阅读说明：本技术 一种基于通信延迟的智能车队纵向跟随控制方法 (Intelligent motorcade longitudinal following control method based on communication delay ) 是由 雷利利 王梓 于 2021-06-30 设计创作，主要内容包括：本发明涉及智能车队纵向跟随控制技术领域,具体涉及一种基于通信延迟的车队纵向跟随控制方法。采用模型预测控制(MPC)设计编队的控制律,实现理想通信情况下的编队保持。然后引入通信延迟,利用卡尔曼滤波算法(KF)对领航车实际状态信息进行预测,并将状态预测误差作为对通信延迟的补偿,既考虑了编队的稳定性,同时又兼顾了车辆编队的控制效果。(The invention relates to the technical field of intelligent motorcade longitudinal following control, in particular to a motorcade longitudinal following control method based on communication delay. And designing a control law of formation by adopting Model Predictive Control (MPC), and realizing formation maintenance under an ideal communication condition. And then introducing communication delay, predicting actual state information of the pilot vehicle by using a Kalman filtering algorithm (KF), and compensating the communication delay by using a state prediction error, thereby not only considering the stability of formation, but also considering the control effect of vehicle formation.)

1. A communication delay-based intelligent fleet longitudinal following control method is characterized by comprising the following specific steps: establishing an upper-layer control kinematic equation based on a model prediction control principle; establishing a Kalman filtering model, and optimizing a vehicle state space model under the influence of communication delay by adopting a delay error compensation method to obtain an optimized vehicle state space model; and (4) performing MPC algorithm solution based on the optimized vehicle state space model to obtain expected control input, establishing a lower layer controller, and feeding the control input back to the controlled vehicle to realize acceleration or braking of the vehicle.

2. The intelligent fleet longitudinal following control method according to claim 1, wherein said step of building said upper control kinematics equation based on model predictive control principle comprises:

the longitudinal motion process of the intelligent vehicles in the fleet is regarded as a nonlinear third-order model and is described by the following differential equation:

wherein s is a longitudinal position measured by the intelligent vehicle from an inertial reference position, v and a are the speed and the acceleration of the vehicle respectively, and eta is the control input of the engine;

and (3) linearizing the feedback of the original nonlinear model to obtain:

η＝ma_des+C_dv²+d_m+2τC_dva；

wherein a is_desI.e. the desired acceleration, C, determined by the upper level controller_dRepresents the aerodynamic drag coefficient, d_mτ is the time constant of the vehicle engine;

consider a identically configured smart vehicle in a fleet of vehicles traveling on a straight road, where three closely follow_iRepresenting the distance, v, of the ith vehicle from an inertial reference point_i、a_iRespectively representing the speed and the acceleration of the ith vehicle, and l representing the length of the vehicle body, wherein the inter-vehicle distance error between the ith vehicle and the previous vehicle is as follows:

e_s＝s_i-1-s_i-d_i，des-l；

wherein d is_i，aesThe expected distance between the ith vehicle and the preceding vehicle is selected, how to select the expected distance depends on a safety distance algorithm, namely a vehicle distance strategy, and different vehicle distance strategies have different effects;

considering the stability of the fleet and the traffic flow, the safe distance algorithm with variable headway is adopted as a control research method, and the distance error between the ith vehicle and the preceding vehicle is shown as the following formula, wherein the expected distance is as follows:

d_i，des＝d₀+h_i(v_i)*v_i；

wherein d is₀Represents the minimum safe distance, h_iThe time interval of the locomotive;

the speed difference between the front vehicle and the front vehicle is as follows:

e_v＝v_i-1-v_i；

wherein v is_i-1Representing the speed of the i-1 st vehicle;

the state space model of the vehicle in the intelligent fleet longitudinal following control system based on the communication delay is derived as follows:

in the above formula, the first and second carbon atoms are,

x＝[e_s e_v a_i v_i]^Tu＝a_desω＝a_i-1；

where x, u, and ω are the state, control input, and disturbance input, respectively.

