A multi-target dynamic weighting full-speed-range vehicle adaptive cruise control method

Abstract

The application is a kind of multi-target dynamic weighting full-speed domain vehicle adaptive cruise control method. The kinematic information of the vehicle and the preceding vehicle is collected; a method for establishing a longitudinal following discrete state space model considering the acceleration disturbance of the preceding vehicle is given, and safety, following performance, economy and comfort control objectives are formulated; based on the objective benefit function of each control objective, the probability distribution of mixed strategy is solved through non-cooperative game and used as the objective weight coefficient; considering the subjective preference of the driver for the vehicle control objective, a subjective weight model is established based on the blind number theory to solve the subjective weight coefficient; a subjective and objective combined weighting model is established based on game theory, the optimal subjective and objective weight is dynamically assigned to the model predictive controller for real-time solving, and the optimal expected acceleration under the subjective and objective is obtained. The application realizes the multi-target weight of the control system, and comprehensively improves the safety, following performance, economy and comfort under the premise of considering the multi-target conflict and the demand of the driver.

Description

TECHNICAL FIELD

[0001] The present application belongs to the field of intelligent driving, and particularly relates to a multi-target dynamic weighting full-speed-range vehicle adaptive cruise control method. BACKGROUND

[0002] With the acceleration of intelligent driving technology, adaptive cruise control system is widely studied. Adaptive cruise control can simplify the driver's operation of the vehicle. When the driver drives the vehicle in a complex traffic scene, the adaptive cruise control system controls the driving and braking of the vehicle, so that the host vehicle and the preceding vehicle maintain a certain distance, and the safety of the driver is ensured. However, the driver has multiple performance requirements for vehicle driving, such as safety, following performance, economy and comfort, that is, the goals of the adaptive cruise control system, and different goals are strongly coupled, which easily causes control conflicts. When too much emphasis is placed on a certain goal, the remaining goals will be reduced. At the same time, the driver's subjective driving demand in the full-speed-range range is time-varying, and the control target weight of the adaptive cruise control system should respond to the driving demand. Therefore, how to balance the relationship between the safety, following performance, economy and comfort of the adaptive cruise control system, comprehensively consider the objective multi-target conflict and the subjective driving demand time-varying factor, dynamically weight the multi-target, and promote the popularization of the automatic cruise control system have great significance.

[0003] At present, some patents have solved the problem of vehicle adaptive cruise control. In the existing published patents, for example, CN202110676661.6 invents a kind of self-adaptive PID automobile driving cruise control method, which improves safety, but does not comprehensively consider the vehicle following performance, economy and comfort of multiple vehicle key performances, and cannot improve the performance of the vehicle while ensuring the safety of the vehicle, which is difficult to meet the driving needs of the driver in different driving scenes. At the same time, improper parameter setting of PID algorithm can easily cause oscillation problem. CN201811366959.1 invents a kind of multi-target adaptive cruise control method, device and equipment, which coordinates multiple control targets. However, the weight of each target is determined through repeated experiments in offline state. In the actual operation process of the adaptive cruise control system, the weight coefficient will no longer change. Since all driving states of the vehicle cannot be exhausted in offline state, it is difficult to determine the weight coefficient of multiple targets depending on limited simulation conditions. At the same time, the driving state of the vehicle changes all the time, and the relative importance of the driver's demand for multiple targets also changes constantly, so fixed weighting will be difficult to adapt to changing vehicle driving state. CN202211646001.4 invents a kind of individualized adaptive cruise system based on deep reinforcement learning and its control method, which meets the demand of different style drivers for the performance of adaptive cruise control system. However, the three control modules designed involve multiple parameters and a large range of search, which is poor in real-time performance. It is difficult to design the multi-target reward function of the adaptive cruise system using reinforcement learning, which has the problem of difficult training. At the same time, due to the limitation of acceleration response and vehicle physical conditions, safety, following performance, economy and comfort often conflict in the process of vehicle driving and braking in complex traffic scenes, and the game relationship of multiple performance targets needs to be considered to give reasonable weight coefficient. Therefore, under the premise of ensuring the safety of vehicle driving, considering the objective multi-target conflict and subjective driver target demand time-varying in the full speed domain, the multi-target weight coefficient of the adaptive cruise control system is dynamically weighted, which has important engineering and application value for improving the comprehensive performance of multiple targets and improving the driving experience of the driver. SUMMARY

[0004] The purpose of the present application is to provide a kind of multi-target dynamic weighting full speed domain vehicle adaptive cruise control method to balance the safety, following performance, economy and comfort of adaptive cruise control multi-target performance, which can consider the demand of the driver, coordinate the contradiction between multiple targets, and realize the dynamic adjustment of the performance of the vehicle when the driver drives the vehicle.

