A decision-making planning method for intelligent connected vehicles under heterogeneous traffic flow

Abstract

The present invention discloses a decision-making and planning method for intelligent networked vehicles under heterogeneous traffic flow, and the steps are as follows: identify the intelligent networked status of other vehicles on a local road section and calculate the traffic capacity of each lane; obtain the motion state of other vehicles on the road in the future period of time according to the recognition results of the intelligent networked status of other vehicles; establish an omnidirectional collision risk assessment model to calculate the driving risk of intelligent networked vehicles in real time; make a decision on whether the vehicle needs to change lanes according to the traffic capacity, driving risk and driving risk change rate of the lane where the intelligent networked vehicle is located and the adjacent lane, and then determine the lane changing trajectory according to the traffic efficiency of the adjacent lane and the change of driving risk from changing lanes to the adjacent lane. The present invention can use the roadside base station to accurately obtain the future motion state of some vehicles with networked capabilities, without the need to use a prediction method to obtain it indirectly, which improves the calculation efficiency while also improving the accuracy of the decision-making and planning stage.

Description

Technical Field

[0001] The present invention belongs to the technical field of intelligent transportation systems, and in particular relates to a decision-making planning method for intelligent connected vehicles under heterogeneous traffic flows. Background Art

[0002] With the development of the automobile industry, the mileage of highways and the number of cars in my country are also increasing year by year. While the transportation industry is developing rapidly, it has also brought some new problems: environmental pollution and traffic accidents, which seriously affect people's production and life. In response to the current traffic problems, the American Intelligent Transportation Association took the lead in proposing the concept of intelligent transportation system. It believes that the driver is the most unstable and random link in the entire driving process, and especially points out the important role of smart cars in traffic problems. The American Society of Automotive Engineers (SAE) divides autonomous driving into six levels: L0 (manual driving), L1 (assisted driving), L2 (partial autonomous driving), L3 (conditional autonomous driving), L4 (advanced autonomous driving), and L5 (fully autonomous driving). Among them, L4 and L5 level intelligent vehicles can achieve highly or completely autonomous driving. Studies have found that intelligent vehicles perform better than ordinary drivers in dealing with emergency conditions. On this basis, intelligent connected vehicles are equipped with advanced on-board sensor equipment, controllers and actuators, and combined with modern communication technology, which can realize information interaction (V2X) between vehicles and platforms such as vehicles, roads and clouds, and have functions such as environmental perception, decision-making planning and collaborative control. As the strategic commanding heights of automotive technology, it has become the future development goal of the traditional automotive industry. Against the background of a new round of technological revolution represented by 5G, big data and cloud computing, my country has successively issued a number of development plans to guide the development of intelligent connected vehicles. It can be seen that the development of intelligent connected vehicles is fully in line with the industrial reform trend of "intelligent manufacturing" in my country and has strong practical significance.

[0003] In recent years, the penetration rate of vehicles with autonomous driving functions in the domestic market has increased rapidly. At this stage, autonomous driving vehicles with different levels of autonomous driving functions will gradually penetrate into the traditional road traffic environment at a certain ratio. It takes a long penetration process for traffic flow to evolve from pure manual driving to pure autonomous driving. There must be heterogeneous traffic flow mixed with autonomous driving vehicles and manual driving vehicles. At the same time, with the development of intelligent network technology, vehicles with different levels of intelligent network connection will also appear in road traffic one after another. In the future, there will be a long-term phenomenon of highly complex heterogeneous traffic flows with different levels of autonomous driving and different levels of intelligent network connection mixing on the road.

[0004] However, the decision-making and planning layer is the core link of intelligent connected vehicles. At present, the decision-making and planning methods for intelligent connected vehicles are mostly concentrated on homogeneous traffic flows with the same degree of intelligence and networking. When designing decision-making and planning algorithms, most researchers will assume that the vehicle environment is a homogeneous traffic flow, without considering that heterogeneous traffic flows such as intelligent vehicles, connected vehicles and ordinary vehicles will exist in the long term in the future. Summary of the invention

[0005] In view of the deficiencies of the above-mentioned prior art, the purpose of the present invention is to provide a decision-making and planning method for intelligent connected vehicles under heterogeneous traffic flows, so as to solve the problem that the existing decision-making and planning methods do not fully consider the mixed conditions of heterogeneous traffic flows, resulting in low efficiency, low precision and poor reliability in the decision-making and planning stage. The method of the present invention makes full use of the vehicle-road-cloud network facilities, identifies the intelligent network status and traffic capacity of vehicles traveling on local sections through roadside base stations, adopts diversified methods for heterogeneous vehicles to obtain their future motion state information, and combines the omnidirectional collision risk assessment model to monitor the driving risk of intelligent connected vehicles in real time, and finally makes decisions and plans based on their real-time risks and road traffic capacity, making full use of intelligent network technology, which can greatly improve the calculation efficiency, precision and reliability of the decision-making and planning stage.

