A multi-vehicle collaborative decision-making method in a human-machine mixed driving environment

Abstract

The present invention relates to a multi-vehicle collaborative decision-making method in a human-machine mixed driving environment, including obtaining vehicle attributes and vehicle status near the conflict area, updating the game vehicle list based on the game range, switching the corresponding model to solve the strategy according to the type and number of vehicles in the list; respectively inferring the action strategy of HV through the CAV and HV interaction model, the CAV collaborative decision model, and the human-machine mixed driving interaction and collaborative model coupled between the two to finally solve the CAV optimal strategy. Compared with the prior art, this method simultaneously considers the collaboration between CAVs and the interaction between CAVs and HVs, and adapts to the uncertainty in the dynamic interaction process of human driving vehicles through real-time adjustment of optimal parameters, so as to achieve efficient and safe passage of vehicles in mixed driving scenarios; in addition, this method can also adapt to the personalized driving needs of CAVs.

Description

Technical Field

[0001] The present invention relates to the field of autonomous driving, and in particular to a multi-vehicle collaborative decision-making method in a human-machine mixed driving environment. Background Art

[0002] With the rapid development of autonomous vehicles, human-driving vehicles (HV) and connected and automated vehicles (CAV) will share road resources in the form of mixed human-machine traffic flow for a long time in the future. How to ensure safety and improve efficiency in mixed human-machine driving scenarios is an important topic in future autonomous driving research. Single-vehicle intelligent autonomous driving has environmental perception limitations in complex interactive environments, making it difficult to ensure interactive safety; it often exhibits relatively conservative interactive behaviors, making it difficult to ensure interactive efficiency; it is difficult to make the best decision when facing complex scenarios, and there are practical problems such as high equipment costs. With the increase in CAV penetration, it is possible to consider using vehicle-to-vehicle collaboration to resolve driving conflicts in complex environments and improve safety and efficiency.

[0003] However, most of the current research on vehicle-to-vehicle collaboration focuses on vehicle collaboration in a pure CAV environment or on the interaction between CAV and HV from the perspective of single-vehicle intelligence. There is a lack of comprehensive consideration of the interaction between CAV and HV in mixed-driving scenarios and the collaboration between CAVs. Most current studies only treat HVs as dynamic obstacles or simply express them using a following model. The dynamic interaction characteristics of HVs are not effectively captured, and the interactivity between CAVs is not well reflected. In addition, existing studies have not considered the differences in individual needs between collaborative CAVs, and the resulting models are difficult to meet the personalized driving needs of CAVs. Summary of the invention

[0004] The purpose of the present invention is to provide a multi-vehicle collaborative decision-making method in a human-machine mixed driving environment in order to overcome the defects of the above-mentioned prior art. It is a collaborative decision-making method that can realize CAV workshop collaboration and CAV and HV interaction coupling. In the specific model method design, the personalized driving preference of CAV is realized through the benefit function, and the twin game method is used to realize the dynamic adjustment of the model according to the uncertainty of HV interaction, and then the model is solved to obtain the optimal vehicle action decision result.

[0005] The purpose of the present invention can be achieved by the following technical solutions:

[0006] A multi-vehicle collaborative decision-making method in a human-machine mixed driving environment generally involves constructing three decision-making models: a CAV and HV interaction model, a CAV collaborative decision-making model, and a human-machine mixed driving interaction and collaboration model; generally involves the following steps:

[0007] S1 Vehicle attribute and status information update step: input attribute information of all vehicles in the scene, including vehicle type, kinematic parameters, etc.; vehicle status information, including vehicle position, vehicle speed, acceleration, etc.

[0008] S2 Game object list update and model switching steps: According to the vehicle status, determine the vehicles entering the near conflict area and the conflict area, and update the interactive and cooperative vehicles included in the game object list; determine the actual execution model according to the type and number of game objects.

[0009] S3 CAV and HV interaction modeling steps: Establish a model of CAV and HV interaction.

[0010] S4 CAV collaborative decision-making modeling steps: Establish a model of collaborative decision-making within the CAV vehicle.

