CN109927725B - Self-adaptive cruise system with driving style learning capability and implementation method - Google Patents
Self-adaptive cruise system with driving style learning capability and implementation method Download PDFInfo
- Publication number
- CN109927725B CN109927725B CN201910077516.9A CN201910077516A CN109927725B CN 109927725 B CN109927725 B CN 109927725B CN 201910077516 A CN201910077516 A CN 201910077516A CN 109927725 B CN109927725 B CN 109927725B
- Authority
- CN
- China
- Prior art keywords
- vehicle
- following
- driving style
- learning
- acceleration
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
- 238000000034 method Methods 0.000 title claims abstract description 69
- 230000002787 reinforcement Effects 0.000 claims abstract description 29
- 230000004927 fusion Effects 0.000 claims abstract description 23
- 230000003044 adaptive effect Effects 0.000 claims abstract description 19
- 230000008447 perception Effects 0.000 claims abstract description 18
- 230000006399 behavior Effects 0.000 claims abstract description 12
- 238000013461 design Methods 0.000 claims abstract description 12
- 230000001133 acceleration Effects 0.000 claims description 63
- 230000006870 function Effects 0.000 claims description 42
- 239000011159 matrix material Substances 0.000 claims description 29
- 239000013598 vector Substances 0.000 claims description 27
- 238000004422 calculation algorithm Methods 0.000 claims description 21
- 238000005457 optimization Methods 0.000 claims description 16
- 238000004891 communication Methods 0.000 claims description 7
- 238000009795 derivation Methods 0.000 claims description 6
- 230000002159 abnormal effect Effects 0.000 claims description 4
- 230000008859 change Effects 0.000 claims description 4
- 230000009471 action Effects 0.000 claims description 3
- 238000004364 calculation method Methods 0.000 claims description 3
- 238000007405 data analysis Methods 0.000 claims description 3
- 230000004069 differentiation Effects 0.000 claims description 3
- 239000000446 fuel Substances 0.000 claims description 3
- 238000005070 sampling Methods 0.000 claims description 3
- 238000012549 training Methods 0.000 claims description 3
- KUEUWHJGRZKESU-UHFFFAOYSA-N Niceritrol Chemical compound C=1C=CN=CC=1C(=O)OCC(COC(=O)C=1C=NC=CC=1)(COC(=O)C=1C=NC=CC=1)COC(=O)C1=CC=CN=C1 KUEUWHJGRZKESU-UHFFFAOYSA-N 0.000 claims description 2
- 238000011156 evaluation Methods 0.000 claims description 2
- 230000008569 process Effects 0.000 description 5
- 238000010586 diagram Methods 0.000 description 4
- 238000005516 engineering process Methods 0.000 description 4
- 238000011161 development Methods 0.000 description 3
- 238000004458 analytical method Methods 0.000 description 2
- 238000012545 processing Methods 0.000 description 2
- 241000256602 Isoptera Species 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000001914 filtration Methods 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 238000009434 installation Methods 0.000 description 1
- 238000012827 research and development Methods 0.000 description 1
- 238000012360 testing method Methods 0.000 description 1
Images
Landscapes
- Control Of Driving Devices And Active Controlling Of Vehicle (AREA)
Abstract
The invention belongs to the technical field of intelligent auxiliary driving and automobile safety of automobiles, and particularly relates to an adaptive cruise system with driving style learning capability and an implementation method, wherein the adaptive cruise system changes the adaptive cruise following performance of the system through online learning of driver style behaviors. The invention designs an adaptive cruise control system with driving style self-learning capability and an implementation method thereof by considering the problem of automobile adaptive cruise control. The system is suitable for an L2-level automatic driving vehicle, aims to realize self-adaption to follow the front vehicle in an actual driving scene through environment perception and information fusion, and simultaneously, the driving strategy of the system considers the driving style of a specific driver, so that the system has consistent satisfaction and acceptance for different crowds. To achieve this goal, the present invention uses a combination of linear quadratic control and reinforcement learning online learning methods to better optimize system performance.
Description
Technical Field
The invention belongs to the technical field of intelligent auxiliary driving and automobile safety of automobiles, and particularly relates to an adaptive cruise system with driving style learning capability and an implementation method, wherein the adaptive cruise system changes the adaptive cruise following performance of the system through online learning of driver style behaviors.
Background
The intelligent driving assistance technology gradually develops, becomes a main development direction of vehicle technology research and development, and draws wide attention worldwide. Cruise systems are gradually evolving in the context of vehicle system automation and intelligence. In the design of the currently developed adaptive cruise system control system, in addition to the more traditional methods such as PID and fuzzy control, an advanced control algorithm based on a model is used in the adaptive cruise system, and the following performance is improved by using a model predictive control method in the patents of application numbers 201810313067.9 and 201710826862.3. However, these methods still mainly consider vehicle following performance, such as safety, comfort and economy.
With the continuous improvement and maturity of vehicle technology, a higher development direction is provided for the adaptive cruise. In order to better improve the system performance, the cruise control of the vehicle not only needs to ensure indexes such as safety and the like under the traffic flow, but also needs to consider the subjective intention of a driver. Since the driving styles of drivers are greatly different among human drivers, the design of the control system precedes the actual use of the vehicle. Therefore, when the control system needs to be better adapted to the driving style of different specific drivers, a self-learning capability is further required for the system. Such methods are currently immature. For example, in the patent application No. 2010106159140, the driver is classified and learned by off-line data driving analysis, and the driver's acceptance of the system is improved by changing parameters of the adaptive cruise control system in the form of vehicle buttons or the like. The patent of application No. 201710812719.9 also learns driver driving habits and adjusts controller parameters by off-line learning and key press. Unlike the jerry of application No. 201710812719.9, the present application can autonomously learn the driving style of the driver.
Disclosure of Invention
In order to solve the problems, the invention designs an adaptive cruise system with driving style learning capability and an implementation method thereof by considering the problem of automobile adaptive cruise control. The system is suitable for an L2-level automatic driving vehicle, aims to realize self-adaption to follow the front vehicle in an actual driving scene through environment perception and information fusion, and simultaneously, the driving strategy of the system considers the driving style of a specific driver, so that the system has consistent satisfaction and acceptance for different crowds. To achieve this goal, the present invention uses a combination of linear quadratic control and reinforcement learning online learning methods to better optimize system performance.