[0006] To achieve the above object, the technical solution adopted by the present invention is as follows:

[0007] The intelligent connected vehicle decision-making and planning method under heterogeneous traffic flow of the present invention comprises the following steps:

[0008] (1) The intelligent connected vehicle receives the verification information stream sent by the roadside base station, identifies the intelligent connected status of other vehicles on the local road section, and calculates the traffic capacity of each lane;

[0009] (2) obtaining the movement status of other vehicles on the road within a period of time in the future based on the intelligent network connection status identification result of other vehicles in step (1);

[0010] (3) establishing an omnidirectional collision risk assessment model, inputting the motion state information obtained in step (2) into the omnidirectional collision risk assessment model, and calculating the driving risk of the intelligent connected vehicle in real time;

[0011] (4) Based on the traffic capacity, driving risk, and driving risk change rate of the lane where the intelligent connected vehicle is located and the adjacent lanes, a decision is made on whether the vehicle needs to change lanes. The lane changing trajectory is then determined based on the traffic efficiency of the adjacent lane and the change in driving risk when changing lanes to the adjacent lane.

[0012] Furthermore, the step of identifying the intelligent connected status of other vehicles on the local road section in step (1) is as follows: the roadside base station sends a verification information stream to the vehicles traveling on the road section, and the vehicle returns a corresponding verification information stream after receiving the verification information stream to reflect its reception status. If the roadside base station receives the verification information stream returned by the vehicle, it is classified as an intelligent connected vehicle, and the vehicle is judged to be in automatic or manual driving state based on the returned information; if the roadside base station does not receive the verification information stream returned by the vehicle, the vehicle is classified as a non-intelligent connected vehicle.

[0013] Furthermore, the calculation method of the traffic capacity of each lane in step (1) is:

[0014] Q i =VR i ×D i (1)

[0015] In the formula, Q i is the average traffic volume of the i-th lane, indicating the lane capacity; VR i is the average lane speed of the i-th lane; D i is the average traffic density of the i-th lane.

[0016] Furthermore, the steps of using a diversified method to obtain the motion status of other vehicles on the road in the future period of time in step (2) are as follows:

[0017] (21) If the identification result in step (1) is that the other vehicle is an intelligent network-connected vehicle and is in an automatic driving state, directly use the network-connected communication to obtain the future motion state of the other vehicle and calibrate the obtained information in combination with the roadside base station information;

[0018] (22) If the identification result in step (1) is that the other vehicle is an intelligent network-connected vehicle and is in manual driving state, the state of the driver and the vehicle state information of the vehicle are obtained by using network-connected communication, and the driving intention is identified by combining the driver state and the vehicle state, and the future motion state of the vehicle is predicted;

[0019] (23) If the recognition result in step (1) is that the other vehicle is a non-intelligent network-connected vehicle, the vehicle status information is obtained using the on-board sensing sensor, and the obtained vehicle status information is used to identify the driving intention of the vehicle and predict the future motion state.

[0020] Furthermore, the driving intention recognition step in step (22) is as follows:

[0021] (221) Collecting driver status information and vehicle status information offline, including: driver's line of sight focus, heart rate, breathing rate, head rotation angle, vehicle speed, vehicle acceleration, yaw angular velocity, steering wheel angle, steering wheel angular velocity, vehicle deviation from lane centerline position, and vehicle lateral position to establish a driver intention recognition data set;

[0022] (222) The driver's lane-changing intention corresponding to each set of data in the data set established in step (221) is calibrated, and the weights of all parameters contained in the data set are updated using the ReliefF algorithm. The weights of all parameters are sorted to select the characteristic parameters that best reflect the driver's lane-changing intention. The weight W(A) of any parameter A is calculated as follows:

[0023]

[0024]

[0025] In the formula, diff(A,R,H j ) represents sample R and sample H j The difference in parameter A, p(C) is the proportion of classes C≠class(R), p(classs(R)) is the proportion of samples of the same class as sample R, M j (C) represents the jth nearest neighbor sample in class C≠class(R); k is the number of samples selected with the same classification as parameter A; m is the sample category; max and min are functions for finding the maximum and minimum values, respectively;

[0026] (223) Using the feature parameters selected in step (222) as the input layer of the LSTM neural network, using the driver's lane change intention corresponding to each group of parameters as the output layer of the LSTM neural network, the LSTM neural network is trained to identify the driver's driving intention. The specific steps are as follows:

[0027] (2231) Calculate the forget gate:

[0028] f t =σ(W f ·[h t-1 ,X t ])+b f ) (4)

[0029] In the formula, f t is the forget gate at the current moment, with a value range of 0 to 1; W f is the weight value of the forget gate; X t is the input value at the current moment; h t-1 is the output value of the previous moment; b f is the forget gate bias; σ is the sigmoid function;

[0030] (2232) Calculate the input gate:

[0031] i t =σ(W i ·[h t-1 ,X t ])+b i ) (5)

[0032] In the formula, i t is the input gate at the current moment, with a value range of 0 to 1; W i is the input gate weight value; b i Bias for input gate;