[0011] S5 Human-machine mixed driving interaction collaborative modeling steps: Couple the above CAV and HV interaction model and CAV collaborative decision-making model to carry out human-machine mixed driving interaction collaborative model, and solve the model to predict the actions of HV in the game object list and solve the optimal action strategy of CAV, that is, acceleration.

[0012] S6 vehicle status update step: CAV implements vehicle status update through kinematic formulas based on the output acceleration decision results, and solves the updated speed and position; the acceleration, speed, and position of the HV are re-collected from the environment.

[0013] Compared with the prior art, the present invention has the following advantages:

[0014] (1) The present invention proposes to simultaneously consider the coordination between networked autonomous driving vehicles and the interaction between networked autonomous driving vehicles and human-driven vehicles in mixed-driving scenarios, which is more in line with real mixed-driving scenarios and can effectively solve the safety and efficiency problems existing in the autonomous driving of a single vehicle.

[0015] (2) The present invention fully considers the uncertainty in the dynamic interaction process of human-driven vehicles by identifying and dynamically adjusting the weight-related parameters of the component weights of the human-driven vehicle benefit function.

[0016] (3) The collaborative decision-making model for connected autonomous vehicles based on cooperative game proposed in the present invention can also adapt to the personalized driving needs of connected autonomous vehicles. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 Flow chart of the method of the present invention

[0018] Figure 2 The algorithm flow chart of S3 of the present invention is

[0019] Figure 3 The algorithm flow chart of S4 of the present invention is

[0020] Figure 4 Schematic diagram of the data flow relationship between the key steps in the algorithm of the present invention

[0021] Figure 5 Example scenario DETAILED DESCRIPTION

[0022] Definition of conflict zone and near-conflict zone:

[0023] The conflict area is the area containing conflict points caused by the intersection of vehicle trajectories, such as the interior of intersections on urban roads and highways;

[0024] The present invention defines the near conflict area as a distance before the vehicle reaches the conflict area, and the present invention uses the parking sight distance length before the stop line as the calculation range of the near conflict area. The present invention takes the near conflict area and the vehicles inside the conflict area into consideration.

[0025] Stopping sight distance is a professional term that refers to the shortest driving distance required for a vehicle to brake and stop when it encounters an obstacle in front of it in the same lane.

[0026] The technical solution of the present invention is applicable to urban scenarios, highway scenarios, etc. The method of the present invention detects and collects vehicle status information in near-conflict areas and conflict areas through networked roadside equipment, and performs strategy calculation and decision information release. It is technically feasible.

[0027] The technical solution of the present invention is further introduced below in conjunction with the accompanying drawings and embodiments.

[0028] Example

[0029] A multi-vehicle collaborative decision-making method in a human-machine mixed driving environment generally involves constructing three decision-making models: a CAV and HV interaction model, a CAV collaborative decision-making model, and a human-machine mixed driving interaction and collaboration model; generally involves the following steps:

[0030] S1 Vehicle attribute and status information update step: input attribute information of all vehicles in the scene, including vehicle type, kinematic parameters, etc.; vehicle status information, including vehicle position, vehicle speed, acceleration, etc.

[0031] S2 Game object list update and model switching steps: According to the vehicle status, determine the vehicles entering the near conflict area and the conflict area, and update the interactive and cooperative vehicles included in the game object list; determine the actual execution model according to the type and number of game objects.

[0032] S3 CAV and HV interaction modeling steps: Establish a model of CAV and HV interaction.

[0033] S4 CAV collaborative decision-making modeling steps: Establish a model of collaborative decision-making within the CAV vehicle.

[0034] S5 Human-machine mixed driving interaction collaborative modeling steps: Couple the above CAV and HV interaction model and CAV collaborative decision-making model to carry out human-machine mixed driving interaction collaborative model, and solve the model to predict the actions of HV in the game object list and solve the optimal action strategy of CAV, that is, acceleration.

[0035] S6 vehicle status update step: CAV implements vehicle status update through kinematic formulas based on the output acceleration decision results, and solves the updated speed and position; the acceleration, speed, and position of the HV are re-collected from the environment.