3. The intelligent fleet longitudinal following control method based on communication delay of claim 1, wherein a kalman filter model is established, and a delay error compensation method is adopted to optimize a vehicle state space model under the influence of communication delay, so as to obtain the optimized vehicle state space model, the method comprises the following specific steps:

assuming that the following vehicle displacement is s_iThe vehicle speed is v_iLongitudinal acceleration of a_iThe displacement of the front vehicle is s_i-1The vehicle speed is v_i-1Longitudinal acceleration of a_i-1；

Assuming that two vehicles run at a constant speed change between a certain time k and a time k +1, that is, the longitudinal acceleration value is a certain value between the times, and the time interval is set to be T, the relationship equation of the longitudinal parameter between the time k and the time k +1 can be listed according to the relative position relationship of the two vehicles, as shown in the following formula:

v_i-1(k+1)＝v_i-1(k)+a_i-1(k)T；

a_i-1(k+1)＝a_i-1(k)；

selecting front vehicle displacement s_i-1Speed v of the preceding vehicle_i-1Longitudinal acceleration a of the front vehicle_i-1Is the state variable X (k) and the observation variable Z (k) of the system, i.e. X (k) Z (k) s_i-1(k)，v_i-1(k)，a_i-1(k)]^T；

Setting the process noise matrix and the measurement noise matrix of the system as W and V respectively, and setting the variance as Q and R, and then obtaining the following formula:

X(k+1)＝DX(k)+W；

Z(k)＝HX(k)+V；

after a state space equation of the system is obtained, a recursion equation set for estimating the state of the front vehicle by adopting a Kalman filtering method can be obtained, and the new prediction equation is as follows:

and (3) state prediction:

and (3) covariance prediction:

P(k+1|k)＝DP(k|k)D^T+Q；

filtering gain:

K(k+1)＝P(k+1|k)H^T[HP(k+1|k)H^T+R]^-1；

and (3) updating the state:

and (3) covariance updating:

P(k+1|k+1)＝[I_n-K(k+1)H]P(k+1|k)；

by utilizing a filtering calculation prototype, all following vehicles can predict the state of the received front vehicle information in one step; assuming that state information of a vehicle ahead at the time k and at the time s is received by a following vehicle observer, and the difference between the time k and the time s is N filtering cycle durations T; obtaining an actual one-step state prediction recurrence sequence of s time → k time by a recurrence mode, wherein N is k-s:

is provided withFor the one-step state prediction of the system without time delay, the one-step state prediction error is as follows:

estimating the state before the s moment, and enabling:

then there isThe information of a system state transition matrix D and an s-moment system containing N-1 filtering cycles at the s → k moment;

the final calculated state prediction difference value is set as:

compensated front vehicle displacementAnd the speed of the front vehicleTo achieve the purpose.

Will be provided withAndreplacing s in interval error formula and speed difference formula_i-1And v_i-1Obtaining a compensated pitch error e_sSum velocity difference e_v：

4. The intelligent fleet longitudinal following control method according to claim 1, wherein said method comprises the steps of performing MPC algorithm solution based on an optimized vehicle state space model to obtain a desired control input, and establishing a lower controller to feed the control input back to the controlled vehicle, wherein said steps of accelerating or braking the vehicle are as follows:

the following discrete linear state space model is obtained after conversion:

y(k)＝Cx(k)；

whereinWhich is indicative of the state of the system,which represents the output that the system can measure,for controlling the inputIs an interference input; assuming that the state vector and the interference vector are in each sampling period t_sThe interference of the k time to the future time is predicted as the interference itself, namely, ω (k + j | k) ═ ω (k); the prediction of the future N sampling instants can be achieved by the following iterative model:

wherein N is called the prediction time domain;

the performance indicator function is defined as:

wherein y is_ref(k + j | k) is a reference trajectory which is related to the output measurement value up to the current time, or a predetermined trajectory, signRepresenting a quadratic function, Q, R being the error and input weighting matrices, respectively; the above formula shows that the performance index mainly considers two aspects of the error magnitude of the output and the reference and the energy magnitude of the input, and can be rewritten into the following vector form:

defining vector e (k) as the deviation of the system free response from the future target trajectory:

E(k)＝Y_ref(k)-M_xx(k)-M_uu(k-1)-M_ωω(k)；

the standard form of quadratic programming available:

and establishing a lower layer controller for controlling the acceleration and braking of the automobile:

when the automobile is accelerated, the expected torque of the engine is calculated according to the expected acceleration of the inverse longitudinal dynamics model, and then the expected accelerator opening is obtained through an inverse table look-up method by combining the rotating speed information of the engine; acceleration control the relationship between the desired engine torque and the desired acceleration is:

according to the engine torque characteristic of the vehicle, the engine speed omega can be known_eAnd a desired torque T_e，desThe corresponding desired throttle opening is obtained under the conditions of:

a_des＝f(T_e，des，ω_e)；

when an automobile is braked, an expected braking force is required to be obtained according to an expected acceleration, then an expected braking pressure is obtained according to the expected braking force, the automobile has no accelerator input during braking, and a vehicle dynamic equation in the braking process can be written as follows:

ma_des＝-F_bdes-F_areo-R_x；

the desired brake pressure may be found as:

Technical Field

The invention relates to the technical field of intelligent motorcade longitudinal following control, in particular to a motorcade longitudinal following control method based on communication delay.

Background

Vehicle formation control is taken as a research hotspot in the field of intelligent transportation, and is expected to reduce traffic accidents, improve traffic capacity and save energy. The vehicle formation control mainly adopts a layered design during design, an upper layer controller calculates the expected self-vehicle acceleration through a control algorithm, and a lower layer controller converts the expected acceleration of the upper layer into the accelerator opening and the brake pedal pressure through an inverse longitudinal dynamics model and then inputs the accelerator opening and the brake pedal pressure into a controlled vehicle model, so that the longitudinal control of the vehicle is realized.

At present, a plurality of domestic and foreign documents for researching vehicle formation control consider the influence of delay on vehicle formation control, and according to the proposed control method, the capability of a vehicle fleet for bearing the delay influence is improved to a certain extent. However, most of these studies are based on a certain model to obtain a consistency convergence condition (system stability), and do not consider the control effect of the system. The invention provides a formation control method based on time delay compensation, which comprises the following steps: and designing a control law of formation by adopting Model Predictive Control (MPC), and realizing formation maintenance under an ideal communication condition. And then introducing communication delay, predicting the state information of the pilot vehicle by using a Kalman filtering algorithm (KF), and compensating the communication delay by using a state estimation error so as to improve the control effect of vehicle formation.

Disclosure of Invention

The invention provides a communication delay-based intelligent fleet longitudinal following control research, aiming at ensuring the stability and control effect of a fleet under the influence of communication delay.

In order to achieve the above purpose, the invention provides the following technical scheme:

a smart fleet longitudinal follow control based on communication delay, comprising:

establishing an upper-layer control kinematic equation based on a model prediction control principle;

establishing a Kalman filtering model, and optimizing a vehicle state space model under the influence of communication delay by adopting a delay error compensation method to obtain an optimized vehicle state space model;

and (4) performing MPC algorithm solution based on the optimized vehicle state space model to obtain expected control input, establishing a lower layer controller, and feeding the control input back to the controlled vehicle to realize acceleration or braking of the vehicle.

Firstly, establishing an upper-layer control kinematic equation based on a model prediction control principle:

the invention takes the longitudinal motion process of the intelligent vehicle in the fleet as a nonlinear third-order model, which can be described by the following differential equation:

where s is the longitudinal position of the smart vehicle measured from the inertial reference position, v, a are the speed and acceleration of the vehicle, respectively, and η is the control input to the engine.

And (3) linearizing the feedback of the original nonlinear model to obtain:

η＝ma_des+C_dv²+d_m+2τC_dva；

wherein a is_desI.e. the desired acceleration, C, determined by the upper level controller_dRepresents the aerodynamic drag coefficient, d_mτ is the time constant of the vehicle engine, which is the mechanical resistance of the vehicle.

The invention considers intelligent vehicles of the same configuration, s, in which three closely follow, in a fleet of vehicles running on a straight road_iRepresenting the distance, v, of the ith vehicle from an inertial reference point_i、a_iRespectively representing the speed and the acceleration of the ith vehicle, and l representing the length of the vehicle body, wherein the inter-vehicle distance error between the ith vehicle and the previous vehicle is as follows:

e_s＝s_i-1-s_i-d_i，des-l

wherein d is_i，desThe expected distance between the ith vehicle and the preceding vehicle is selected, how to select the expected distance depends on a safety distance algorithm, namely a vehicle distance strategy, and different vehicle distance strategies have different effects.