[0005] The technical solution for achieving the object of the application is a multi-target dynamic weighting full-speed-range vehicle adaptive cruise control method, which dynamically adjusts the weight coefficients of multi-targets in the full-speed-range of the adaptive cruise control system by constructing a subjective and objective dynamic optimal weighting model based on game theory, characterized in that it comprises the following steps:

[0006] Step (1): Collecting the kinematic information of the host vehicle and the preceding vehicle through the vehicle-mounted sensors, and obtaining the acceleration of the preceding vehicle and the acceleration change rate of the host vehicle through calculation;

[0007] Step (2): Establishing a longitudinal following discrete state space model considering the disturbance of the acceleration of the preceding vehicle, and formulating safety, following, economy and comfort control objectives;

[0008] Step (3): Establishing a multi-party non-cooperative game model based on the objective payoff function of each control objective, solving the probability distribution of the mixed strategy, and taking it as the objective weight coefficient;

[0009] Step (4): Considering the subjective preference of the driver for the vehicle control objectives in the full-speed-range, establishing a subjective weight model based on the blind number theory, and outputting the subjective weight coefficient;

[0010] Step (5): Establishing a subjective and objective weight vector set by combining the objective weight coefficient and the subjective weight coefficient, establishing a subjective and objective combined weighting model based on game theory, dynamically assigning the optimal subjective and objective weights to the model predictive controller for real-time solving, and obtaining the optimal expected acceleration under the subjective and objective.

[0011] Compared with the prior art, the present application has the following advantages:

[0012] (1) The multi-target dynamic weighting full-speed-range vehicle adaptive cruise control method of the present application formulates safety, following, economy and comfort performance objectives for vehicle driving, takes safety as the hard constraint of the model predictive controller, adopts Min-Max standardization to normalize the following, economy and comfort objectives of the vehicle in complex traffic scenarios, establishes a multi-party non-cooperative game model, solves the probability distribution of the mixed strategy based on the sequential quadratic programming method, and converts it into the current subjective weight coefficient of the control objective; K-means algorithm is used to divide the drivers into aggressive, stable and conservative driving styles, and the subjective weight coefficients of each control objective are calculated based on the blind number theory; a subjective and objective weight vector set is established, and the subjective and objective comprehensive weight coefficients obtained based on the game set model are dynamically assigned to the model predictive controller to obtain the optimal expected acceleration, solving the problem of balancing safety, following, economy and comfort in the prior art.

[0013] (2) The application considers that different target strong coupling easily causes control conflict and that the subjective driving demand of a driver in a full speed range is time-varying, objective multi-target conflict and subjective driving demand time-varying factors are comprehensively considered, the multi-target is dynamically weighted subjectively and objectively, the weight coefficients of the multiple targets can be adaptively adjusted according to the complex driving scene of the vehicle, and the complex traffic scene has good adaptability.

[0014] (3) The application considers the interference of the motion state information of the preceding vehicle on the motion decision of the vehicle, a longitudinal following discrete state space model considering the acceleration disturbance of the preceding vehicle is established, and the accuracy of the adaptive cruise control is improved. BRIEF DESCRIPTION OF DRAWINGS

[0015] Figure 1 The step flow chart of the application.

[0016] Figure 2 The front and rear vehicle relationship diagram described in the application.

[0017] Figure 3 The step flow chart of the objective weight model.

[0018] Figure 4 The step flow chart of the subjective weight model.

[0019] Figure 5 The step flow chart of the subjective and objective combined weighting model. DETAILED DESCRIPTION

[0020] The application will be further described in detail below with reference to the drawings.