The technical scheme of the invention is described as follows by combining the attached drawings:
a method for realizing an adaptive cruise system with driving style learning capability comprises a perception fusion module A, a following vehicle control module B, a driving style self-learning module C and a bottom vehicle execution control module D;
the perception fusion module A is used for obtaining the running state information of the front vehicle, including the speed information and the relative distance of the front vehicle;
the following control module B comprises a vehicle following control module B-a and a preceding vehicle acceleration prediction module B-B, wherein the vehicle following control module B-a is used for establishing a vehicle following model, establishing a control problem and determining an optimized target to obtain a following controller; the front vehicle acceleration predicting module B-B is used for predicting the front vehicle acceleration according to the confirmed front vehicle and the running state information of the front vehicle and the current vehicle;
the driving style self-learning module C is used for learning the driving style of a specific driver based on a reinforcement learning method aiming at the characteristic that the driving style of the specific driver is different, so as to adjust the control parameter of the optimal cruise control problem and achieve the function of a self-learning system;
the vehicle execution control module D is used for performing tracking control and finally outputting a vehicle power driving system and a braking system to control driving of the vehicle;
the perception fusion module A is unidirectionally connected with the following vehicle control module B and the driving style self-learning module C; the following vehicle control module B is connected with the vehicle execution control module D in a one-way mode; the driving style self-learning module C is connected with the following vehicle control module B in a one-way mode; the vehicle execution control module D is connected with the vehicle in a one-way mode.
The method comprises the following steps:
firstly, confirming a front vehicle to be followed through a perception fusion module A self fusion recognition algorithm, and obtaining the running state information of the front vehicle, including the speed information and the relative distance of the front vehicle; and the information such as the speed, the engine moment, the braking deceleration and the like of the current vehicle are obtained from the communication of a vehicle bus, namely a CAN network;
step two, establishing a vehicle following model through a vehicle following control module B-a in the vehicle following control module B, establishing a control problem, determining an optimized target and obtaining a vehicle following controller; the vehicle-following control module B is used for controlling the vehicle-following control module B to carry out a vehicle-following control process according to the current vehicle speed information and the current vehicle speed information;
step three, learning the driving style of a specific driver based on a reinforcement learning method aiming at the characteristic that the driving style of the specific driver has difference through a driving style self-learning module C, and further adjusting the control parameter of the optimal cruise control problem to achieve the function of a self-learning system;
and fourthly, performing tracking control through the bottom vehicle execution control module D, and finally outputting a vehicle power driving system and a braking system to control driving of the vehicle.
The specific method of the second step is as follows:
2.1) vehicle following model establishment: according to the self-adaptive following problem requirement and the longitudinal kinematics and dynamics characteristics of the vehicle, a vehicle following model is established;
in order to describe the longitudinal dynamics of the vehicle in a following scene, the system introduces two state variables,
respectively as follows:
＝d-dref(1a)
where d is the relative distance between the vehicle and the front vehicle, drefIs referenced to the relative distance, dref＝d0+τvh，d0Is the safe parking distance, τ is the expected headway, vhIs the speed of the vehicle, vtIs the front vehicle speed; therefore, the differential equation of the longitudinal dynamics of the vehicle in the following scene can use the above state variables,
represents:
wherein, ahIs the longitudinal acceleration of the vehicle, atIs the front vehicle longitudinal acceleration, τ is the desired headway; it is assumed here that the inner loop acceleration tracking control module of the host vehicle can be expanded to the first order approximation:
wherein, tauiIs the inner loop dynamic time constant, ahIs the longitudinal acceleration of the vehicle, arefIs the desired longitudinal acceleration;
in a continuous system, the state quantities are selected as
The controlled variable is u ═ arefModeling the acceleration of the front vehicle as system disturbance, namely disturbance quantity d ═ ah(ii) a Thus, the continuous state space equation for a vehicle following system pair can be expressed as:
wherein,
ahis the longitudinal acceleration of the vehicle, arefIs the desired longitudinal acceleration, τiIs the inner loop dynamic time constant, τ is the expected headway;
in the design of a controller, a discrete system expression form of a state space equation of a vehicle following system is obtained by using a zero-order holding method, namely ZOH, and a state quantity at the k moment is defined as
Control input u (k) ═ aref(k) Thus, at time k, one can obtain:
x(k+1)＝Ax(k)+Buu(k)+Bdd(k) (5)
wherein
d(k)＝ah(k)，TsIs the sampling time interval, τ is the desired headway, τiIs the inner loop dynamic time constant;
2.2) establishing the optimal cruise control problem as a linear quadratic optimization problem, wherein the specific method comprises the following steps:
2.2.1) establishing an optimal quadratic control problem;
the performance indicator function may be expressed as:
wherein,
is a semi-positive definite symmetric matrix and is,
is a positive definite symmetric matrix, x (k) is a state quantity at the moment k, u (k) is a control quantity at the moment k, and a parameter matrix Q, R represents the requirements of the controller on driving style, comfort and economy, and is defined as:
substituting it into equation (6), the performance indicator function can also be expressed as:
here, it is clear that the penalty term
The following performance is characterized, i.e. the vehicle reaches the steady state with the expected longitudinal directionKeeping an expected following distance with a front vehicle at the vehicle speed and punishing a term
Expectation of minimizing fuel consumption, penalty term
The comfort of the vehicle is expected to be improved, and frequent acceleration and deceleration operations are reduced;
2.2.2) solving the optimal quadratic control problem;
defining a discrete Hamiltonian equation to solve the optimization problem, and for solving the optimization problem equation (6), the discrete Hamiltonian equation is defined as:
wherein,
is a Lagrange multiplier, x (k), u (k), d (k), Q, R, A, Bu,BdAs previously mentioned and unchanged in the derivation of the following formula, the optimal solution that minimizes the hamiltonian h (k) according to the principle of minimization needs to be satisfied:
Qx(k)+ATλ(k+1)-λ(k)＝0 (10b)
due to the inverse matrix R of R-1Is represented by the formula (10 a):
the following form is selected here for λ (k):
λ(k)＝Px(k)+hd(x) (12)
wherein,
and
is a Hamiltonian matrix, and is obtained by substituting equations (11) and (12) into equation of state (5):
while substituting formula (12) for formula (10b) to obtain:
(Q-R)x(k)+ATPx(k+1)+AThd(k+1)-hd(k)＝0 (14)
substituting formula (13) for formula (14) to obtain:
since equation (15) holds for all x (k), the ricati equation is obtained:
due to the fact that
Ricatt equation (16) translates to:
since d (k +1) is unknown at the current time k without loss of generality, assuming that d (k +1) ═ d (k), equation (17) becomes
Thus, an explicit solution for h is;
wherein,
thus, by integrating the formulas (10b), (11), and (12), the optimal control law is:
wherein,
is the controller gain. From this analysis, it follows that the driving style can be differentiated by changing the controller gain.