[0033] (2233) Calculate candidate memory unit information:

[0034]

[0035] In the formula, is the candidate information to be updated to the memory unit at the current moment; W C is the candidate information weight value; b C is the candidate information bias; tanh is the hyperbolic tangent function;

[0036] (2234) Calculate new memory unit information:

[0037]

[0038] In the formula, C t is the new memory unit information at the current moment; C t-1 is the memory unit information of the previous moment;

[0039] (2235) Calculate the LSTM neural network output:

[0040] o t =σ(W o ·[h t-1 ,X t ])+b o ) (8)

[0041] h t =o t tanh(C t ) (9)

[0042] In the formula, o t is the initial output at the current moment; W o is the initial output weight value; b o is the initial bias; h t is the output at the current moment, which is the driver’s driving intention.

[0043] Furthermore, the prediction step of the future motion state of the vehicle in step (22) is as follows:

[0044] (224) A short-term low-speed kinematics prediction model is established to predict the future motion state of the vehicle when it is traveling at a low speed, denoted as

[0045]

[0046] Where X is the longitudinal position of the vehicle; Y is the lateral position of the vehicle; v is the vehicle speed; is the vehicle yaw angle; β is the sideslip angle of the center of mass; l f is the distance from the vehicle's center of mass to the front axle; l r is the distance from the vehicle's center of mass to the rear axle; a is the vehicle's acceleration; δ f is the front wheel turning angle of the vehicle; sin, cos and tan are sine, cosine and tangent functions respectively, X t is the longitudinal position of the vehicle at time t, X t+1 Y is the longitudinal position of the vehicle at time t+1; t is the lateral position of the vehicle at time t, Y t+1 is the lateral position of the vehicle at time t+1; is the vehicle yaw angle at time t, is the vehicle yaw angle at time t+1; v t is the vehicle speed at time t, v t+1 is the vehicle speed at time t+1;

[0047] (225) A short-term high-speed dynamics prediction model is established to predict the future motion state of the vehicle when it is traveling at high speed, denoted as

[0048]

[0049] In the formula, m is the vehicle mass; and are the longitudinal velocity and acceleration of the vehicle in the vehicle coordinate system respectively; and are the lateral velocity and acceleration of the vehicle in the vehicle coordinate system respectively; and are the yaw rate and angular acceleration of the vehicle respectively; C f , C r are the cornering stiffness of the front and rear wheels respectively; I z is the moment of inertia of the vehicle mass around the z-axis;

[0050] (226) Combine the motion state SS in the short-term domain obtained in steps (224) and (225) K or SS D, a fifth-order polynomial fitting is used to generate the vehicle motion state information in the future long-term domain, recorded as

[0051]

[0052] Where t is time; a i and b i are all polynomial coefficients, i=0,1,2,3,4,5; v X and v Y are the components of vehicle speed v on the X-axis and Y-axis respectively.

[0053] Furthermore, the driving intention of the vehicle in step (23) is recognized by using an LSTM neural network method, and the input layer parameters of the LSTM neural network are selected as: target vehicle speed, lateral acceleration, vehicle offset lane centerline position and vehicle lateral position information.

[0054] Furthermore, in the step (23), the method for predicting the future motion state of the vehicle adopts a quintic polynomial combined with a multi-objective optimization method. After obtaining the driving intention, a series of candidate trajectories tra = [L 1 ,L 2 ,…,L n ], and then design the objective function to use the multi-objective optimization method to select a trajectory that conforms to the actual situation as the predicted trajectory, and output the vehicle's future motion state information. The objective function is:

[0055]

[0056] In the formula, a yL (t) is the lateral acceleration of trajectory L; a ymax is the maximum permissible lateral acceleration; LO L is the length of the trajectory L; LO max is the maximum allowed trajectory length; ΔL is the trajectory error between the candidate trajectory and the previous moment; Δ max is the maximum allowed trajectory error; w 1 、w 2 and w 3 is the weight coefficient, and the sum of the three is 1.

[0057] Furthermore, the steps for establishing the omnidirectional collision risk assessment model in step (3) are as follows:

[0058] (31) Establish the omnidirectional collision time OTTC model:

[0059]

[0060] Where, OR is the straight-line distance between the center of mass of the ego vehicle and other vehicles; ΔOV is the relative speed of the ego vehicle and other vehicles; (X 0 ,Y 0 ) is the center of mass of the vehicle; (X i ,Y i ) is the center of mass position of the target vehicle; V 0 and V i are the speeds of the ego vehicle and the other vehicle respectively; θ is the angle between the ego vehicle’s front direction and the line connecting the centers of mass of the two vehicles; sgn is the sign function; and are the yaw angles of the vehicle and other vehicles respectively; π is pi;

[0061] (32) Establish the OTHW model of omnidirectional collision headway:

[0062]

[0063] (33) Establish an omnidirectional safety distance OR safe Model:

[0064]

[0065] Where D 0safe and D isafe are the emergency braking distances of the vehicle and other vehicles respectively;

[0066] (34) Combining the OTTC model, OTHW model and OR established in steps (31)-(33) safe Model, establish an omnidirectional collision risk assessment model:

[0067]

[0068] Where ξ is the omnidirectional collision risk and g is the gravitational acceleration.