[0021] The application is a full speed range vehicle adaptive cruise control method with multi-target dynamic weighting, as shown in the figure, comprising the following steps: Figure 1

[0022] Step 1: Collect the kinematics information of the vehicle and the preceding vehicle through the vehicle-mounted sensor, and obtain the acceleration of the preceding vehicle and the acceleration change rate of the vehicle through calculation;

[0023] Step 2: Establish a longitudinal following discrete state space model considering the acceleration disturbance of the preceding vehicle, and formulate safety, following, economy and comfort control targets;

[0024] Step 3: Based on the objective benefit function of each control target, a multi-party non-cooperative game model is established, the probability distribution of the mixed strategy is solved, and the probability distribution is taken as the objective weight coefficient;

[0025] Step 4: Considering the subjective preference of the driver for the vehicle control target in the full speed range, a subjective weight model is established based on the blind number theory, and the subjective weight coefficient is output;

[0026] ​Step 5: Combine the objective weight coefficient and the subjective weight coefficient to establish the subjective and objective weight vector set, establish the subjective and objective combination weighting model based on game theory, dynamically assign the optimal subjective and objective weight to the model predictive controller for real-time solving, and obtain the optimal expected acceleration under the subjective and objective.

[0027] The "preceding vehicle acceleration and vehicle acceleration change rate" calculation method in step 1 is as follows:

[0028] Step 11, preceding vehicle acceleration calculation:

[0029]

[0030] Where a f is the preceding vehicle acceleration, v f is the preceding vehicle speed, T v is the sampling time of the preceding vehicle speed, and △t v is the sampling period of the preceding vehicle speed.

[0031] Step 12, vehicle acceleration change rate calculation:

[0032]

[0033] Where j r is the vehicle acceleration change rate, a r is the vehicle acceleration, T v is the sampling time of the preceding vehicle speed, and △t v is the sampling period of the preceding vehicle speed.

[0034] The "longitudinal following discrete state space model considering preceding vehicle acceleration disturbance" in step 2 is established as follows:

[0035] Step 21, considering the longitudinal dynamic coupling characteristics of the vehicle and the preceding vehicle, a vehicle longitudinal dynamic model is established.

[0036] The kinematic relationship between the vehicle and the preceding vehicle is:

[0037]

[0038] Where e s is the distance error between the vehicle and the preceding vehicle, d is the actual distance between the vehicle and the preceding vehicle, d des is the expected distance between the vehicle and the preceding vehicle, e v is the speed difference between the vehicle and the preceding vehicle, v r is the vehicle speed.

[0039] Step 22, the vehicle expected acceleration and actual acceleration are sampled with a first-order time delay:

[0040]

[0041] where τ is the time delay coefficient, s is the complex variable, a des is the desired acceleration of the ego vehicle.

[0042] Step 23, select the distance error between the ego vehicle and the preceding vehicle, the ego vehicle speed, the speed difference between the ego vehicle and the preceding vehicle, the ego vehicle acceleration, and the ego vehicle acceleration change rate as state variables, and establish a longitudinal following discrete state space model considering the disturbance of the preceding vehicle acceleration:

[0043]

[0044]

[0045] x(k) = [e s (k),v r (k),e v (k),a r (k),j r (k)] T

[0046] where φ(k) is the system control variable, i.e., the ego vehicle acceleration, is the system disturbance variable, i.e., the preceding vehicle acceleration, and G is the coefficient matrix of the system disturbance variable, and T s is the system sampling period.

[0047] The method for setting the safety, following, economy, and comfort control objectives in step 2 is as follows:

[0048] Step 24, set the safety control objective by comprehensively considering the fixed safety distance and the collision time as:

[0049] d ≥ d max = {t TTC e v ,d s0}

[0050] where t TTC is the collision time, and d s0 is the minimum following safety distance.

[0051] Since safety is the most important control objective of adaptive cruise control, it is set as a hard constraint.