[0036] Furthermore, in S1, the vehicle attribute and status information updating step is specifically as follows:

[0037] The vehicle type information of each vehicle is obtained from the environment, that is, whether the vehicle is a CAV or HV; kinematic related parameters, such as maximum acceleration, maximum deceleration, maximum speed, etc.

[0038] At each time step, the acceleration, speed, and position of the HV in the environment are collected and updated; the CAV updates the vehicle speed and position according to the kinematic formula based on the optimal acceleration output by the model. The vehicle state information is used as the input of the model.

[0039] Furthermore, in S2, the game object list update and model switching steps are specifically as follows:

[0040] All vehicles in the scene S = {S 1 ,S 2 ,…,S k ,…,S K}, which contains I CAVs, represented by M = {M 1 ,M 2 ,…,M i ,…,M I} and J HV, expressed as N = {N 1 ,N 2 ,…,N j ,…,N J}.

[0041] Define the game range, that is, the near conflict area and the conflict area. Vehicles entering the game range are added to the game object list; vehicles exit the game object list after passing the conflict area. Vehicles in the game object list There are P CAVs, which are expressed as Q HVs, expressed as

[0042] According to the types of vehicles and the number of each type of vehicles in the game object list, the model is selected and switched.

[0043] When there is no CAV vehicle or only one vehicle in the game object list, the vehicle can pass freely without considering interactive coordination;

[0044] When all vehicles in the game object list are CAV vehicles, only CAV collaborative decision-making is performed, and the CAV collaborative decision-making model is selected to solve the optimal acceleration strategy of the vehicle;

[0045] When the vehicles in the game object list include both CAV and HV, and there is only one CAV, only CAV and HV need to make interactive decisions, and the CAV and HV interactive model is selected to solve the optimal acceleration strategy of CAV;

[0046] When the vehicles in the game object list include both CAV and HV, and there are multiple CAVs, it is necessary to couple the CAV collaborative decision-making model and the CAV and HV interactive decision-making model to realize human-machine mixed driving interactive collaboration.

[0047] S3 CAV and HV interaction modeling steps: Establish a model for the interaction between CAV and HV. It can be used as an interaction model for a single CAV and HV, or as an outer layer model for S5 human-machine mixed driving interaction and collaborative modeling, taking CAV as a whole to achieve the interaction modeling between HV and multiple CAVs. On the basis of solving the HV action through model prediction, the overall benefit of CAV itself is optimized.

[0048] Furthermore, in S3, the interaction model in the CAV and HV interaction modeling step is specifically:

[0049] Based on game theory, a non-cooperative model is constructed to divide CAV and HV into leaders and followers, respectively.

[0050] Considering the safety, efficiency and comfort requirements during vehicle operation, the profit function f of CAV is designed. p and the profit function f of HV q The profit function forms of the two are consistent, and the individual profit value of a vehicle is the weighted sum of the three.

[0051]

[0052]

[0053] In the formula, is the safety index of CAV, is the efficiency index of CAV, It is the CAV comfort index. are the weight values ​​of the three indicators respectively. They are the safety, efficiency and comfort indicators of HV. are their respective weight values. The specific indicator calculation is shown in the following formula.

[0054]

[0055]

[0056]

[0057]

[0058]

[0059]

[0060] Where, ΔL p is the distance between CAV vehicle p and the collision point with HV vehicle q, v p (k+1) is the speed of CAV vehicle p at the k+1th moment, a p (k), a p (k+1) is the acceleration of CAV vehicle p at the kth and k+1th moments respectively. ΔL q is the distance between the HV vehicle q and the conflict point between the CAV vehicle p, v q (k+1) is the speed of HV vehicle q at the k+1th moment, a q (k), a q (k+1) are the accelerations of the HV vehicle q at the kth and k+1th moments, respectively.

[0061] The safety and efficiency weights of HV are updated in real time based on the weights of the previous moment and according to the similarity between the predicted acceleration of CAV and its actual acceleration. In order to reduce the complexity of model parameter updating, the weights corresponding to the safety and efficiency indicators are converted into the form of functions of the same parameter θ. As shown in the following formula:

[0062]

[0063] In the formula, A and B are constants.