[0069] Furthermore, the specific steps of step (4) are as follows:

[0070] (41) The intelligent connected vehicle calculates the driving risk of its current location in real time and receives the average flow rate of each lane sent by the roadside base station;

[0071] (42) If the driving risk ξ of the current location of the intelligent connected vehicle is greater than the risk threshold ξ max and Greater than the risk change rate threshold That is, ξ>ξ max and Or the average flow rate Q of the current lane c Less than the average flow rate Q of the adjacent lane n , i.e. Q c <Qn ; The intelligent connected vehicle issues a lane change decision command and uses a quintic polynomial method to generate candidate trajectories for changing lanes to the adjacent lane;

[0072] (43) inputting the candidate trajectories generated in step (42) into the omnidirectional collision risk assessment model, evaluating the omnidirectional collision risk of each lane changing trajectory, screening out the candidate trajectories that meet the safe lane changing conditions, and then further screening the candidate trajectories using the objective function designed in the above step (23) to select the optimal lane changing trajectory;

[0073] (44) If step (43) fails to solve the optimal lane-changing trajectory, then max and Emergency braking is performed at Q c n Brake and slow down to follow the vehicle in front.

[0074] Beneficial effects of the present invention:

[0075] 1. The present invention takes into account the impact of the heterogeneous (different intelligence levels) characteristics of vehicles on future roads on the decision-making and planning of intelligent connected vehicles, and makes full use of intelligent network technology and autonomous driving technology to greatly improve the decision-making and planning efficiency of intelligent connected vehicles under heterogeneous traffic flows.

[0076] 2. The present invention can utilize roadside base stations to accurately obtain the future motion states of some vehicles with networking capabilities, without the need to use prediction methods to obtain them indirectly, thereby improving computing efficiency and the accuracy of the decision-making and planning stages. BRIEF DESCRIPTION OF THE DRAWINGS

[0077] Figure 1 It is the principle diagram of the method of the present invention;

[0078] Figure 2 Schematic diagram of heterogeneous traffic flow. DETAILED DESCRIPTION

[0079] In order to facilitate the understanding of those skilled in the art, the present invention is further described below in conjunction with embodiments and drawings. The contents mentioned in the implementation modes are not intended to limit the present invention.

[0080] Reference Figure 1 As shown, a decision-making planning method for intelligent connected vehicles under heterogeneous traffic flow of the present invention comprises the following steps:

[0081] (1) The intelligent connected vehicle receives the verification information stream sent by the roadside base station, identifies the intelligent connected status of other vehicles on the local road section, and calculates the traffic capacity of each lane; Figure 2 As shown;

[0082] ​Among them, the step of identifying the intelligent connected status of other vehicles on the local road section is: the roadside base station sends a verification information stream to the vehicles traveling on the road section, and the vehicle returns the corresponding verification information stream after receiving the verification information stream to reflect its reception status. If the roadside base station receives the verification information stream returned by the vehicle, it is classified as an intelligent connected vehicle, and the vehicle is judged to be in automatic or manual driving state based on the returned information; if the roadside base station does not receive the verification information stream returned by the vehicle, the vehicle is classified as a non-intelligent connected vehicle.

[0083] The calculation method of the traffic capacity of each lane is as follows:

[0084] Q i =VR i ×D i (1)

[0085] In the formula, Q i is the average vehicle flow rate of the i-th lane (vehicles / h), indicating the lane capacity; VR i is the average lane speed of the i-th lane (m / s); D i is the average traffic density of the ith lane (vehicles / km).

[0086] (2) obtaining the movement status of other vehicles on the road within a period of time in the future based on the intelligent network connection status identification result of other vehicles in step (1);

[0087] Among them, the steps of using a diversified method to obtain the motion status of other vehicles on the road in the future are as follows:

[0088] (21) If the identification result in step (1) is that the other vehicle is an intelligent network-connected vehicle and is in an automatic driving state, directly use the network-connected communication to obtain the future motion state of the other vehicle and calibrate the obtained information in combination with the roadside base station information;

[0089] (22) If the identification result in step (1) is that the other vehicle is an intelligent network-connected vehicle and is in manual driving state, the state of the driver and the vehicle state information of the vehicle are obtained by using network-connected communication, and the driving intention is identified by combining the driver state and the vehicle state, and the future motion state of the vehicle is predicted;

[0090] (23) If the recognition result in step (1) is that the other vehicle is a non-intelligent network-connected vehicle, the vehicle status information is obtained using the on-board sensing sensor, and the obtained vehicle status information is used to identify the driving intention of the vehicle and predict the future motion state.