Considering the stability of the motorcade and the traffic flow, the invention adopts the safe distance algorithm with variable headway as a control research method, and the distance error between the ith vehicle and the preceding vehicle is shown as the following formula, wherein the expected distance is as follows:

d_i，des＝d₀+h_i(v_i)*v_i

wherein d is₀Represents the minimum safe distance, h_iThe headway is the headway.

The speed difference between the front vehicle and the front vehicle is as follows:

e_v＝v_i-1-v_i

wherein v is_i-1The speed of the i-1 th vehicle is indicated.

The state space model of the vehicle in the intelligent fleet longitudinal following control system based on the communication delay is derived as follows:

in the above formula, the first and second carbon atoms are,

x＝[e_s e_v a_i v_i]^T u＝a_desω＝a_i-1；

where x, u, and ω are the state, control input, and disturbance input, respectively.

The second step is that: and establishing a Kalman filtering model, and optimizing the vehicle state space model under the influence of communication delay by adopting a delay error compensation method to obtain the optimized vehicle state space model.

Assuming that the following vehicle displacement is s_iThe vehicle speed is v_iLongitudinal acceleration of a_iThe displacement of the front vehicle is s_i-1The vehicle speed is v_i-1Longitudinal acceleration of a_i-1。

The two vehicles are assumed to travel at a constant speed change between a certain time k and a time k +1, that is, the longitudinal acceleration value is constant between the times, and the time interval is set to be T. From the relative position relationship of the two vehicles, the relationship equation between the longitudinal parameter at the time k and the time k +1 can be listed, as shown in the following formula:

v_i-1(k+1)＝v_i-1(k)+a_i-1(k)T

a_i-1(k+1)＝a_i-1(k)

selecting front vehicle displacement s_i-1Speed v of the preceding vehicle_i-1Longitudinal acceleration a of the front vehicle_i-1Is the state variable X (k) and the observation variable Z (k), namely X (k) Z (k) s_i-1(k),v_i-1(k),a_i-1(k)]^T。

Let the process and measurement noise matrices of the system be W and V, respectively, with variances of Q and R. Then the formula is finished to obtain:

X(k+1)＝DX(k)+W

Z(k)＝HX(k)+V

after a state space equation of the system is obtained, a recursion equation set for estimating the state of the front vehicle by adopting a Kalman filtering method can be obtained, and the new prediction equation is as follows:

and (3) state prediction:

and (3) covariance prediction:

P(k+1|k)＝DP(k|k)D^T+Q

filtering gain:

K(k+1)＝P(k+1|k)H^T[HP(k+1|k)H^T+R]^-1

and (3) updating the state:

and (3) covariance updating:

P(k+1|k+1)＝[I_n-K(k+1)H]P(k+1|k)

by using the filtering calculation prototype, all following vehicles can perform one-step state prediction on the received front vehicle information.

Suppose that state information of the vehicle ahead at time k at time s is received by the following vehicle observer and that time k differs from time s by N filter cycle durations T. Obtaining a one-step state prediction recurrence sequence of s time → k time by a recurrence mode, wherein N is k-s:

is provided withFor the one-step state prediction of the system without time delay, the one-step state prediction error is as follows:

estimating the state before the s moment, and enabling:

then there isThe information of the system state transition matrix D and the s-time system of N-1 filtering cycles at the s → k time is contained.

The final calculated state prediction difference value is set as:

compensated front vehicle displacementAnd the speed of the front vehicleAnd is

Will be provided withAndreplacing s in interval error formula and speed difference formula_i-1And v_i-1Obtaining a compensated pitch error e_sSum velocity difference e_v：

The third step: and (4) performing MPC algorithm solution based on the optimized vehicle state space model to obtain expected control input, establishing a lower layer controller, and feeding the control input back to the controlled vehicle to realize acceleration or braking of the vehicle.

The following discrete linear state space model is obtained after conversion:

y(k)＝Cx(k)

whereinWhich is indicative of the state of the system,which represents the output that the system can measure,for controlling the inputIs a jamming input. Assuming that the state vector and the interference vector are in each sampling period t_sIs internally measurable and the interference of time k to future time is predicted as itselfThat is, ω (k + j | k) ═ ω (k). The prediction of the future N sampling instants can be achieved by the following iterative model:

where N is referred to as the prediction time domain.