2.2.3) preceding vehicle acceleration estimation considering uncertainty;
estimating the acceleration of the front vehicle by using the speed of the front vehicle and two first-order low-pass filters, and finally limiting the abnormal change of the acceleration by using a curvature limiter;
2.3) self-learning module of following driving style
Aiming at the characteristic that the driving styles of specific drivers are different, the driving styles of the specific drivers are learned based on a reinforcement learning method, so that the control parameters of the optimal cruise control problem are adjusted, and the function of a self-learning system is achieved; the specific method is as follows;
2.3.1) specific driver driving style definition;
based on a large number of statistical data analysis results of drivers, dividing the drivers into three categories of aggressive, steady and conservative by using a common driving style classification method; aggressive drivers tend to keep a close distance from the vehicle ahead, and accelerate and decelerate frequently; conservative drivers tend to stay farther away from the leading vehicle and experience less acceleration and deceleration; the system uses the variable headway time distance commonly used in the self-adaptive cruise system to represent the behavior style;
2.3.2) reinforcement learning method establishment
Because the actual driver style information is unknown during system design, the system learns the driving style of a specific driver by using a reinforcement learning method; when the Riccati equation is solved, the reinforcement learning is completed, and the system learns the driving style habit of the driver; in the linear quadratic problem in the linear discrete system, the Q function is a state action value function in reinforcement learning, is a quadratic form with respect to the state quantity and the control quantity, and can be expressed as:
wherein,
is the return in reinforcement learning, and P is the Riccatin solution; the system model formula (5) is substituted into the formula (21) to obtain
Wherein,
z(k)＝[xT(k),dT(k),uT(k)]Tfor subsequent differentiation convenience, a kernel matrix is defined:
according to the berman optimality principle, the optimal solution is the control quantity u (k) which enables the function value of the Q function to be minimum, namely:
therefore, the derivation of the formula (22)
Obtaining:
from the equation (22), system parameters A, B can be obtainedd,BuFor the display expression mode of the kernel matrix S, Q and R are solved by using a least square method in function approximation for the kernel matrix S through a learning method;
approximating the Q function using a linear kernel function, which can be written as:
wherein W ═ W1,w2,...,w15]TAs a weight vector, the weight vector is,
the method is a basis function, a Q function in a parameter approximation form is used, and for an infinite time domain γ equal to 1, a bellman equation in reinforcement learning can be written as follows:
gradually updating an approximate optimal value based on driver data through incremental learning, and carrying out approximate solution on 15 unknown parameters of which optimal weight vectors are unknown; converting the solution of the formula (27) into a least square method, namely LS problem solution; determining the number N of training samples in batches, carrying out iterative computation on samples at the moment k for j times, wherein m is an integer constant, and sequentially computing:
to obtain Wj+1The least squares estimate of (d) is:
after the LS algorithm is converged, the system compares the calculated gain vector of the controller with 2 norms of gain parameter vectors of three types of drivers given in the parameter tables of the drivers with different following vehicle behavior styles, and selects the most appropriate driving style type to update the parameters of the controller; otherwise, keeping the original controller parameters, and continuing to learn the next step;
the specific algorithm flow is as follows:
step (1): controller gain initialization weight vector W, N, m in driver parameter table based on different following behavior styles and characteristic parameter tau of each driving styleA,τM,τC,rA,rMAnd rC；
Step (2): updating and calculating to obtain the current data set
And (3): using the LS algorithm: updating the weight vector W by using a formula (29) to obtain Wj+1；
And (4): if the LS algorithm converges, W is obtainedj+1Then, the kernel matrix is updated: using weight vector Wj+1And (3) performing inverse solution on each parameter of the matrix S to obtain a controller gain estimation value:
comparing the parameters obtained by evaluation with 2 norms of three driving style parameter vectors, selecting a group of most suitable parameters in the driver parameter tables of different driving style, updating the parameters of the controller, and setting: j is 1, and the calculation is ended; returning to the following driving style of the vehicle;
and (5): otherwise, setting: j is j + 1; and (4) continuing to the steps (2) - (4).
The invention has the beneficial effects that:
1. establishing an optimal control problem framework aiming at the problem of the cruise system;
2. predicting and considering the acceleration of the front vehicle in the optimization problem, and avoiding unnecessary acceleration and braking to improve the comfort;
3. the control problem considers the driving style of the driver comprehensively, and the self-learning method is used for learning the driving style, so that the system can carry out the driving style self-learning aiming at each specific driver.
4. And the acceleration of the front vehicle is estimated by using two first-order low-pass filters, so that the prediction complexity is reduced.
Drawings
FIG. 1 is a block diagram of the system architecture of the present invention;
FIG. 2 is a schematic general flow diagram of the driving style self-learning cruise control system of the present invention;
FIG. 3 is a schematic diagram of a forward vehicle acceleration estimation module;
FIG. 4 is a graph showing following behavior of the system regardless of the driving style of the driver in the embodiment;
fig. 5 is a following behavior representation graph of the system after learning by considering the driving style of the driver in the embodiment.
Detailed Description
In order to adapt to higher requirements of development of automobile technology on an intelligent driving system, the system has self-learning capability on the driving style under the condition of ensuring the performance of the self-adaptive vehicle following system. The invention provides an adaptive cruise control system with driving style self-learning capability, which comprehensively considers two factors of vehicle following performance and driving style of a driver. Firstly, a second-order vehicle following model is established for the self-adaptive vehicle following problem, an optimization problem solution is established based on a quadratic optimization type control method, the driving style is expressed as unknown controller parameters in a vehicle following controller, in order to enable the system to have the self-learning capability for the driving style of specific driving, a function approximation method and a least square method are combined to solve the unknown system parameters, and the requirements of the system for drivers with different driving styles on the driving performance are effectively improved.
An adaptive cruise system with driving style learning capability, whose structural block diagram is shown in fig. 1, mainly includes: the system comprises a perception fusion module A, a following vehicle control module B, a following vehicle driving style self-learning module C and a vehicle execution control module D. The perception fusion module A adopts a scheme of radar camera fusion in an environment perception sensing part, confirms a front vehicle to be followed through a self fusion recognition algorithm, and obtains the running state information of the front vehicle, including the speed information and the relative distance of the front vehicle; and obtaining the information of the current vehicle such as speed, engine torque, braking deceleration and the like from the CAN network communication of the vehicle bus. The information is used by the vehicle following control module B-a and the vehicle following driving style self-learning module C. In the following control module B, a vehicle following model is established according to the confirmed running state information of the front vehicle, the front vehicle and the current vehicle, and meanwhile, a control problem is established, and an optimization target is determined. A preceding vehicle acceleration prediction module (see a module B-B in FIG. 1) for predicting the preceding vehicle acceleration according to the confirmed driving state information of the preceding vehicle and the current vehicle; and the following controller (such as a module B-a in fig. 1) is used for designing a vehicle following control model and an optimization solving algorithm according to the states and the prediction information so as to realize a self-adaptive following task for the front vehicle. A following driving style self-learning module (such as a module C shown in a figure 1) is designed aiming at the characteristic that the driving styles of specific drivers are different, the driving style of the specific drivers is defined and evaluated (such as a module C-a shown in a figure 1) to define the relevant parameters of controllers of the driving styles of the drivers with different driving styles, and the driving style attributes of the specific drivers are evaluated by using mean square deviation; the driving style learning module (such as the module C-b in fig. 1) is not known about the optimal cruise controller parameters for a specific driver, so that the system model parameters need to be learned by using a reinforcement learning method; based on specific driver data, a reinforced learning process is completed by solving the Riccati equation in a linear quadratic problem, and then control parameters of the optimal cruise control problem are adjusted, so that the function of a self-learning system is achieved. Fig. 2 shows a flowchart of the overall technical solution of the present invention, which is specifically implemented as follows: in a vehicle perception fusion module (such as a module A shown in figure 1), an environment perception sensing part adopts a scheme of radar camera fusion, confirms a vehicle to be followed in front through a self fusion recognition algorithm, and obtains running state information of the vehicle in front, wherein the running state information comprises speed information and relative distance of the vehicle in front; the information of the current vehicle speed, the engine torque, the braking deceleration and the like is obtained through the communication of a vehicle bus (CAN) network. In the case of system control, the preceding vehicle acceleration prediction submodule (such as modules B-B in FIG. 1) estimates the preceding vehicle acceleration; the following control module (such as a module B-a in figure 1) obtains an optimized control quantity according to the confirmed running state information of the front vehicle, the front vehicle and the current vehicle as well as the established vehicle following model and control problem, acts on the bottom vehicle execution control module (such as a module D in figure 1) to perform tracking control, and finally outputs a vehicle power driving system and a braking system to control the driving of the vehicle; under the condition that the system is not controlled, the specific driver driving style defining and evaluating module judges whether the driver is consistent with the current driving style parameters (such as a module C-a in figure 1), the following driving style self-learning module (such as a module C-b in figure 1) learns the driving style by using a learning method, and after the learning is converged, the system control parameters are updated to the following control module.