[0091] Specifically, the driving intention recognition step in step (22) is as follows:

[0092] (221) Collecting driver status information and vehicle status information offline, including: driver's line of sight focus, heart rate, breathing rate, head rotation angle, vehicle speed, vehicle acceleration, yaw angular velocity, steering wheel angle, steering wheel angular velocity, vehicle deviation from lane centerline position, and vehicle lateral position to establish a driver intention recognition data set;

[0093] (222) The driver's lane-changing intention corresponding to each set of data in the data set established in step (221) is calibrated, and the weights of all parameters contained in the data set are updated using the ReliefF algorithm. The weights of all parameters are sorted to select the characteristic parameters that best reflect the driver's lane-changing intention. The weight W(A) of any parameter A is calculated as follows:

[0094]

[0095]

[0096] In the formula, diff(A,R,H j ) represents sample R and sample H j The difference in parameter A, p(C) is the proportion of classes C≠class(R), p(classs(R)) is the proportion of samples of the same class as sample R, M j (C) represents the jth nearest neighbor sample in class C≠class(R); k is the number of samples selected with the same classification as parameter A; m is the sample category; max and min are functions for finding the maximum and minimum values, respectively;

[0097] (223) Using the feature parameters selected in step (222) as the input layer of the LSTM neural network, using the driver's lane change intention corresponding to each group of parameters as the output layer of the LSTM neural network, the LSTM neural network is trained to identify the driver's driving intention. The specific steps are as follows:

[0098] (2231) Calculate the forget gate:

[0099] f t =σ(W f ·[h t-1 ,X t ])+b f ) (4)

[0100] In the formula, f t is the forget gate at the current moment, with a value range of 0 to 1; W f is the weight value of the forget gate; X t is the input value at the current moment; h t-1 is the output value of the previous moment; b f is the forget gate bias; σ is the sigmoid function;

[0101] (2232) Calculate the input gate:

[0102] i t =σ(W i ·[h t-1 ,X t ])+b i ) (5)

[0103] In the formula, i t is the input gate at the current moment, with a value range of 0 to 1; W i is the input gate weight value; b i Bias for input gate;

[0104] (2233) Calculate candidate memory unit information:

[0105]

[0106] In the formula, is the candidate information to be updated to the memory unit at the current moment; W C is the candidate information weight value; b C is the candidate information bias; tanh is the hyperbolic tangent function;

[0107] (2234) Calculate new memory unit information:

[0108]

[0109] In the formula, C t is the new memory unit information at the current moment; C t-1 is the memory unit information of the previous moment;

[0110] (2235) Calculate the LSTM neural network output:

[0111] o t =σ(W o ·[h t-1 ,X t ])+b o ) (8)

[0112] h t =o t tanh(C t ) (9)

[0113] In the formula, o t is the initial output at the current moment; W o is the initial output weight value; b o is the initial bias; h t is the output at the current moment, which is the driver’s driving intention.

[0114] Specifically, the prediction step of the future motion state of the vehicle in step (22) is as follows:

[0115] (224) A short-term low-speed kinematics prediction model is established to predict the future motion state of the vehicle when it is traveling at a low speed, denoted as

[0116]

[0117] Where X is the longitudinal position of the vehicle; Y is the lateral position of the vehicle; v is the vehicle speed; is the vehicle yaw angle; β is the sideslip angle of the center of mass; l f is the distance from the vehicle's center of mass to the front axle; l r is the distance from the vehicle's center of mass to the rear axle; a is the vehicle's acceleration; δ f is the front wheel turning angle of the vehicle; sin, cos and tan are sine, cosine and tangent functions respectively, X t is the longitudinal position of the vehicle at time t, X t+1 Y is the longitudinal position of the vehicle at time t+1; t is the lateral position of the vehicle at time t, Y t+1 is the lateral position of the vehicle at time t+1; is the vehicle yaw angle at time t, is the vehicle yaw angle at time t+1; v t is the vehicle speed at time t, v t+1 is the vehicle speed at time t+1;

[0118] (225) A short-term high-speed dynamics prediction model is established to predict the future motion state of the vehicle when it is traveling at high speed, denoted as

[0119]

[0120] In the formula, m is the vehicle mass; and are the longitudinal velocity and acceleration of the vehicle in the vehicle coordinate system respectively; and are the lateral velocity and acceleration of the vehicle in the vehicle coordinate system respectively; and are the yaw rate and angular acceleration of the vehicle respectively; C f , C r are the cornering stiffness of the front and rear wheels respectively; I z is the moment of inertia of the vehicle mass around the z-axis;

[0121] (226) Combine the motion state SS in the short-term domain obtained in steps (224) and (225) K or SS D, a fifth-order polynomial fitting is used to generate the vehicle motion state information in the future long-term domain, recorded as

[0122]

[0123] Where t is time; a i and b i are all polynomial coefficients, i=0,1,2,3,4,5; v X and v Y are the components of vehicle speed v on the X-axis and Y-axis respectively.

[0124] The driving intention of the vehicle in step (23) is identified by using an LSTM neural network method, and the input layer parameters of the LSTM neural network are selected as: target vehicle speed, lateral acceleration, vehicle offset lane centerline position and vehicle lateral position information.