The performance indicator function is defined as:

wherein y is_ref(k + j | k) is a reference trajectory which is related to the output measurement value up to the current time, or is a predetermined trajectory, symbolRepresenting a quadratic function, Q, R are error and input weighting matrices, respectively. The above formula shows that the performance index mainly considers two aspects of the error magnitude of the output and the reference and the energy magnitude of the input, and can be rewritten into the following vector form:

defining vector e (k) as the deviation of the system free response from the future target trajectory:

E(k)＝Y_ref(k)-M_xx(k)-M_uu(k-1)-M_ωω(k)

the standard form of quadratic programming available:

and establishing a lower layer controller for controlling the acceleration and braking of the automobile:

when the automobile is accelerated, the expected torque of the engine is calculated according to the inverse longitudinal dynamics model from the expected acceleration, and then the expected accelerator opening is obtained through an inverse table look-up method by combining the engine rotating speed information. Acceleration control process the relationship between the desired torque and the desired acceleration of the engine is:

according to the engine torque characteristic of the vehicle, the engine speed omega can be known_eAnd a desired torque T_e,desThe corresponding desired throttle opening is obtained under the conditions of:

a_des＝f(T_e,des,ω_e)

when an automobile is braked, it is necessary to obtain a desired braking force from a desired acceleration and then obtain a desired braking pressure from the desired braking force. The automobile has no accelerator input during braking, and the vehicle dynamics equation of the braking process can be written as follows:

ma_des＝-F_bdes-F_areo-R_x

the desired brake pressure may be found as:

the invention has the characteristics and beneficial effects that:

the invention provides a formation control method based on time delay compensation, which comprises the following steps: and designing a control law of formation by adopting Model Predictive Control (MPC), and realizing formation maintenance under an ideal communication condition. And then introducing communication delay, predicting actual state information of the pilot vehicle by using a Kalman filtering algorithm (KF), and compensating the communication delay by using a state prediction error, thereby not only considering the stability of formation, but also considering the control effect of vehicle formation.

Drawings

The invention will be described in detail with reference to the following figures:

FIG. 1 is a schematic illustration of a three-vehicle platoon driving in an embodiment of the present invention;

FIG. 2 is an overall control architecture diagram in an embodiment of the present invention;

FIG. 3 is an engine output torque MAP MAP in an embodiment of the present invention;

fig. 4 is a diagram of the effect of fleet control taking into account communication delays in an embodiment of the present invention.

Detailed Description

The technical solutions in the embodiments of the present invention will be described in detail and clearly with reference to the accompanying drawings. The described embodiments are only some of the embodiments of the present invention.

As shown in fig. 1 and fig. 2, the technical solution of the present invention for solving the above technical problem is:

establishing an upper-layer control kinematic equation based on a model prediction control principle;

establishing a Kalman filtering model, and optimizing a vehicle state space model under the influence of communication delay by adopting a delay error compensation method to obtain an optimized vehicle state space model;

and (4) performing MPC algorithm solution based on the optimized vehicle state space model to obtain expected control input, establishing a lower layer controller, and feeding the control input back to the controlled vehicle to realize acceleration or braking of the vehicle.

The establishment of the upper-layer control kinematic equation based on the model prediction control principle comprises the following steps:

establishing an intelligent vehicle longitudinal motion nonlinear third-order model:

where s is the longitudinal position of the smart vehicle measured from the inertial reference position, v, a are the speed and acceleration of the vehicle, respectively, and η is the control input to the engine.

And (3) linearizing the feedback of the original nonlinear model:

η＝ma_des+C_dv²+d_m+2τC_dva；

wherein a is_desI.e. the desired acceleration, C, determined by the upper level controller_dRepresents the aerodynamic drag coefficient, d_mτ is the time constant of the vehicle engine, which is the mechanical resistance of the vehicle.

The state space model of the vehicle in the intelligent fleet longitudinal following control system based on the communication delay is derived as follows:

in the above formula, the first and second carbon atoms are,

x＝[e_s e_v a_i v_i]^T u＝a_desω＝a_i-1；、

where x, u, and ω are the state, control input, and disturbance input, respectively. e.g. of the type_s、e_v、a_i、v_i、a_i-1The distance error, the speed error, the ith vehicle acceleration, the ith vehicle speed and the (i-1) th vehicle acceleration are respectively.

