represents the longitudinal velocity of the active vehicle.

8. The risk prediction and control method for the merging area of ​​an autonomous driving vehicle according to claim 5 or 7, characterized in that: The S5 includes: The decision-making process of active vehicles should meet the following requirements: In formula (24), J p (k) is the objective function at time k, Δy(k) represents the output error, i.e., the difference between the expected output and the actual output; ΔY(k) represents the difference between the expected lateral position and the actual lateral position; represents the rate of change of the control input; Q1, Q2, and R represent the weight matrices of the output error, lateral position error, and rate of change of the control input, respectively; Δy T , ΔY T , They are Δy(k), ΔY(k), The transposed matrix of In formula (25), u * (k) represents the optimal control input sequence, argmin represents the optimal control input sequence by minimizing J p (k) optimizes the vehicle control input u'(k); u'(k) represents the optimized control input sequence obtained by the model predictive control MPC and game process, which is obtained by combining equations (13), (21), and (23); A k represents the active vehicle state matrix; B k Represents the active vehicle control matrix; ultimately the optimal vehicle control input sequence is obtained.

Images