[0125] In the step (23), the method for predicting the future motion state of the vehicle adopts a quintic polynomial combined with a multi-objective optimization method. After obtaining the driving intention, a series of candidate trajectories tra = [L 1 ,L 2 ,…,L n ], and then design the objective function to use the multi-objective optimization method to select a trajectory that conforms to the actual situation as the predicted trajectory, and output the vehicle's future motion state information. The objective function is:

[0126]

[0127] In the formula, a yL (t) is the lateral acceleration of trajectory L; a ymax is the maximum permissible lateral acceleration; LO L is the length of the trajectory L; LO max is the maximum allowed trajectory length; ΔL is the trajectory error between the candidate trajectory and the previous moment; Δ max is the maximum allowed trajectory error; w 1 、w 2 and w 3 is the weight coefficient, and the sum of the three is 1.

[0128] (3) establishing an omnidirectional collision risk assessment model, inputting the motion state information obtained in step (2) into the omnidirectional collision risk assessment model, and calculating the driving risk of the intelligent connected vehicle in real time;

[0129] The steps to establish the omnidirectional collision risk assessment model are as follows:

[0130] (31) Establish the omnidirectional collision time OTTC model:

[0131]

[0132] Where, OR is the straight-line distance between the center of mass of the ego vehicle and other vehicles; ΔOV is the relative speed of the ego vehicle and other vehicles; (X 0 ,Y 0 ) is the center of mass of the vehicle; (X i ,Y i ) is the center of mass position of the target vehicle; V 0 and V i are the speeds of the ego vehicle and other vehicles respectively; θ is the angle between the ego vehicle’s front direction and the line connecting the two vehicles’ centers of mass (counterclockwise is positive); sgn is the sign function; and are the yaw angles of the vehicle and other vehicles respectively; π is pi;

[0133] (32) Establish the OTHW model of omnidirectional collision headway:

[0134]

[0135] (33) Establish an omnidirectional safety distance OR safe Model:

[0136]

[0137] Where D 0safe and D isafe are the emergency braking distances of the vehicle and other vehicles respectively;

[0138] (34) Combining the OTTC model, OTHW model and OR established in steps (31)-(33) safe Model, establish an omnidirectional collision risk assessment model:

[0139]

[0140] Where ξ is the omnidirectional collision risk and g is the gravitational acceleration.

[0141] (4) Based on the traffic capacity, driving risk, and driving risk change rate of the lane where the intelligent connected vehicle is located and the adjacent lanes, a decision is made on whether the vehicle needs to change lanes. The lane change trajectory is then determined based on the traffic efficiency of the adjacent lane and the change in driving risk when changing lanes to the adjacent lane.

[0142] (41) The intelligent connected vehicle calculates the driving risk of its current location in real time and receives the average flow rate of each lane sent by the roadside base station;

[0143] (42) If the driving risk ξ of the current location of the intelligent connected vehicle is greater than the risk threshold ξ max and Greater than the risk change rate threshold That is, ξ>ξ max and Or the average flow rate Q of the current lane c Less than the average flow rate Q of the adjacent lane n , i.e. Q c n ; The intelligent connected vehicle issues a lane change decision command and uses a quintic polynomial method to generate candidate trajectories for changing lanes to the adjacent lane;

[0144] (43) inputting the candidate trajectories generated in step (42) into the omnidirectional collision risk assessment model, evaluating the omnidirectional collision risk of each lane changing trajectory, screening out the candidate trajectories that meet the safe lane changing conditions, and then further screening the candidate trajectories using the objective function designed in the above step (23) to select the optimal lane changing trajectory;

[0145] (44) If step (43) fails to solve the optimal lane-changing trajectory, then max and Emergency braking is performed at Q c n Brake and slow down to follow the vehicle in front.

[0146] The present invention has many specific application paths. The above is only a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements can be made without departing from the principle of the present invention. These improvements should also be regarded as the protection scope of the present invention.​​