[0052] Step 25, for the driver, it is an important requirement that the adaptive cruise control system can stably follow the preceding vehicle. Following includes the actual two-vehicle distance following the expected two-vehicle distance and the ego vehicle speed following the speed of the preceding vehicle. The two-norm of the error between the actual two-vehicle distance and the expected two-vehicle distance and the error between the ego vehicle speed and the preceding vehicle speed is weighted and summed to obtain the following control objective:

[0053] JF = w s e s 2 + w v e v 2

[0054] wherein J F is a following vehicle target, w s is a weight coefficient of actual two-vehicle distance and error of expected two-vehicle distance, w v is a weight coefficient of vehicle speed and error of front vehicle speed.

[0055] Step 26, the economy, i.e., the energy consumption target is minimum, and the energy consumption is related to the acceleration of the vehicle, and the economy control target is as follows:

[0056]

[0057] wherein J E is an economy target, is a weight coefficient of vehicle acceleration.

[0058] Step 27, the rate of change of vehicle acceleration can be weighted after taking its two-norm, which is the comfort control target:

[0059]

[0060] wherein J C is a comfort target, is a weight coefficient of rate of change of vehicle acceleration.

[0061] Step 28, since the following vehicle, economy and comfort control targets of the vehicle have different dimensions and dimension units, Min-Max standardization is adopted, and the original control target data is linearly transformed, so that the result value is mapped between [0, 1]. The normalized following vehicle control target J F * , the economy control target J E * and the comfort control target J C * are obtained, and the conversion function is as follows:

[0062]

[0063] wherein J max is the maximum value of the control target sample data, and J min is the minimum value of the control target sample data.

[0064] The method for establishing the multi-party non-cooperative game model in the step 3, as shown in Figure 3 , comprises the following steps:

[0065] Step 31, the weights of the three of car-following, economy and comfort can be converted into a finite player non-cooperative game model to solve the problem.

[0066] The finite player non-cooperative game is represented by a tuple:

[0067] Γ=(N,{S i} i∈N ,{u i} i∈N )

[0068] Where N is the finite set of players, i.e. the three players of car-following, economy and comfort; S i is the pure strategy space of player i, i.e. the weight coefficients of safety, economy and comfort at the current sampling time; u i is the reward function of player i, i.e. the revenue that player i can get by taking the weight coefficient at the current sampling time.

[0069] Step 32, the optimal solution of the revenue of player i is recorded as:

[0070] minβ i -u i (σ)

[0071]

[0072]

[0073]

[0074] Where the mixed strategy distribution σ * is called the Nash equilibrium of the game Γ:

[0075]

[0076] Where (σ -i ,s i j ) represents the mixed strategy combination of player i using its jth pure strategy, i.e. the jth pure strategy of the ith player is assigned a mixed strategy with probability 1. For the three players of car-following, economy and comfort, it is impossible to obtain better returns than in the Nash equilibrium by only changing their own mixed strategy without changing the strategies of others. The optimization of each player means maximizing its revenue when other players play according to the Nash equilibrium strategy.

[0077] Step 33, according to the existence theorem of Nash equilibrium, there exists:

[0078]

[0079] where σ * is the Nash equilibrium of the game Γ, β i* is the optimal expected payoff of player i.

[0080] The method for solving the probability distribution of mixed strategies in step 3 comprises the following steps:

[0081] Step 34, the problem of calculating the Nash equilibrium of Γ can be simplified into a problem of solving an optimization problem M with the optimal value of zero. In order to solve the problem M, it is re-expressed as a vector in an m+z-dimensional Euclidean space. Take y as a vector with a length of m+z, and arrange the strategies of players 1 to z in order, with a total of m strategies.

[0082] Transform the variables in M, and the optimization problem M is converted into the following form:

[0083] min f(y)

[0084] s.t.g(y)≤0

[0085] h(y)=0

[0086] y i ≥0

[0087] where:

[0088] f(y)=∑(β i -u i (σ))

[0089]

[0090]

[0091] Step 35, to solve this nonlinear minimization problem with nonlinear constraints, a quasi-Newton method based on sequential quadratic programming (SQP) is used.

[0092] The method for establishing the subjective weight model in step 4, as shown in Figure 4 comprises the following steps:

[0093] Step 41, the K-means algorithm is used to divide the drivers into aggressive, stable, and conservative driving styles.