[0064] Establish an interaction model, and CAV predicts the optimal acceleration that HV may take when it makes a certain acceleration action decision based on the benefit function of HV. On this basis, the self-benefit is calculated and the CAV’s own acceleration strategy space is Search for the optimal solution that maximizes its own benefits As shown below:

[0065]

[0066] st

[0067]

[0068]

[0069]

[0070]

[0071]

[0072] In the formula, is the list of CAV objects participating in the game M g The acceleration strategy space of CAV acceleration strategy space can be adopted Take the strategy space for HV at this time middle The corresponding HV income, In order to make Maximum optimal acceleration strategy; For when CAV takes When HV adopts the optimal acceleration strategy The corresponding CAV overall benefit is The optimal acceleration strategy for CAV that maximizes the benefit. for The minimum and maximum values ​​of for The minimum and maximum values ​​of is a safety constraint, where is the time it takes for the CAV vehicle p to reach the conflict point according to the current state, is the time it takes for the HV vehicle q to reach the conflict point according to the current state. The optimal acceleration strategy of HV obtained according to the model As a model in The prediction of HV is denoted as Used for subsequent twin games.

[0073] In order to more accurately capture the real-time interactive uncertainty of HV actions, the twin game method is used to update the HV weight-related parameters θ.

[0074] At the initial moment, the weight values ​​corresponding to the three CAV indicators are obtained based on the natural driving data calibration , and CAV fixed parameter values , and the corresponding HV initial weight value related parameters are θ0 . In θ 0 Within the definition domain, r (r is set to an odd number) values ​​are discretely taken to form the initial θ value space Θ 0 .

[0075]

[0076] In the formula, the value space Θ 0 The specific method for selecting the values ​​of each θ is as follows: At the initial moment, in order to select the values ​​as comprehensively as possible, the range of the value space is determined as The value space except the minimum value Optimal parameters at the initial moment Maximum Outside, in [0,θ 0 ]、 Take (r-3) / 2 values ​​at equal distances and add them to the value space to form a value space Θ with r elements. 0 .

[0077] Twin game solution: At each subsequent time step, based on the value space Θ of the previous moment k For each θ value in , the CAV and HV interaction model is used to solve the HV to obtain a series of HV optimal acceleration strategies corresponding to the θ value.

[0078] Fitness calculation optimal parameters OK: Calculate Θ k The optimal acceleration strategy of HV for each θ value in With actual The degree of difference is shown in the following formula.

[0079]

[0080] α is defined as fitness. The larger the α value, the smaller the difference between the calculated value and the true value, and the higher the fitness. The value with the highest fitness α among all the θ values ​​is taken as the optimal parameter for the next moment.

[0081] Value list Θ k+1 Update: According to the optimal parameters at the next moment And its corresponding fitness size α * , confirm the number of values ​​r at the next moment. In this embodiment, the rules are formulated as shown in Table 1.

[0082] Table 1 Correspondence between the number of twin values ​​and fitness

[0083]

[0084] When the fitness reaches a certain level, the optimal Perform model calculations and stop playing the twin game until the fitness exceeds the level where twins are no longer needed.

[0085] Set the value interval length to Minimum value of the value limit Maximum The calculation method is shown in the following formula.

[0086]

[0087]

[0088] If calculated Then Accordingly

[0089] If calculated Then Accordingly

[0090] Finally, the value space Θ of θ at the next moment k+1 exist The nearby discrete values ​​are obtained as shown in the following formula.

[0091]

[0092] S4 CAV collaborative decision-making modeling steps: Establish a CAV vehicle internal collaborative decision-making model. It can be used as a vehicle collaborative decision-making model in a pure CAV scenario, and can also be used as an inner model for human-machine mixed driving interactive collaborative modeling to solve CAV individual acceleration.

[0093] Furthermore, in S4, the collaborative decision model in the CAV collaborative decision modeling step is specifically:

[0094] Considering the requirements of safety, efficiency and comfort during vehicle operation, the characteristic function of CAV is designed, and the individual benefit value of the vehicle is the weighted sum of the three. The calculation method of each indicator is consistent with the CAV and HV interaction model.

[0095]

[0096]

[0097]

[0098]

[0099] In the formula, p′ is the number of M in the game object list. g Other CAV vehicles except CAV vehicle p, i.e. Calculate the safety inside the CAV. The weight value is Can be adjusted according to the CAV's individual preferences.