The design and the specific working process of each module of the self-adaptive cruise control system with the driving style self-learning capability are as follows:
1) perception fusion module A
The perception fusion module A adopts a scheme of radar camera fusion in an environment perception sensing part, confirms a front vehicle to be followed through a self fusion recognition algorithm, and obtains the running state information of the front vehicle, including the speed information and the relative distance of the front vehicle; the information of the current vehicle speed, the engine torque, the braking deceleration and the like is obtained through the communication of a vehicle bus (CAN) network. The information is used for the vehicle following module and the driving style self-learning module. The feasible installation scheme of the radar camera is that a product-level camera is additionally arranged on the inner wall of the front windshield, and the camera can output lane line information and front obstacle (vehicles, pedestrians, bicycles and the like) information through an internal image processing algorithm. A product-level millimeter wave radar is additionally arranged at the rear part of a vehicle grille and can be a 77Ghz forward long-distance radar, the radar can provide at least 40 original target information and can output 6 dangerous target queues after processing;
2) follow car control module B
The following sub-modules are provided, and the vehicle follows a control module (shown as a module B-a in FIG. 1) and a preceding vehicle acceleration prediction module (shown as a module B-B in FIG. 1). Establishing a vehicle following model, establishing a control problem, determining an optimized target and obtaining a following controller; a preceding vehicle acceleration prediction module that predicts a preceding vehicle acceleration based on the confirmed preceding vehicle and the traveling state information of the preceding vehicle and the current vehicle; the method comprises the following working processes:
2.1) vehicle following model establishment (as module B-a in FIG. 1): in the module, a vehicle following model is established according to the self-adaptive following problem requirement and the longitudinal kinematics and dynamics characteristics of the vehicle;
in order to describe the longitudinal dynamics of the vehicle in a following scene, the system introduces two state variables,
respectively as follows:
＝d-dref(1a)
where d is the relative distance between the vehicle and the front vehicle, drefIs referenced to the relative distance, dref＝d0+τvh，d0Is the safe parking distance, τ is the expected headway, vhIs the speed of the vehicle, vtIs the front vehicle speed. Therefore, the differential equation of the longitudinal dynamics of the vehicle in the following scene can use the above state variables,
represents:
wherein, ahIs the longitudinal acceleration of the vehicle, atIs the front vehicle longitudinal acceleration, τ is the desired headway. It is assumed here that the inner loop acceleration tracking control module of the host vehicle can be expanded to the first order approximation:
wherein, tauiIs the inner loop dynamic time constant, ahIs the longitudinal acceleration of the vehicle, arefIs the desired longitudinal acceleration.
In a continuous system, the state quantities are selected as
The controlled variable is u ═ arefModeling the acceleration of the front vehicle as system disturbance, namely disturbance quantity d ═ ah. Thus, the continuous state space equation for a vehicle following system pair can be expressed as:
wherein,
ahis the longitudinal acceleration of the vehicle, arefIs the desired longitudinal acceleration, τiIs the inner loop dynamic time constant, and τ is the desired headway.
Further in the controller design, a discrete system expression form of a state space equation of the vehicle following system can be obtained by using a zero-order hold method (ZOH), and the state quantity at the k moment is defined as
Control input u (k) ═ aref(k) Thus, at time k, one can obtain:
x(k+1)＝Ax(k)+Buu(k)+Bdd(k) (5)
wherein
d(k)＝ah(k)，TsIs the sampling time interval, τ is the desired headway, τiIs the inner loop dynamic time constant.
2.2) establishment of optimal cruise control problem
The safety, comfort and economy in the following performance of the vehicle and the driving style of a driver are taken into comprehensive consideration as multi-objective optimization indexes, and the optimal cruise control problem is established as a linear quadratic optimization problem.
2.2.1) optimal quadratic control problem establishment
In summary of the above, the performance indicator function can be expressed as:
wherein,
is a semi-positive definite symmetric matrix and is,
is a positive definite symmetric matrix. x (k) is a state quantity at time k, and u (k) is a control quantity at time k. The parameter matrix Q, R represents the controller's requirements for driving style, comfort and economy, which is defined here as:
substituting it into equation (5), the performance indicator function can also be expressed as:
here, it is clear that the penalty term
The following performance is characterized, namely, when the steady state is reached, the vehicle keeps a desired following distance with the front vehicle at a desired longitudinal speed. Penalty term
Expectation of minimizing fuel consumption, penalty term
It is desirable to improve the comfort of the vehicle and reduce frequent acceleration and deceleration operations.
2.2.2) solving of optimal quadratic control problem
Defining the discrete hamiltonian equation to solve the optimization problem, for solving the optimization problem equation (6), the discrete hamiltonian equation can be defined as:
wherein,
is a Lagrange multiplier, x (k), u (k), d (k), Q, R, A, Bu,BdAs previously mentioned and unchanged in the derivation of the following formula, the optimal solution that minimizes the hamiltonian h (k) according to the principle of minima needs to be satisfied:
Qx(k)+ATλ(k+1)-λ(k)＝0 (10b)
due to the inverse matrix R of R-1Is represented by the formula (10 a):
the following form is selected here for λ (k):
λ(k)＝Px(k)+hd(x) (12)
wherein,
and
is a hamiltonian matrix. Substituting equations (11), (12) into equation of state (5) yields:
while substituting formula (12) for formula (10b) to obtain:
(Q-R)x(k)+ATPx(k+1)+AThd(k+1)-hd(k)＝0 (14)
substituting formula (13) for formula (14) to obtain:
since equation (15) holds for all x (k), the ricati equation is obtained:
due to the fact that
Ricatt equation (16) translates to:
since d (k +1) is unknown at the current time k without loss of generality, assuming that d (k +1) ═ d (k), equation (17) becomes
Thus, an explicit solution for h is;
wherein,
thus, by integrating the formulas (10b), (11), and (12), the optimal control law is:
wherein,
is the controller gain. Thus analyzed to obtainThus, the driving style can be differentiated by changing the controller gain.