[0094] Step 42, the credibility of a driver is represented by , and the credibility of a group of drivers P1, P2, …, P n is respectively

[0095] The credibility of a driver P i about the group of drivers P1, P2, …, Pn The overall credibility is:

[0096]

[0097] Driver groups P1, P2, ..., P n The overall credibility is:

[0098]

[0099] Driver P i The absolute overall credibility is:

[0100] θ=αα i

[0101] Step 43: Provide the evaluation range for the importance of the control objective. The value range for the importance of the control objective is [0,1]. Define the control objective as "very important" as [0.8,1], "important" as [0.6,0.8], "moderate" as [0.3,0.6], and "not important" as [0,0.3].

[0102] Step 44: Drivers' demands for economy, safety, and comfort change under complex traffic scenario information. The importance of each control objective to the driver often does not fall at a certain point γ. i It falls on γ, rather than on top. i Nearby and with a confidence level of α i Within the interval, the importance coefficients given by each driver to the control objective C are [a1,b1], [a2,b2], ..., [a...]. n ,b n The objective C is expressed in blind number form as follows:

[0103]

[0104] The mean of the blinded numbering objective f(γ) is:

[0105]

[0106] The variance of the blinded numbering objective f(γ) is:

[0107]

[0108] The standard deviation of the blinded target f(γ) is:

[0109]

[0110] The reliability of the target importance is:

[0111] R f(γ) =1-σ f(γ) / Ef(γ)

[0112] where R f(γ) is the target importance reliability, σ f(γ) is the target importance standard deviation, E f(γ) is the target importance mean.

[0113] Step 45, if the target importance reliability is greater than the pre-specified control value 0.85, the reliability test is passed; otherwise, the first round of evaluation results are fed back to each driver, and the driver is required to correct the original target importance evaluation interval value, and then the relevant calculation and test are performed again until the reliability test is passed.

[0114] Step 46, after each target importance reliability is passed, the subjective weight of each target can be calculated by using the formula:

[0115]

[0116] The "subjective and objective combined weighting model" in step 5 is established as shown in the following steps: Figure 5

[0117] Step 51, an objective and subjective weight vector set is established

[0118] where ω1 is the objective weight, and ω2 is the subjective weight.

[0119] The linear combination of the objective and subjective weight vectors is:

[0120] W = λ1ω1 T + λ2ω2 T

[0121] Step 52, based on the game aggregation model, the target function is established with the minimum deviation as the objective:

[0122]

[0123] which is equivalent to the linear equation group of the first-order derivative optimization:

[0124]

[0125] The combined coefficients λ1 and λ2 after optimization are calculated and normalized:

[0126]

[0127] Step 53, finally, the comprehensive weight of the multi-control target is given:

[0128] W * = λ1 * ω1 T ​+ λ2 * ω2 T

Claims

1. A multi-objective dynamic weighting full-speed-range vehicle adaptive cruise control method, characterized in that, The application discloses a method for dynamically adjusting the weight coefficient of a multi-objective in a full-speed domain of an adaptive cruise control system based on a game theory-based subjective and objective dynamic optimal weighting model. Step (1): collecting kinematic information of a host vehicle and a preceding vehicle through a vehicle-mounted sensor, and obtaining the acceleration of the preceding vehicle and the acceleration change rate of the host vehicle through calculation; Step (2): establishing a longitudinal following discrete state space model considering the acceleration interference of the preceding vehicle, and formulating safety, following, economy and comfort control objectives; Step (3): establishing a multi-party non-cooperative game model based on an objective payoff function of each control objective, and solving the probability distribution of a mixed strategy, which is taken as an objective weight coefficient; Step (4): considering the subjective preference of a driver for the vehicle control objectives in the full-speed domain, establishing a subjective weight model based on a blind number theory, and outputting a subjective weight coefficient; Step (5): establishing a subjective and objective weight vector set by combining the objective weight coefficient and the subjective weight coefficient, establishing a subjective and objective dynamic optimal weighting model based on the game theory, dynamically assigning an optimal subjective and objective weight to a model predictive controller for real-time solving, and obtaining an optimal expected acceleration under the subjective and objective.