[0100] According to the basic principles of cooperative game, the characteristic function, optimization objectives and constraints are designed, the benefits are distributed according to the Shapley principle, and the optimal actions of the vehicles are planned. The Shapley value calculation is shown in the following formula:

[0101]

[0102] In the formula, ψ p is the Shapley value of CAV vehicle p. s is M g The subset of represents the set of possible cooperative game objects of participants in the calculation process. v(s) is the profit generated by the participant set s. |s|, |M g | respectively represent s, M g The number of vehicles included.

[0103] v(s) is the revenue generated by the participant set for s, which is calculated as:

[0104]

[0105] The model is established with the goal of maximizing the overall total benefit of CAV vehicles. It is shown in the following formula:

[0106]

[0107] st

[0108]

[0109]

[0110] In the formula, is the safety constraint between CAVs.

[0111] Further, in S5, the interactive decision model in the human-machine mixed driving interactive decision modeling step is specifically:

[0112] Considering the interaction between CAV and HV and the coordination within CAV, a two-layer optimization model is constructed. The outer layer is the interaction between CAV as a whole and HV as shown in the following formula:

[0113]

[0114] st

[0115]

[0116]

[0117]

[0118]

[0119]

[0120]

[0121]

[0122] In the formula, is the overall benefit of CAV, is the sum of the benefits of CAVs in the game object list. Safety constraint considers the safety between CAV and HV. The other quantities have the same meaning as the CAV and HV interaction model.

[0123] The inner layer is the internal coordination of CAV, and the optimal acceleration strategy of CAV vehicles is obtained by solving the CAV collaborative decision model. However, in the model, when calculating the safety index, all CAVs and HVs in the game object list except itself need to be considered.

[0124] Furthermore, in S6, after selecting an appropriate model according to the game object list to solve the optimal acceleration strategy, the vehicle state is updated.

[0125] The state of the CAV is expressed and updated according to the kinematic formula:

[0126]

[0127]

[0128] In the formula, To participate in the game CAV vehicle speed, acceleration, and position.

[0129] The real status of HV is re-collected from the environment to obtain its actual position speed Acceleration

[0130] Example scenario Figure 5 shown.

[0131] The above description is only a description of the preferred embodiments of the present application, and is not intended to limit the scope of the present application. Any changes or modifications made by any person skilled in the art based on the above disclosed technical contents shall be deemed as equivalent effective embodiments and shall fall within the scope of protection of the technical solution of the present application.