2.2.3) preceding vehicle acceleration estimation taking uncertainty into account
In the optimal following problem, the acceleration information of the preceding vehicle is an important consideration factor. It is therefore generally necessary to estimate the leading vehicle acceleration using the difference in velocity information. There are generally several ways: a direct difference method, but noise information affects estimation accuracy; a Kalman filtering method; vehicle-to-vehicle communication, but communication delay and information security problems exist. To simplify the problem, as shown in fig. 3, the vehicle speed of the preceding vehicle is used, the acceleration of the preceding vehicle is estimated by two first-order low-pass filters, and finally, a curvature limiter is used to limit the abnormal change of the acceleration.
2.3) self-learning module of following driving style
Aiming at the characteristic that the driving styles of specific drivers are different, the driving styles of the specific drivers are learned based on a reinforcement learning method, so that the control parameters of the optimal cruise control problem are adjusted, and the function of a self-learning system is achieved.
2.3.1) driver-specific Driving Style definition
Based on a large number of statistical data analysis results of drivers, the drivers are classified into three categories of aggressive, steady and conservative by using a common driving style classification method. Aggressive drivers tend to keep a close distance from the vehicle ahead, and accelerate and decelerate frequently; while conservative drivers tend to stay farther away from the leading vehicle and experience less acceleration and deceleration. The system uses the variable headway commonly used in adaptive cruise systems to characterize this style of behavior. Accordingly, the controller-related parameters of the present system are set accordingly as shown in table 1:
TABLE 1 driver parameters for different following behavior styles
2.3.2) reinforcement learning method establishment
Since the actual driver style information is not known at the time of system design, the system learns a particular driver driving style using a reinforcement learning method. When the Riccati equation is solved, the reinforcement learning is completed, and the system learns the driving style habit of the driver. In the linear quadratic problem in the linear discrete system, the Q function is a state action value function in reinforcement learning, is a quadratic form with respect to the state quantity and the control quantity, and can be expressed as:
wherein
Is the return in reinforcement learning, and P is the Riccatin solution. For x (k), u (k), d (k), Q, R, A, Bu,BdThe same symbols in this subsection still retain their original physical and numerical meanings. The system model formula (5) is substituted into the formula (21) to obtain
Wherein,
z(k)＝[xT(k),dT(k),uT(k)]T. For subsequent differentiation convenience, a kernel matrix is defined:
according to the berman optimality principle, the optimal solution is the control quantity u (k) which enables the function value of the Q function to be minimum, namely:
therefore, the derivation of the formula (22)
Obtaining:
from the equation (22), system parameters A, B can be obtainedd,BuAnd Q and R are used for solving the display expression mode of the kernel matrix S by using a least square method in function approximation through a learning method.
Approximating the Q function using a linear kernel function, which can be written as:
wherein W ═ W1,w2,...,w15]TAs a weight vector, the weight vector is,
the method is a basis function, a Q function in a parameter approximation form is used, and for an infinite time domain γ equal to 1, a bellman equation in reinforcement learning can be written as follows:
through incremental learning, the approximate optimal value is gradually updated based on the driver data, and 15 unknown parameters of which the optimal weight vectors are unknown are approximately solved. The solution of equation (27) is converted to a least squares problem solution (LS). Determining the number N of training samples in batches, carrying out iterative computation on samples at the moment k for j times, wherein m is an integer constant, and sequentially computing:
to obtain Wj+1The least squares estimate of (d) is:
after the LS algorithm is converged, the system compares the calculated controller gain vector with 2 norms of the given three types of driver style gain parameter vectors in the table 1, and selects the most appropriate driving style type to update the controller parameters; otherwise, keeping the original controller parameters and continuing to learn the next step.
Therefore, in the process of learning the driving style based on the driving style learning algorithm of reinforcement learning, the specific algorithm flow is as follows:
step (1): initializing weight vectors W, N, m based on the controller gains in Table 1, and characteristic parameters τ of the respective driving stylesA,τM,τC,rA,rMAnd rC；
Step (2): updating and calculating to obtain the current data set
And (3): using the LS algorithm: updating the weight vector W by using a formula (29) to obtain Wj+1；
And (4): if the LS algorithm converges, W is obtainedj+1Then, the kernel matrix is updated: using weight vector Wj+1And (3) performing inverse solution on each parameter of the matrix S to obtain a controller gain estimation value:
comparing the evaluated parameters with 2 norms of three driving style parameter vectors, selecting the most appropriate group of parameters in the table 1, updating the parameters of the controller, and setting: j equals 1 and the calculation ends. And returns the vehicle to follow the driving style.
And (5): otherwise, setting: j equals j + 1. And (4) continuing to the steps (2) - (4).
For the self-adaptive cruise control system with driving style self-learning capability and the implementation method, the effectiveness of the method is verified by using an actual system parameter, and the design parameters of a system controller are shown as q in formula (7)11＝0.15,q22＝0.73,q230.2, inner loop dynamic time constant τi0.9, safety distance d03, front vehicle acceleration estimation module parameter τ1＝0.4,τ20.2. After actually collected drivers are subjected to vehicle following style learning, the vehicle following performance of the drivers is finally verified in a test run, fig. 4 is the vehicle following performance of the system without considering the driving style of the drivers, and fig. 5 is the vehicle following performance of the system after the driving style of the drivers is considered and the learning is performed.