2. The method of claim 1, wherein, In step 1, the "obtaining the acceleration of the preceding vehicle and the acceleration change rate of the host vehicle through calculation" is specifically as follows: Step (11): calculating the acceleration of the preceding vehicle: Wherein, a f is the front vehicle acceleration, v f is the front vehicle speed, T v is the sampling time of the front vehicle speed, △t v is the sampling period of the front vehicle speed; Step (12): calculating the acceleration change rate of the host vehicle: Wherein, j r is the acceleration rate of the host vehicle, a r is the acceleration of the host vehicle, T a is the sampling time of the acceleration of the host vehicle, △t a is the sampling period of the acceleration of the host vehicle.

3. The method of claim 2, wherein, In step 2, the "establishing a longitudinal following discrete state space model considering the acceleration interference of the preceding vehicle" is specifically as follows: Step (21): establishing a longitudinal dynamics model between the host vehicle and the preceding vehicle by comprehensively considering the longitudinal dynamics coupling characteristic relationship between the host vehicle and the preceding vehicle; The kinematic relationship between the host vehicle and the preceding vehicle is as follows: where e s is the error in the distance between the host vehicle and the preceding vehicle, d is the actual distance between the host vehicle and the preceding vehicle, d des is the desired distance between the host vehicle and the preceding vehicle, e v is the difference between the speed of the host vehicle and the speed of the preceding vehicle, v r is the speed of the host vehicle. Step (22): taking a first-order time delay to represent the expected acceleration and the actual acceleration of the host vehicle: where τ is a time delay coefficient, s is a complex variable, a des is the desired acceleration of the vehicle; Step (23): selecting the distance error between the host vehicle and the preceding vehicle, the speed of the host vehicle, the speed difference between the host vehicle and the preceding vehicle, the acceleration of the host vehicle and the acceleration change rate of the host vehicle as state variables, and establishing a longitudinal following discrete state space model considering the acceleration interference of the preceding vehicle: x(k) = [e s (k),v r (k),e v (k),a r (k),j r (k)] T Wherein, φ(k) is the system control variable, i.e. the acceleration of the vehicle, is the system disturbance variable, i.e. the acceleration of the preceding vehicle, G matrix is the coefficient matrix of the system disturbance variable, T s is the system sampling period.

4. The method of claim 3, wherein, In step 2, the "formulating safety, following, economy and comfort control objectives" is specifically as follows: Step (24): setting the safety control objective as follows by comprehensively considering a fixed safety distance and a collision time: d ≥ d max = {t TTC e v , d s0} where t is the time to collision, d is the minimum following distance, and v is the vehicle speed. TTC s0 is the time to collision, d is the minimum following distance, and v is the vehicle speed.​ Since safety is the most important control objective of the adaptive cruise control, the safety is set as a hard constraint; Step (25): taking the two-norm of the error between the actual distance between the two vehicles and the expected distance between the two vehicles and the error between the speed of the host vehicle and the speed of the preceding vehicle, and then performing weighted summation to obtain the following control objective: J F = w s e s 2 + w v e v 2 wherein J F is the car-following performance target, w s is the weight coefficient of the actual two-vehicle distance and the error of the expected two-vehicle distance, w v is the weight coefficient of the host vehicle speed and the error of the front vehicle speed. Step (26): the economy control objective is as follows: wherein J E is an economic objective, is a weight coefficient for the acceleration of the vehicle; Step (27): taking the two-norm of the acceleration change rate of the vehicle, and then performing weighting to obtain the comfort control objective: wherein J C is a comfort target, is a weight coefficient of the vehicle acceleration rate of change; Step (28): Since the car-following, economy and comfort control objectives have different dimensions and dimension units, Min-Max standardization is adopted, and the original control objective data is linearly transformed so that the result value is mapped between [0, 1]; the normalized car-following control objective J F * , the normalized economy control objective J E * and the normalized comfort control objective J C * are obtained. The conversion function is as follows: where J max is a maximum value of the control target sample data, J min is a minimum value of the control target sample data.