Claims

1. A multi-vehicle collaborative decision-making method in a human-machine mixed driving environment, It is characterized in that The overall plan involves building three decision-making models: CAV and HV interaction model, CAV collaborative decision-making model, and human-machine hybrid driving interaction and collaboration model. The overall plan includes the following steps: S1 Vehicle attribute and status information update step: input attribute information of all vehicles in the scene, including vehicle type and kinematic parameters; vehicle status information, including vehicle position, vehicle speed, and acceleration; S2 Game object list update and model switching step: According to the vehicle status, determine the vehicles entering the near conflict area and the conflict area, and update the interactive and cooperative vehicles included in the game object list; according to the type and number of game objects, determine the actual execution model; S3 CAV and HV interaction modeling steps: Establish a model of CAV and HV interaction; S4 CAV collaborative decision-making modeling steps: Establish a model of collaborative decision-making within the CAV vehicle; S5 Human-machine mixed driving interaction collaborative modeling steps: Couple the above CAV and HV interaction model and CAV collaborative decision model to conduct a human-machine mixed driving interaction collaborative model, solve the model, predict the actions of the HV in the game object list, and solve the optimal action strategy of the CAV, that is, acceleration; S6 vehicle status update step: CAV implements vehicle status update through kinematic formulas based on the output acceleration decision result, and solves the updated speed and position; the acceleration, speed, and position of HV are re-collected from the environment; In S3, CAV and HV interaction modeling: CAVs and HVs are divided into leaders and followers, respectively; Based on the requirements of safety, efficiency and comfort during vehicle operation, the profit function f of CAV is designed. p and the profit function f of HV q ; In the formula, is the safety index of CAV, is the efficiency index of CAV, It is the CAV comfort index; are the weight values ​​of the three indicators respectively; similarly, They are the safety, efficiency and comfort indicators of HV. are their respective weight values; The specific indicator calculation is shown in the following formula: In the formula, ΔL p is the distance between CAV vehicle p and the collision point with HV vehicle q, v p (k+1) is the speed of CAV vehicle p at the k+1th moment, a p (k), a p (k+1) are the accelerations of the CAV vehicle p at the kth and k+1th moments respectively; ΔL q is the distance between the HV vehicle q and the conflict point between the CAV vehicle p, v q (k+1) is the speed of HV vehicle q at the k+1th moment, a q (k), a q (k+1) are the accelerations of the HV vehicle q at the kth and k+1th moments respectively; The CAV predicts the optimal acceleration of the HV when it makes a certain acceleration action decision based on the HV's benefit function. Specific implementation: In the CAV's own acceleration strategy space Search for the optimal solution that maximizes its own benefits As shown below: st In the formula, is the list of CAV objects participating in the game M g The acceleration strategy space of CAV acceleration strategy space can be adopted Take the strategy space for HV at this time middle The corresponding HV income, In order to make Maximum optimal acceleration strategy; For when CAV takes When HV adopts the optimal acceleration strategy The corresponding CAV overall benefit is The optimal acceleration strategy for CAV that maximizes the benefit; at the same time, for The minimum and maximum values ​​of for The minimum and maximum values ​​of is a safety constraint, where is the time it takes for the CAV vehicle p to reach the conflict point according to the current state, is the time it takes for the HV vehicle q to reach the conflict point according to the current state; The optimal acceleration strategy will be achieved in CAV The optimal acceleration strategy of HV obtained according to the model As a model in The prediction of HV is denoted as Used for subsequent twin games.

2. The method according to claim 1, It is characterized in that In step S1, the vehicle attribute and status information updating step is specifically as follows: Obtain vehicle type information of each vehicle from the environment, i.e., whether the vehicle is a CAV or HV; kinematics related parameters, including maximum acceleration, maximum deceleration, and maximum speed; At each time step, the current acceleration, speed, and position of the HV in the environment are collected and updated; the CAV updates the vehicle speed and position according to the kinematic formula based on the optimal acceleration output by the model; The vehicle status information is used as the input of the model.

3. The method according to claim 1, It is characterized in that In step S2, the game object list update and model switching steps are specifically as follows: All vehicles in the scene S = {S 1 ,S 2 ,…,S k ,…,S K }, which contains I CAVs, represented by M = {M 1 ,M 2 ,…,M i ,…,M I } and J vehicles HV, expressed as N = {N 1 ,N 2 ,…,N j ,…,N J }; Define the game range, that is, the near conflict area and the conflict area; vehicles entering the game range are added to the game object list; vehicles exit the game object list after passing the conflict area; vehicles in the game object list There are P CAVs, which are expressed as Q HVs, expressed as According to the types of vehicles and the number of each type of vehicles in the game object list, the model is selected and switched. When there is no CAV vehicle or only one vehicle in the game object list, the vehicle can pass freely without considering interactive coordination; When all vehicles in the game object list are CAV vehicles, only CAV collaborative decision-making is performed, and the CAV collaborative decision-making model is selected to solve the optimal acceleration strategy of the vehicle; When the vehicles in the game object list include both CAV and HV, and there is only one CAV, only CAV and HV need to make interactive decisions, and the CAV and HV interactive model is selected to solve the optimal acceleration strategy of CAV; When the vehicles in the game object list include both CAV and HV, and there are multiple CAVs, it is necessary to couple the CAV collaborative decision-making model and the CAV and HV interactive decision-making model to realize collaborative decision-making in the human-machine mixed driving scenario.

4. The method according to claim 1, It is characterized in that The twin game method is used to update the HV weight related parameters θ; At the initial moment, the weight values ​​corresponding to the three CAV indicators are obtained based on the natural driving data calibration And CAV fixed parameter values And the corresponding HV initial weight value related parameters are θ 0 ; in θ 0 Then, r values ​​are discretely taken within the domain to form the initial θ value space Θ 0 , r is set to an odd number; In the formula, the value space Θ 0 The method for selecting the values ​​of θ is as follows: At the initial moment, the range of the value space is determined as The value space except the minimum value Optimal parameters at the initial moment Maximum Outside, in [0,θ 0 ]、 Take (r-3) / 2 values ​​at equal distances and add them to the value space to form a value space Θ with r elements. 0 ; Twin game solution: At each subsequent time step, based on the value space Θ of the previous moment k For each θ value in , the CAV and HV interaction model is used to solve the HV to obtain a series of HV optimal acceleration strategies corresponding to the θ value. Fitness calculation optimal parameters OK: Calculate Θ k The optimal acceleration strategy of HV for each θ value in With actual The degree of difference is shown in the following formula: α is defined as fitness. The larger the α value is, the smaller the difference between the calculated value and the true value is, and the higher the fitness is. The value with the highest fitness α among all the θ values ​​is taken as the optimal parameter for the next moment. Value list Θ k+1 Update: According to the optimal parameters at the next moment And its corresponding fitness size α * , confirm the number of values ​​r at the next moment; when the fitness reaches a certain level, the optimal Perform model calculations and stop playing the twin game until the fitness exceeds the level where twins are not needed; Set the value interval length to Minimum value of the value limit Maximum The calculation method is shown in the following formula: If calculated Then Accordingly If calculated Then Accordingly Finally, the value space Θ of θ at the next moment k+1 exist The nearby discrete values ​​are obtained as shown in the following formula:

5. The method according to claim 4, It is characterized in that The corresponding relationship between the number of twin values ​​and fitness is shown in Table 1: Table 1 Correspondence between the number of twin values ​​and fitness 6. The method according to claim 1, It is characterized in that In step S4, the collaborative decision model in the CAV collaborative decision modeling step is specifically: Considering the requirements of safety, efficiency and comfort during vehicle operation, the characteristic function of CAV is designed, and the individual benefit value of the vehicle is the weighted sum of the three. The calculation method of each indicator is consistent with the CAV and HV interaction model. In the formula, p′ is the number of M in the game object list. g Other CAV vehicles except CAV vehicle p, i.e. Calculate the safety inside the CAV; the weight value Adjust according to the CAV’s personalized preferences; According to the basic principles of cooperative game, the characteristic function, optimization objectives and constraints are designed, the benefits are distributed according to the Shapley principle, and the optimal action of the vehicle is planned; the Shapley value is calculated as shown in the following formula: In the formula, ψ p is the Shapley value of CAV vehicle p; s is M g , which represents the set of possible cooperative game objects of participants in the calculation process; v(s) is the profit generated by the participant set s; |s|, |M g | respectively represent s, M g The number of vehicles included; v(s) is the revenue generated by the participant set for s, which is calculated as: The model is established with the goal of maximizing the overall total benefit of CAV vehicles; As shown below: st In the formula, is the safety constraint between CAVs.

7. The method according to claim 1, It is characterized in that In step S5, the interactive decision model in the human-machine hybrid driving interactive decision modeling step is specifically: Considering the interaction between CAV and HV and the coordination within CAV, a two-layer optimization model is constructed. The outer layer is the interaction between CAV as a whole and HV as shown in the following formula: st In the formula, is the overall benefit of CAV, and is the sum of the benefits of CAVs included in the game object list; the safety constraint considers the safety between CAV and HV; the other quantities have the same meaning as the CAV and HV interaction model; The inner layer is the internal coordination of CAV, and the optimal acceleration strategy of CAV vehicles is obtained by solving the CAV collaborative decision model. However, in the model, when calculating the safety index, all CAVs and HVs in the game object list except itself need to be considered.

8. The method according to claim 1, It is characterized in that In step S6, after selecting an appropriate model according to the game object list to solve the optimal acceleration strategy, the vehicle state is updated; The state of the CAV is expressed and updated according to the kinematic formula: In the formula, To participate in the game CAV vehicle speed, acceleration, and position; The real status of HV is re-collected from the environment to obtain its actual position speed Acceleration

Images