Claims

1. A decision-making and planning method for intelligent connected vehicles under heterogeneous traffic flow, It is characterized in that Here are the steps: (1) The intelligent connected vehicle receives the verification information stream sent by the roadside base station, identifies the intelligent connected status of other vehicles on the local road section, and calculates the traffic capacity of each lane; (2) obtaining the movement status of other vehicles on the road within a period of time in the future based on the intelligent network connection status identification result of other vehicles in step (1); (3) establishing an omnidirectional collision risk assessment model, inputting the motion state information obtained in step (2) into the omnidirectional collision risk assessment model, and calculating the driving risk of the intelligent networked vehicle in real time; (4) Based on the traffic capacity, driving risk, and driving risk change rate of the lane where the intelligent connected vehicle is located and the adjacent lanes, a decision is made on whether the vehicle needs to change lanes. The lane change trajectory is then determined based on the traffic efficiency of the adjacent lane and the change in driving risk when changing lanes to the adjacent lane. The steps of using a diversified method to obtain the motion status of other vehicles on the road in the future period of time in step (2) are as follows: (21) If the identification result in step (1) is that the other vehicle is an intelligent network-connected vehicle and is in an automatic driving state, directly use the network-connected communication to obtain the future motion state of the other vehicle and calibrate the obtained information in combination with the roadside base station information; (22) If the identification result in step (1) is that the other vehicle is an intelligent network-connected vehicle and is in manual driving state, the state of the driver and the vehicle state information of the vehicle are obtained by using network-connected communication, and the driving intention is identified by combining the driver state and the vehicle state, and the future motion state of the vehicle is predicted; (23) If the identification result in step (1) is that the other vehicle is a non-intelligent network-connected vehicle, the vehicle status information is obtained by using the vehicle-mounted sensing sensor, and the obtained vehicle status information is used to identify the driving intention of the vehicle and predict the future motion state; The driving intention recognition step in step (22) is as follows: (221) Collecting driver status information and vehicle status information offline, including: driver's line of sight focus, heart rate, breathing rate, head rotation angle, vehicle speed, vehicle acceleration, yaw angular velocity, steering wheel angle, steering wheel angular velocity, vehicle deviation from lane centerline position, and vehicle lateral position to establish a driver intention recognition data set; (222) The driver's lane-changing intention corresponding to each set of data in the data set established in step (221) is calibrated, and the weights of all parameters contained in the data set are updated using the ReliefF algorithm. The weights of all parameters are sorted to select the characteristic parameters that best reflect the driver's lane-changing intention. The weight W(A) of any parameter A is calculated as follows: In the formula, diff(A,R,H j ) represents sample R and sample H j The difference in parameter A, p(C) is the proportion of classes C≠class(R), p(classs(R) is the proportion of samples of the same class as sample R, M j (C) represents the jth nearest neighbor sample in class C≠class(R); k is the number of samples selected with the same classification as parameter A; m is the sample category; max and min are functions for finding the maximum and minimum values, respectively; (223) Using the feature parameters selected in step (222) as the input layer of the LSTM neural network, using the driver's lane change intention corresponding to each group of parameters as the output layer of the LSTM neural network, the LSTM neural network is trained to identify the driver's driving intention. The specific steps are as follows: (2231) Calculate the forget gate: f t =σ(W f ·[h t-1 ,X t ]+b f ) (4) In the formula, f t is the forget gate at the current moment, with a value range of 0 to 1; W f is the weight value of the forget gate; X t is the input value at the current moment; h t-1 is the output value of the previous moment; b f is the forget gate bias; σ is the sigmoid function; (2232) Calculate the input gate: i t =σ(W i ·[h t-1 ,X t ]+b i ) (5) In the formula, i t is the input gate at the current moment, with a value range of 0 to 1; W i is the input gate weight value; b i Bias for input gate; (2233) Calculate candidate memory unit information: In the formula, is the candidate information to be updated to the memory unit at the current moment; W C is the candidate information weight value; b C is the candidate information bias; tanh is the hyperbolic tangent function; (2234) Calculate new memory unit information: In the formula, C t is the new memory unit information at the current moment; C t-1 is the memory unit information of the previous moment; (2235) Calculate the LSTM neural network output: the t =σ(W o ·[h t-1 ,X t ]+b o ) (8) h t =o t ·tanh(C t ) (9) In the formula, o t is the initial output at the current moment; W o is the initial output weight value; b o is the initial bias; h t is the output at the current moment, i.e., the driver’s driving intention; The steps for predicting the future motion state of the vehicle in step (22) are as follows: (224) A short-term low-speed kinematics prediction model is established to predict the future motion state of the vehicle when it is traveling at a low speed, denoted as Where X is the longitudinal position of the vehicle; Y is the lateral position of the vehicle; v is the vehicle speed; is the vehicle yaw angle; β is the sideslip angle of the center of mass; l f is the distance from the vehicle's center of mass to the front axle; l r is the distance from the vehicle's center of mass to the rear axle; a is the vehicle's acceleration; δ f is the front wheel turning angle of the vehicle; sin, cos and tan are sine, cosine and tangent functions respectively, X t is the longitudinal position of the vehicle at time t, X t+1 Y is the longitudinal position of the vehicle at time t+1; t is the lateral position of the vehicle at time t, Y t+1 is the lateral position of the vehicle at time t+1; is the vehicle yaw angle at time t, is the vehicle yaw angle at time t+1; v t is the vehicle speed at time t, v t+1 is the vehicle speed at time t+1; (225) A short-term high-speed dynamics prediction model is established to predict the future motion state of the vehicle when it is traveling at high speed, denoted as In the formula, m is the vehicle mass; and are the longitudinal velocity and acceleration of the vehicle in the vehicle coordinate system respectively; and are the lateral velocity and acceleration of the vehicle in the vehicle coordinate system respectively; and are the yaw rate and angular acceleration of the vehicle respectively; C f , C r are the cornering stiffness of the front and rear wheels respectively; I z is the moment of inertia of the vehicle mass around the z-axis; (226) Combine the motion state SS in the short-term domain obtained in steps (224) and (225) K or SS D , a fifth-order polynomial fitting is used to generate the vehicle motion state information in the future long-term domain, recorded as Where t is time; a i and b i are all polynomial coefficients, i=0,1,2,3,4,5; v X and v Y are the components of vehicle speed v on the X-axis and Y-axis respectively; The steps for establishing the omnidirectional collision risk assessment model in step (3) are as follows: (31) Establish the omnidirectional collision time OTTC model: Where, OR is the straight-line distance between the center of mass of the ego vehicle and other vehicles; ΔOV is the relative speed of the ego vehicle and other vehicles; (X 0 ,Y 0 ) is the center of mass of the vehicle; (X i ,Y i ) is the center of mass position of the target vehicle; V 0 and V i are the speeds of the ego vehicle and the other vehicle respectively; θ is the angle between the ego vehicle’s front direction and the line connecting the centers of mass of the two vehicles; sgn is the sign function; and are the yaw angles of the vehicle and other vehicles respectively; π is pi; (32) Establish the OTHW model of omnidirectional collision headway: (33) Establish an omnidirectional safety distance OR safe Model: Where D 0safe and D isafe are the emergency braking distances of the vehicle and other vehicles respectively; (34) Combining the OTTC model, OTHW model and OR established in steps (31)-(33) safe Model, establish an omnidirectional collision risk assessment model: Where ξ is the omnidirectional collision risk and g is the gravitational acceleration.