5. The method of claim 4, wherein, In step 3, the "establishing a multi-party non-cooperative game model" is specifically as follows: Step (31): assigning weights to the following, economy and comfort, and converting the weights into a finite player non-cooperative game model to solve the problem; The finite player non-cooperative game is represented by a tuple: Γ = (N, {S i} i∈N ,{u i} i∈N ) where N is a finite set of players, i.e. the three players of car-following, economy and comfort; S i is the pure strategy space of player i, i.e. the weight coefficients of safety, economy and comfort at the current sampling time; u i is the payoff function of player i, i.e. the revenue that player i can get by taking the weight coefficients at the current sampling time; Step (32): recording the optimal solution of the payoff of the player i as follows: minβ i -u i (σ) where the mixed strategy distribution σ * is called the Nash equilibrium of the game Γ: where (σ -i ,s i j ) denotes the mixed strategy combination of player i using its jth pure strategy, i.e. the jth pure strategy of the ith player is assigned the mixed strategy with probability 1. Step (33): according to the existence theorem of the Nash equilibrium, there is: where σ * is the Nash equilibrium of the game Γ, β i* is the optimal expected payoff of player i.

6. The method of claim 5, wherein, In step 3, the "solving the probability distribution of the mixed strategy" is specifically as follows: Step (34): Simplify the problem of calculating the Nash equilibrium of Γ into the problem of solving an optimization problem M with the optimal value of zero; in order to solve problem M, restate it as a vector in an m+z-dimensional Euclidean space, take y as a vector with a length of m+z, and arrange the strategies of players 1 to z in order, a total of m strategies; transform the variables in M, and the optimization problem M is converted into the following form: min f(y) s.t.g(y)≤0 h(y)=0 y i ≥0 Where: f(y) = ∑(β i - u i (σ)) Step (36): To solve this nonlinear minimization problem with nonlinear constraints, a quasi-Newton method based on sequential quadratic programming is used.

7. The method of claim 6, wherein, Step 4 "Establishing a subjective weight model based on blind number theory" is as follows: Step (41): Use the K-means algorithm to divide the drivers into aggressive, stable, and conservative driving styles; Step (42): The credibility of the driver is used to indicate that a set of drivers P1, P2, …, P n The credibility of each driver is respectively Driver P i The overall credibility of the group of drivers P1, P2,..., P n is: Driver group P1, P2,..., P n The overall credibility of P is: Driver P i The absolute overall credibility of P is: θ = αα i Step (43): Give the evaluation interval of the importance of the control target; the value interval of the importance of the control target is [0, 1], define "very important" as [0.8, 1], "important" as [0.6, 0.8], "general" as [0.3, 0.6], and "not important" as [0, 0.3]; Step (44): the change of driver's demand for economy, safety and comfort is made under the complex traffic scene information, and the importance degree of driver to each control target often does not fall on a certain point γ i , but falls in an interval near γ i with a reliability of α i , and the importance degree coefficients given by each driver to the control target C are [a1, b1], [a2, b2], …, [a n , b n ], and the target C is expressed in the form of blind number as: The mean of the blind number target f(γ) is: The variance of the blind number target f(γ) is: The standard deviation of the blind number target f(γ) is: The reliability of the target importance is: R f(γ) = 1 - σ f(γ) / E f(γ) wherein R f(γ) is the target importance reliability, σ f(γ) is the target importance standard deviation, E f(γ) is the target importance mean; Step (45): If the target reliability is greater than the pre-specified control value of 0.85, the reliability test passes; otherwise, the first round of evaluation results is fed back to each driver, and the driver is required to modify the original target importance evaluation interval value, and then the relevant calculations and tests are performed again until the reliability test passes; Step (46): After each target importance reliability passes the test, the subjective weight of each target is calculated using the following formula:

8. The method of claim 7, wherein, Step 5 "Establishing a subjective and objective combined weighting model based on game theory" is as follows: Step (51): Establishing the set of subjective and objective weight vectors Where ω1 is the objective weight and ω2 is the subjective weight; The linear combination of the subjective and objective weight vectors is: W = λ1ω1 T + λ2ω2 T Step (52): Based on the game aggregation model, a target function is established with the goal of minimizing deviation: Equivalent conversion to a linear system of first-order derivatives of the optimization: Calculate the optimized combination coefficients λ1 and λ2, and perform normalization processing: Step (53): Finally, give the comprehensive weight of multiple control targets: W * = λ1 * ω1 T + λ2 * ω2 T .

Images