Claims (1)
1. A method for realizing an adaptive cruise system with driving style learning capability comprises a perception fusion module (A), a following control module (B), a driving style self-learning module (C) and a vehicle execution control module (D);
the perception fusion module (A) is used for obtaining the running state information of the front vehicle, including the speed information and the relative distance of the front vehicle;
the following control module (B) comprises a vehicle following control module (B-a) and a preceding vehicle acceleration predicting module (B-B), and the vehicle following control module (B-a) is used for establishing a vehicle following model, simultaneously establishing a control problem, determining an optimized target and obtaining a following controller; a preceding vehicle acceleration prediction module (B-B) for predicting a preceding vehicle acceleration on the basis of the confirmed preceding vehicle and the traveling state information of the preceding vehicle and the current vehicle;
the driving style self-learning module (C) is used for learning the driving style of a specific driver based on a reinforcement learning method aiming at the characteristic that the driving style of the specific driver is different, so as to adjust the control parameter of the optimal cruise control problem and achieve the function of a self-learning system;
the vehicle execution control module (D) is used for performing tracking control and finally outputting a vehicle power driving system and a braking system to control the driving of the vehicle;
the perception fusion module (A) is unidirectionally connected with the car following control module (B) and the driving style self-learning module (C); the following control module (B) is unidirectionally connected with the vehicle execution control module (D); the driving style self-learning module (C) is unidirectionally connected with the following vehicle control module (B); the vehicle execution control module (D) is connected with the vehicle in a one-way mode; the method is characterized by comprising the following steps:
firstly, confirming a front vehicle to be followed through a perception fusion module (A) self fusion recognition algorithm, and obtaining the running state information of the front vehicle, including the speed information and the relative distance of the front vehicle; and the information such as the speed, the engine moment, the braking deceleration and the like of the current vehicle are obtained from the communication of a vehicle bus, namely a CAN network;
step two, a vehicle following model is established through a vehicle following control module (B-a) in a vehicle following control module (B), a control problem is established at the same time, an optimization target is determined, and a vehicle following controller is obtained; a preceding vehicle acceleration prediction module (B-B) in the following vehicle control module (B), which uses the speed of the preceding vehicle, estimates the preceding vehicle acceleration through two first-order low-pass filters, and predicts the preceding vehicle acceleration by using a curvature limiter to limit the abnormal change of the acceleration;
step three, learning the driving style of a specific driver based on a reinforcement learning method aiming at the characteristic that the driving style of the specific driver has difference through a driving style self-learning module (C), and further adjusting the control parameter of the optimal cruise control problem to achieve the function of a self-learning system;
fourthly, tracking control is carried out through a bottom vehicle execution control module (D), and finally a vehicle power driving system and a braking system are output to control driving of the vehicle;
the specific method of the second step is as follows:
2.1) vehicle following model establishment: according to the self-adaptive following problem requirement and the longitudinal kinematics and dynamics characteristics of the vehicle, a vehicle following model is established;
in order to describe the longitudinal dynamics of the vehicle in a following scene, the system introduces two state variables,
respectively as follows:
＝d-dref(1a)
where d is the relative distance between the vehicle and the front vehicle, drefIs referenced to the relative distance, dref＝d0+τvh，d0Is the safe parking distance, τ is the expected headway, vhIs the speed of the vehicle, vtIs the front vehicle speed; therefore, the differential equation of the longitudinal dynamics of the vehicle in the following scene uses the state variables,
represents:
wherein, ahIs the longitudinal acceleration of the vehicle, atIs the front vehicle longitudinal acceleration, τ is the desired headway; it is assumed here that the inner loop acceleration tracking control module of the host vehicle is developed with a first order approximation as:
wherein, tauiIs the inner loop dynamic time constant, ahIs the longitudinal acceleration of the vehicle, arefIs the desired longitudinal acceleration;
in a continuous system, the state quantities are selected as
The controlled variable is u ═ arefModeling the acceleration of the front vehicle as system disturbance, namely disturbance quantity d ═ ah(ii) a Thus, the continuous state space equation for a vehicle following system pair is expressed as:
wherein,
ahis the longitudinal acceleration of the vehicle, arefIs the desired longitudinal acceleration, τiIs the inner loop dynamic time constant, τ is the expected headway;
in the design of a controller, a discrete system expression form of a state space equation of a vehicle following system is obtained by using a zero-order holding method, namely ZOH, and a state quantity at the k moment is defined as
Control input u (k) ═ aref(k) Thus, at time k, we get:
x(k+1)＝Ax(k)+Buu(k)+Bdd(k) (5)
wherein
d(k)＝ah(k)，TsIs the sampling time interval, τ is the desired headway, τiIs the inner loop dynamic time constant;
2.2) establishing the optimal cruise control problem as a linear quadratic optimization problem, wherein the specific method comprises the following steps:
2.2.1) establishing an optimal quadratic control problem;
the performance index function is expressed as:
wherein,
is a semi-positive definite symmetric matrix and is,
is a positive definite symmetric matrix, x (k) is a state quantity at the moment k, u (k) is a control quantity at the moment k, and parametersThe matrix Q, R represents the controller's requirements for driving style, comfort and economy, defined here as:
substituting it into equation (6), the performance indicator function is expressed as:
here, it is clear that the penalty term
The following performance is characterized, namely when the vehicle reaches a steady state, the vehicle keeps an expected following distance with a front vehicle at an expected longitudinal speed, and a penalty term
Expectation of minimizing fuel consumption, penalty term
The comfort of the vehicle is expected to be improved, and frequent acceleration and deceleration operations are reduced;
2.2.2) solving the optimal quadratic control problem;
defining a discrete Hamiltonian equation to solve the optimization problem, and for solving the optimization problem equation (6), the discrete Hamiltonian equation is defined as:
wherein,
is a Lagrange multiplier, x (k), u (k), d (k), Q, R, A, Bu,BdAs previously described and not changed during the derivation of the following formulae, according to the principle of minimaThe minimum optimal solution of the Hamiltonian H (k) needs to satisfy:
Qx(k)+ATλ(k+1)-λ(k)＝0 (10b)
due to the inverse matrix R of R-1From (10a) there can be obtained:
the following form is selected here for λ (k):
λ(k)＝Px(k)+hd(x) (12)
wherein,
and
is a Hamiltonian matrix, and is obtained by substituting equations (11) and (12) into equation of state (5):
while substituting formula (12) for formula (10b) to obtain:
(Q-R)x(k)+ATPx(k+1)+AThd(k+1)-hd(k)＝0 (14)
substituting formula (13) for formula (14) to obtain:
since equation (15) holds for all x (k), the ricati equation is obtained:
due to the fact that
Ricatt equation (16) translates to:
since d (k +1) is unknown at the current time k without loss of generality, assuming that d (k +1) ═ d (k), equation (17) becomes
Thus, an explicit solution for h is;
wherein,
thus, by integrating the formulas (10b), (11), and (12), the optimal control law is:
wherein,
the method comprises the steps of (1) changing the value of the controller gain to distinguish the driving style;
2.2.3) preceding vehicle acceleration estimation considering uncertainty;
estimating the acceleration of the front vehicle by using the speed of the front vehicle and two first-order low-pass filters, and finally limiting the abnormal change of the acceleration by using a curvature limiter;
2.3) self-learning module of following driving style
Aiming at the characteristic that the driving styles of specific drivers are different, the driving styles of the specific drivers are learned based on a reinforcement learning method, so that the control parameters of the optimal cruise control problem are adjusted, and the function of a self-learning system is achieved; the specific method is as follows;
2.3.1) specific driver driving style definition;
based on a large number of statistical data analysis results of drivers, dividing the drivers into three categories of aggressive, steady and conservative by using a common driving style classification method; aggressive drivers tend to keep a close distance from the vehicle ahead, and accelerate and decelerate frequently; conservative drivers tend to stay farther away from the leading vehicle and experience less acceleration and deceleration; the system uses the variable headway time distance commonly used in the self-adaptive cruise system to represent the behavior style;
2.3.2) reinforcement learning method establishment
Because the actual driver style information is unknown during system design, the system learns the driving style of a specific driver by using a reinforcement learning method; when the Riccati equation is solved, the reinforcement learning is completed, and the system learns the driving style habit of the driver; in the linear quadratic problem in the linear discrete system, the Q function is a state action value function in reinforcement learning, is a quadratic form with respect to a state quantity and a control quantity, and is expressed as:
wherein,
is the return in reinforcement learning, and P is the Riccatin solution; the system model formula (5) is substituted into the formula (21) to obtain
Wherein,
z(k)＝[xT(k),dT(k),uT(k)]Tfor subsequent differentiation convenience, a kernel matrix is defined:
according to the berman optimality principle, the optimal solution is the control quantity u (k) which enables the function value of the Q function to be minimum, namely:
therefore, the derivation of the formula (22)
Obtaining:
from the formula (22), system parameters A, B are obtainedd,BuFor the display expression mode of the kernel matrix S, Q and R are solved by using a least square method in function approximation for the kernel matrix S through a learning method;
approximating the Q function using a linear kernel function, which can be written as:
wherein W ═ W1,w2,...,w15]TAs a weight vector, the weight vector is,
the method is a basis function, a Q function in a parameter approximation form is used, and for an infinite time domain γ equal to 1, a bellman equation in reinforcement learning is written as follows:
gradually updating an approximate optimal value based on driver data through incremental learning, and carrying out approximate solution on 15 unknown parameters of which optimal weight vectors are unknown; converting the solution of the formula (27) into a least square method, namely LS problem solution; determining the number N of training samples in batches, carrying out iterative computation on samples at the moment k for j times, wherein m is an integer constant, and sequentially computing:
to obtain Wj+1The least squares estimate of (d) is:
after the LS algorithm is converged, the system compares the calculated gain vector of the controller with 2 norms of gain parameter vectors of three types of drivers given in the parameter tables of the drivers with different following vehicle behavior styles, and selects the most appropriate driving style type to update the parameters of the controller; otherwise, keeping the original controller parameters, and continuing to learn the next step;
the specific algorithm flow is as follows:
step (1): controller gain initialization weight vector W, N, m in driver parameter table based on different following behavior styles and characteristic parameter tau of each driving styleA,τM,τC,rA,rMAnd rC；
Step (2): updating and calculating to obtain the current data set
And (3): using the LS algorithm: updating the weight vector W by using a formula (29) to obtain Wj+1；
And (4): if the LS algorithm converges, W is obtainedj+1Then, the kernel matrix is updated: using weight vector Wj+1And (3) performing inverse solution on each parameter of the matrix S to obtain a controller gain estimation value:
comparing the parameters obtained by evaluation with 2 norms of three driving style parameter vectors, selecting a group of most suitable parameters in the driver parameter tables of different driving style, updating the parameters of the controller, and setting: j is 1, and the calculation is ended; returning to the following driving style of the vehicle;
and (5): otherwise, setting: j is j + 1; and (4) continuing to the steps (2) - (4).
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910077516.9A CN109927725B (en) | 2019-01-28 | 2019-01-28 | Self-adaptive cruise system with driving style learning capability and implementation method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910077516.9A CN109927725B (en) | 2019-01-28 | 2019-01-28 | Self-adaptive cruise system with driving style learning capability and implementation method |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN109927725A CN109927725A (en) | 2019-06-25 |
| CN109927725B true CN109927725B (en) | 2020-11-03 |
Family
ID=66985173
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910077516.9A Active CN109927725B (en) | 2019-01-28 | 2019-01-28 | Self-adaptive cruise system with driving style learning capability and implementation method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109927725B (en) |
Families Citing this family (37)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110341710B (en) * | 2019-07-02 | 2021-07-23 | 浙江吉利控股集团有限公司 | Self-adaptive cruise control method, device and equipment |
| DE112019007598B4 (en) * | 2019-09-05 | 2024-05-08 | Mitsubishi Electric Corporation | INFERENCE DEVICE, DEVICE CONTROL SYSTEM AND LEARNING DEVICE |
| CN110568758B (en) * | 2019-09-12 | 2022-07-01 | 中汽研（天津）汽车工程研究院有限公司 | Parameter self-adaptive transverse motion LQR control method for automatically driving automobile |
| CN110949386B (en) * | 2019-11-28 | 2021-07-20 | 江苏大学 | Vehicle adaptive cruise control system and method capable of recognizing driving tendency |
| CN111267847B (en) * | 2020-02-11 | 2021-08-17 | 吉林大学 | Personalized self-adaptive cruise control system |
| CN111452630A (en) * | 2020-04-14 | 2020-07-28 | 江西精骏电控技术有限公司 | New energy automobile motor controller and control method thereof |
| CN111474934A (en) * | 2020-04-26 | 2020-07-31 | 东风汽车集团有限公司 | Personalized cruise control system based on high-precision map |
| CN111738294B (en) * | 2020-05-21 | 2024-05-14 | 深圳海普参数科技有限公司 | AI model training method, AI model using method, computer device, and storage medium |
| CN111547064B (en) * | 2020-05-26 | 2022-07-12 | 吉林大学 | Driving style recognition and classification method for automobile adaptive cruise system |
| CN111897213A (en) * | 2020-06-18 | 2020-11-06 | 中山大学 | Automobile self-adaptive cruise control method under model uncertainty |
| CN112033429B (en) * | 2020-09-14 | 2022-07-19 | 吉林大学 | Target-level multi-sensor fusion method for intelligent automobile |
| CN112158200B (en) * | 2020-09-25 | 2022-04-01 | 厦门大学 | Intelligent electric vehicle following control system and method based on driver characteristics |
| CN112172813B (en) * | 2020-10-14 | 2022-03-04 | 长安大学 | Car following system and method for simulating driving style based on deep inverse reinforcement learning |
| CN112109708B (en) * | 2020-10-26 | 2023-07-14 | 吉林大学 | Self-adaptive cruise control system considering driving behavior and control method thereof |
| CN112466119B (en) * | 2020-11-26 | 2021-11-16 | 清华大学 | Method and system for predicting vehicle following speed of vehicle by using vehicle-road cooperative data |
| CN112677982B (en) * | 2020-12-25 | 2022-06-14 | 重庆大学 | Vehicle longitudinal speed planning method based on driver characteristics |