2. According to claim 1, the intelligent connected vehicle decision-making planning method under heterogeneous traffic flow, It is characterized in that The step of identifying the intelligent network connection status of other vehicles on the local road section in step (1) is as follows: the roadside base station sends a verification information flow to the vehicles traveling on the road section, and the vehicles return a corresponding verification information flow after receiving the verification information flow to reflect their reception status. If the roadside base station receives the verification information flow returned by the vehicle, it is classified as an intelligent network connection vehicle, and the vehicle is judged to be in an automatic or manual driving state based on the returned information; If the roadside base station does not receive the verification information flow returned by the vehicle, the vehicle will be classified as a non-intelligent connected vehicle.

3. The intelligent connected vehicle decision-making and planning method under heterogeneous traffic flow according to claim 1, It is characterized in that The calculation method of the traffic capacity of each lane in step (1) is: Q i =VR i ×D i (1) In the formula, Q i is the average traffic volume of the i-th lane, indicating the lane capacity; VR i is the average lane speed of the i-th lane; D i is the average traffic density of the i-th lane.

4. The intelligent connected vehicle decision-making and planning method under heterogeneous traffic flow according to claim 1, It is characterized in that The driving intention of the vehicle in step (23) is identified by using an LSTM neural network method, and the input layer parameters of the LSTM neural network are selected as: target vehicle speed, lateral acceleration, vehicle offset lane centerline position and vehicle lateral position information.

5. The intelligent connected vehicle decision-making and planning method under heterogeneous traffic flow according to claim 1, It is characterized in that In the step (23), the method for predicting the future motion state of the vehicle adopts a quintic polynomial combined with a multi-objective optimization method. After obtaining the driving intention, a series of candidate trajectories tra = [L 1 ,L 2 ,…,L n ], and then design the objective function to use the multi-objective optimization method to select a trajectory that conforms to the actual situation as the predicted trajectory, and output the vehicle's future motion state information. The objective function is: In the formula, a yL (t) is the lateral acceleration of trajectory L; a ymax is the maximum permissible lateral acceleration; LO L is the length of the trajectory L; LO max is the maximum allowed trajectory length; ΔL is the trajectory error between the candidate trajectory and the previous moment; Δ max is the maximum allowed trajectory error; w 1 、w 2 and w 3 is the weight coefficient, and the sum of the three is 1.

6. The intelligent connected vehicle decision-making and planning method under heterogeneous traffic flow according to claim 1, It is characterized in that The specific steps of step (4) are as follows: (41) The intelligent connected vehicle calculates the driving risk of its current location in real time and receives the average flow rate of each lane sent by the roadside base station; (42) If the driving risk ξ of the current location of the intelligent connected vehicle is greater than the risk threshold ξ max and Greater than the risk change rate threshold That is, ξ>ξ max and Or the average flow rate Q of the current lane c Less than the average flow rate Q of the adjacent lane n , i.e. Q c n ; The intelligent connected vehicle issues a lane change decision command and uses a quintic polynomial method to generate candidate trajectories for changing lanes to the adjacent lane;​ (43) inputting the candidate trajectories generated in step (42) into the omnidirectional collision risk assessment model, evaluating the omnidirectional collision risk of each lane changing trajectory, screening out the candidate trajectories that meet the safe lane changing conditions, and then further screening the candidate trajectories using the objective function designed in the above step (23) to select the optimal lane changing trajectory; (44) If step (43) fails to solve the optimal lane-changing trajectory, then max and Emergency braking is performed at Q c n Brake and slow down to follow the vehicle in front.​

Images