The establishing of the Kalman filtering model and the optimizing of the vehicle state space model under the influence of communication delay by adopting a delay error compensation method comprise the following steps:

establishing a state space equation of the system:

X(k+1)＝DX(k)+W

Z(k)＝HX(k)+V

wherein x (k) ═ z (k) ═ s_i-1(k),v_i-1(k),a_i-1(k)]^T，s_i-1(k)、v_i-1(k)、 a_i-1(k) Respectively, the displacement, the speed and the acceleration of the front vehicle, and W, V are respectively a process noise matrix and a measurement noise matrix of the system, and the variances of the matrixes are respectively Q and R.

Where T is the sampling period.

After the state space equation of the system is obtained, a recursion equation set for estimating the state of the front vehicle by adopting a Kalman filtering method can be obtained as follows:

and (3) state prediction:

and (3) covariance prediction:

P(k+1|k)＝DP(k|k)D^T+Q

filtering gain:

K(k+1)＝P(k+1|k)H^T[HP(k+1|k)H^T+R]^-1

and (3) updating the state:

and (3) covariance updating:

P(k+1|k+1)＝[I_n-K(k+1)H]P(k+1|k)

and by utilizing a filtering calculation prototype, all following vehicles can perform one-step state prediction on the received front vehicle information, then predict the front vehicle state information under the condition of no time delay to obtain a state prediction error, and compensate the state misdetection error into the state quantity of the front vehicle to obtain an optimized state variable.

The MPC algorithm solution is carried out based on the optimized vehicle state space model, expected control input is obtained, a lower layer controller is established, the control input is fed back to a controlled vehicle, and the acceleration or braking of the vehicle is realized by the following steps:

the following discrete linear state space model is obtained:

y(k)＝Cx(k)

whereinWhich is indicative of the state of the system,which represents the output that the system can measure,for controlling the inputIs a jamming input. Assuming that the state vector and the interference vector are in each sampling period t_sIt is also possible to predict the interference at time k with respect to the future time as itself, i.e., ω (k + j | k) ═ ω (k). The prediction of the future N sampling instants can be achieved by the following iterative model:

where N is referred to as the prediction time domain.

The performance indicator function is defined as:

wherein y is_ref(k + j | k) is a reference trajectory which is related to the output measurement value up to the current time, or is a predetermined trajectory, symbolRepresenting a quadratic function, Q, R are error and input weighting matrices, respectively. Tong (Chinese character of 'tong')And obtaining the expected acceleration by solving the performance index function.

And establishing a lower layer controller for controlling the acceleration and braking of the automobile:

when the automobile is accelerated, the expected torque of the engine is calculated according to the inverse longitudinal dynamics model from the expected acceleration, and then the expected accelerator opening is obtained through an inverse table look-up method by combining the engine rotating speed information. Acceleration control process the relationship between the desired torque and the desired acceleration of the engine is:

the invention selects a front wheel driven type B-type hatchback car, the engine torque characteristic curve of which is shown in figure 3, according to the engine torque characteristic curve chart, the engine speed omega can be known_eAnd the desired torque T_e，desThe corresponding expected throttle opening is obtained by a table look-up method under the conditions of (1):

a_des＝f(T_e，des，ω_e)

when an automobile is braked, it is necessary to obtain a desired braking force from a desired acceleration and then obtain a desired braking pressure from the desired braking force. The automobile has no accelerator input during braking, and the vehicle dynamics equation of the braking process can be written as follows:

ma_des＝-F_bdes-F_areo-R_x

the desired brake pressure may be found as:

the engine output torque characteristic in the embodiment of the invention is shown in fig. 3.

In order to verify the effectiveness of the algorithm, the acceleration of the lead vehicle is changed in a sinusoidal cycle within 0-50 s, the lead vehicle runs at a constant speed after 50s, the control effect finally achieved according to the method provided by the invention is shown in fig. 4, and it can be seen from the figure that the actual acceleration of the lead vehicle can better follow the expected acceleration and the actual following distances of the 2 nd and 3 rd vehicles can better follow the expected following distances under the condition of considering communication delay, so that the algorithm provided by the invention can well maintain the vehicle queue control effect.

The above description is only one embodiment of the present invention, and should not be taken as limiting the invention, and any modifications, equivalents, improvements, etc. made within the spirit and scope of the present invention are intended to be included therein.