| CN112721949B (en) * | 2021-01-12 | 2022-07-12 | 重庆大学 | Method for evaluating longitudinal driving personification degree of automatic driving vehicle |
| CN112896161B (en) * | 2021-02-08 | 2022-06-21 | 杭州电子科技大学 | Electric automobile ecological self-adaptation cruise control system based on reinforcement learning |
| CN112937592B (en) * | 2021-03-23 | 2022-04-15 | 武汉理工大学 | Method and system for identifying driving style based on headway |
| CN113406955B (en) * | 2021-05-10 | 2022-06-21 | 江苏大学 | Complex network-based automatic driving automobile complex environment model, cognitive system and cognitive method |
| CN113060146B (en) * | 2021-05-12 | 2023-04-07 | 中国第一汽车股份有限公司 | Longitudinal tracking control method, device, equipment and storage medium |
| CN113232652B (en) * | 2021-06-16 | 2022-10-25 | 武汉光庭信息技术股份有限公司 | Vehicle cruise control method and system based on kinematics model |
| CN113428164B (en) * | 2021-07-21 | 2023-01-03 | 上汽通用五菱汽车股份有限公司 | Driving habit learning method and device and computer readable storage medium |
| CN113401125B (en) * | 2021-07-29 | 2022-10-11 | 中国第一汽车股份有限公司 | Longitudinal car following control method and device, electronic equipment and storage medium |
| CN114167406B (en) * | 2021-11-24 | 2024-06-18 | 英博超算(南京)科技有限公司 | System for reducing false alarm of radar through fusion |
| CN114132333B (en) * | 2021-12-14 | 2024-08-16 | 阿维塔科技(重庆)有限公司 | Intelligent driving system optimization method and device and computer readable storage medium |
| CN114435379B (en) * | 2022-01-07 | 2023-06-20 | 所托(杭州)汽车智能设备有限公司 | Economical driving control method and device for vehicle |
| CN114030472B (en) * | 2022-01-10 | 2022-05-20 | 智道网联科技(北京)有限公司 | Control method, device and equipment for adaptive cruise and readable storage medium |
| CN114488799B (en) * | 2022-01-11 | 2024-02-23 | 吉林大学 | Parameter optimization method for controller of automobile self-adaptive cruise system |
| CN114312777A (en) * | 2022-01-17 | 2022-04-12 | 北京格睿能源科技有限公司 | Fuel cell heavy-truck predictive cruise control method and system |
| CN114492043B (en) * | 2022-01-27 | 2023-12-19 | 吉林大学 | Personalized driver following modeling method considering perception limited characteristics |
| CN114771520B (en) * | 2022-03-31 | 2024-08-02 | 中南大学 | Electric automobile economical self-adaptive cruise control method and system based on reinforcement learning |
| CN114879644B (en) * | 2022-05-26 | 2024-08-16 | 吉林大学 | Method for rapidly calibrating parameters of controller of automobile self-adaptive cruise system |
| CN116052412B (en) * | 2022-11-23 | 2023-08-18 | 兰州大学 | Automatic driving vehicle control method integrating physical information and deep reinforcement learning |
| DE102023201074A1 (en) | 2023-02-09 | 2024-08-14 | Volkswagen Aktiengesellschaft | Method for assisting a driver in controlling a motor vehicle and motor vehicle |
| CN117037524B (en) * | 2023-09-26 | 2023-12-22 | 苏州易百特信息科技有限公司 | Lane following optimization method and system under intelligent parking scene |
| CN117644870B (en) * | 2024-01-30 | 2024-03-26 | 吉林大学 | Driving anxiety detection and vehicle control method and system based on context awareness |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107544518A (en) * | 2017-10-17 | 2018-01-05 | 芜湖伯特利汽车安全系统股份有限公司 | The ACC/AEB systems and vehicle driven based on personification |
| CN107832517A (en) * | 2017-11-01 | 2018-03-23 | 合肥创宇新能源科技有限公司 | ACC lengthwise movement modeling methods based on relative motion relation |
Family Cites Families (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US7774121B2 (en) * | 2007-07-31 | 2010-08-10 | Gm Global Technology Operations, Inc. | Curve speed control system with adaptive map preview time and driving mode selection |
| US8260515B2 (en) * | 2008-07-24 | 2012-09-04 | GM Global Technology Operations LLC | Adaptive vehicle control system with driving style recognition |
| DE102010028140A1 (en) * | 2010-04-23 | 2013-03-21 | Zf Lenksysteme Gmbh | Driver assistance method of motor vehicle, involves performing automatic course correction such that vehicle departure from currently traveled lane is prevented, while compensating movement of steering handle of steering system |
| CN104925057A (en) * | 2015-06-26 | 2015-09-23 | 武汉理工大学 | Automotive self-adaptive cruising system with multi-mode switching system and control method thereof |
| DE102016205260A1 (en) * | 2016-03-31 | 2017-10-05 | Bayerische Motoren Werke Aktiengesellschaft | Method for automatically adjusting the speed of a vehicle |
| CN107499262A (en) * | 2017-10-17 | 2017-12-22 | 芜湖伯特利汽车安全系统股份有限公司 | ACC/AEB systems and vehicle based on machine learning |
| CN108944930B (en) * | 2018-07-05 | 2020-04-21 | 合肥工业大学 | Automatic car following method and system for simulating driver characteristics based on LSTM |
-
2019
- 2019-01-28 CN CN201910077516.9A patent/CN109927725B/en active Active
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107544518A (en) * | 2017-10-17 | 2018-01-05 | 芜湖伯特利汽车安全系统股份有限公司 | The ACC/AEB systems and vehicle driven based on personification |
| CN107832517A (en) * | 2017-11-01 | 2018-03-23 | 合肥创宇新能源科技有限公司 | ACC lengthwise movement modeling methods based on relative motion relation |
Also Published As
| Publication number | Publication date |
|---|---|
| CN109927725A (en) | 2019-06-25 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109927725B (en) | Self-adaptive cruise system with driving style learning capability and implementation method | |
| CN109624986B (en) | Driving style learning cruise control system and method based on mode switching | |
| CN110568760B (en) | Parameterized learning decision control system and method suitable for lane changing and lane keeping | |
| CN107808027B (en) | Self-adaptive car following method based on improved model predictive control | |
| CN112162555B (en) | Vehicle control method based on reinforcement learning control strategy in hybrid vehicle fleet | |
| CN111284489B (en) | Intelligent networked automobile random prediction cruise control system | |
| CN109910909B (en) | Automobile track internet interactive prediction method for multi-automobile motion state | |
| JP2023504223A (en) | Adaptive control of automated or semi-autonomous vehicles | |
| McKinnon et al. | Learn fast, forget slow: Safe predictive learning control for systems with unknown and changing dynamics performing repetitive tasks | |
| US11579574B2 (en) | Control customization system, control customization method, and control customization program | |
| Chu et al. | Self-learning optimal cruise control based on individual car-following style | |
| Kuutti et al. | End-to-end reinforcement learning for autonomous longitudinal control using advantage actor critic with temporal context | |
| CN113874865A (en) | Method and device for determining model parameters of a control strategy of a technical system by means of a Bayesian optimization method | |
| CN113791615A (en) | Hybrid vehicle queue distributed model prediction control method | |
| Shi et al. | Offline reinforcement learning for autonomous driving with safety and exploration enhancement | |
| Ure et al. | Enhancing situational awareness and performance of adaptive cruise control through model predictive control and deep reinforcement learning | |
| CN115598983B (en) | Unmanned vehicle transverse and longitudinal cooperative control method and device considering time-varying time delay | |
| CN111391831A (en) | Automobile following speed control method and system based on preceding automobile speed prediction | |
| Lim et al. | Automatic weight determination in model predictive control for personalized car-following control | |
| Gharib et al. | Multi-objective optimization of a path-following MPC for vehicle guidance: A Bayesian optimization approach | |
| CN110456790B (en) | Intelligent networking electric automobile queue optimization control method based on adaptive weight | |
| Li et al. | Reinforcement learning based lane change decision-making with imaginary sampling | |
| CN115352443B (en) | Self-adaptive cruise control method and device based on by-pass vehicle cut-in recognition | |
| CN114347998B (en) | Vehicle auxiliary driving control method, system, equipment and medium | |
| CN113635900B (en) | Channel switching decision control method based on energy management in predicted cruising process |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |