Distributed driving vehicle path tracking method based on variable time domain predictive control

Abstract

The invention discloses a distributed driving vehicle path tracking method based on variable time domain model predictive control, and belongs to the technical field of distributed driving electric vehicle control. Based on the five-degree-of-freedom vehicle dynamics model and the linear tire model, an MPC path tracking control model is established; adaptively adjusting a prediction time domain in the MPC path tracking control model by using a fuzzy controller to obtain a variable time domain MPC path tracking control model; and converting the path tracking problem of the variable time domain MPC path tracking control model into a constrained quadratic programming optimization problem, solving to obtain a front wheel rotation angle increment at the current moment, superposing the front wheel rotation angle increment at the current moment with a front wheel rotation angle at the previous moment to obtain a front wheel rotation angle at the current moment as a control quantity, and driving a vehicle to track an expected path. The path tracking precision of the distributed driving electric vehicle is improved, and the driving stability is enhanced.

Description

Technical Field

[0001] The present invention belongs to the technical field of distributed drive electric vehicle control, and in particular relates to a distributed drive vehicle path tracking method based on variable time domain model predictive control. Background Art

[0002] In actual driving, distributed drive electric vehicles will inevitably face dangerous conditions such as high speed, low adhesion, and emergency steering, which are likely to cause the vehicle to become unstable and deviate from the expected trajectory. Therefore, giving full play to the advantages of multi-motor power redundancy of distributed drive electric vehicles and improving the driving stability and path tracking accuracy of vehicles under dangerous conditions are key factors in ensuring vehicle safety. At present, experts and scholars in this field have achieved some research results.

[0003] Chen Chaoxin took distributed rear-wheel drive autonomous driving vehicles as the research object, and proposed a collision avoidance control strategy that comprehensively ensures path tracking accuracy and driving stability in response to the possible instability problems under emergency steering collision avoidance conditions. First, based on the five-degree-of-freedom vehicle model and the linear tire model, a path tracking controller was designed based on the model predictive control (MPC) theory. Then, based on the fuzzy control theory, the yaw angular velocity and the sideslip angle of the center of mass were used as the control targets, and the yaw stability controller was designed. Finally, the desired additional yaw moment output by the stability controller was achieved by distributing the left and right rear wheel driving torques to ensure the driving stability of the vehicle when tracking the desired collision avoidance path. Zhang Lei et al. proposed a path tracking and direct yaw moment coordination control strategy based on a hierarchical architecture for distributed four-wheel drive electric vehicles. The upper layer of the control strategy established an MPC controller based on the three-degree-of-freedom vehicle model, considered multiple nonlinear constraints, and used theoretical analysis to optimize the prediction time domain and control time domain to obtain the desired additional yaw moment and front wheel turning angle. The lower layer of the control strategy optimizes the longitudinal force distribution of the four drive wheels to minimize the tire load rate.

[0004] At present, in the research on path tracking and stability control of distributed drive electric vehicles, most of the research is just a simple superposition of path tracking control algorithm and stability control, and the coordination relationship between the two different control systems is not clear. For example, under non-dangerous working conditions, due to the different control targets of the two systems, in order to make the center of mass sideslip angle and yaw rate follow the set expected values, the additional yaw torque output by the stability control system may increase the degree of vehicle deviation from the expected path, thereby reducing the path tracking accuracy. How to use the coordination relationship between the two different control systems, reasonably coordinate the timing of the intervention of the stability control strategy, and avoid the premature intervention of the stability control strategy to affect the path tracking accuracy is a problem to be solved in this field. The present invention introduces the quantitative judgment of the stability state of the vehicle into the path tracking and stability control of the distributed drive electric vehicle, and comprehensively guarantees the path tracking accuracy and driving stability of the vehicle by real-time judgment of the vehicle stability state and reasonable coordination of the timing of the intervention of the stability control strategy. Summary of the invention

[0005] In order to overcome the above technical problems, the present invention proposes a distributed drive vehicle path tracking method based on time-varying model predictive control, which reasonably coordinates the timing of intervention of the stability control strategy. When the vehicle is in a relatively stable state, the control target focuses on improving the path tracking performance, and when the vehicle is about to be in an unstable state, the control target focuses on ensuring the driving stability of the vehicle. The quantitative judgment of the vehicle's stability state is introduced into the path tracking and stability control of the distributed drive electric vehicle. Through real-time judgment of the vehicle's stability state, the timing of intervention of the stability control strategy is reasonably coordinated to comprehensively ensure the path tracking accuracy and driving stability of the vehicle.

[0006] The present invention establishes an MPC path tracking control model and derives a prediction equation based on the MPC control method theory, according to a five-degree-of-freedom vehicle dynamics model and a linear tire model; a fuzzy controller is used in the MPC path tracking control model, and adaptive adjustment of the prediction time domain in the controller is achieved through fuzzy rules; further, in order to ensure the path tracking accuracy and control continuity of the vehicle, an optimization objective function and constraint conditions are designed, and the path tracking problem is converted into a constrained quadratic programming optimization problem, and the front wheel steering angle control amount is solved to achieve the vehicle tracking of the desired path, thereby improving the path tracking control stability of the distributed drive electric vehicle.

[0007] To achieve the above object, the technical solution adopted by the present invention is as follows:

[0008] A distributed drive vehicle path tracking method based on time-varying model predictive control, comprising:

[0009] Based on the five-degree-of-freedom vehicle dynamics model and the linear tire model, an MPC path tracking control model is established;

[0010] The prediction time domain in the MPC path tracking control model is adaptively adjusted using the fuzzy controller to obtain a variable time domain MPC path tracking control model.

[0011] The path tracking problem of the variable time domain MPC path tracking control model is transformed into a constrained quadratic programming optimization problem. The front wheel angle increment at the current moment is obtained by solving it. The front wheel angle increment at the current moment is superimposed on the front wheel angle at the previous moment to obtain the front wheel angle at the current moment as the control variable to drive the vehicle to track the desired path.

[0012] The beneficial effects of the present invention are:

[0013] (1) The present invention has precise path tracking capability. In various working conditions, such as lane changing, serpentine and double lane changing, the actual driving trajectory of the vehicle is highly consistent with the expected trajectory. For example, when driving at 45km / h in lane changing conditions, the maximum tracking error does not exceed 0.07m; in serpentine conditions, the maximum tracking error does not exceed 0.05m, which fully demonstrates good path tracking accuracy.

[0014] (2) The present invention also has excellent driving stability. During the driving process of the vehicle, stability parameters such as the center of mass slip angle and yaw rate can be maintained within a reasonable range. For example, in the above working conditions, the center of mass slip angle does not exceed 0.15 rad, and the yaw rate curve can also follow the expected curve well, ensuring the driving stability of the vehicle.

[0015] (3) The present invention is aimed at emergency obstacle avoidance paths with sudden changes in curvature, such as double lane changes, which are challenging. The method basically meets the path tracking requirements at low and medium vehicle speeds, and can cope with high-speed driving conditions to a certain extent, demonstrating its adaptability to complex working conditions.

[0016] (4) The present invention can overcome the problem of increased tracking error and decreased control accuracy of the traditional time-domain MPC path tracking controller when driving at high speeds by adaptively adjusting the prediction time domain. For example, under double lane change conditions, the maximum path tracking error of the time-varying time-domain MPC at different vehicle speeds is significantly reduced compared to the time-domain MPC. At vehicle speeds of 36km / h, 54km / h and 72km / h, the maximum path tracking error is reduced by 28.6%, 23.5% and 17.5%, respectively, and the variance of the path tracking error value is reduced by 38.5%, 46.7% and 32.6%, respectively.

[0017] (5) The present invention optimizes the dynamic response of the vehicle. The variable time domain MPC controller can reduce the jitter of the front wheel steering angle output curve in the later stage, making the output curve smoother and improving the driving experience of passengers. Especially at the end of the steering under high-speed conditions, it can significantly reduce the jitter of the center of mass side slip angle curve and the yaw rate curve, and improve the driving stability of the vehicle. For example, at speeds of 54km / h and 72km / h, the variable time domain MPC controller can reduce the amplitude of the center of mass side slip angle curve by 6.4% and 6.6% respectively.

[0018] (6) The present invention can also adapt to different vehicle speed changes. The prediction time domain adaptive adjustment based on fuzzy logic control enables the controller to optimize the control parameters in real time according to factors such as vehicle speed and vehicle stability, and maintain a good control effect at different vehicle speeds, thereby enhancing the controller's adaptability to various driving conditions. For example, when driving at high speed, the prediction time domain is appropriately increased to plan steering operations in advance; when driving at low speed, the prediction time domain is appropriately reduced to improve real-time performance and control accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] Figure 1 A schematic diagram of a distributed drive vehicle path tracking method based on time-varying domain model predictive control according to an embodiment of the present invention;

[0020] Figure 2 A flow chart of a distributed drive vehicle path tracking method based on time-varying model predictive control according to an embodiment of the present invention;

[0021] Figure 3 To predict the impact of time domain on path tracking performance;

[0022] Figure 4 It is a time-variable domain fuzzy controller;

[0023] Figure 5 is the vehicle stability state index membership function;

[0024] Figure 6 is the path tracking lateral error membership function;

[0025] Figure 7 is the membership function of the adjustment amount in the prediction time domain;

[0026] Figure 8 is a three-dimensional graph of inference rules;

[0027] Fig. 9 This is the simulation result of lane changing condition (45km / h);

[0028] Fig.10 This is the simulation result of the serpentine condition (45km / h);

[0029] Fig.11This is a schematic diagram of the ISO 3888 double lane shift standard;

[0030] Fig.12 It is the reference path for double lane-changing conditions;

[0031] Fig.13 The simulation results of the tracking path for the double lane-changing condition;

[0032] Fig.14 The simulation results of vehicle state parameters in double lane-changing condition;

[0033] Fig.15 The simulation comparison results of double lane change conditions (36km / h);

[0034] Fig.16 The simulation comparison results of double lane change conditions (54km / h);

[0035] Fig.17 The simulation comparison results of double lane-changing conditions (72km / h). DETAILED DESCRIPTION

[0036] The present invention will be further described below in conjunction with specific embodiments. It should be understood that these embodiments are only used to illustrate the present invention and are not intended to limit the scope of the present invention. In addition, it should be understood that after reading the content taught by the present invention, those skilled in the art can make various changes or modifications to the present invention, and these equivalent forms fall within the scope limited by the appended claims of the application equally.

[0037] like Figure 1 As shown, the present invention establishes an MPC path tracking control model based on the model predictive control (MPC) theory, a five-degree-of-freedom vehicle dynamics model and a linear tire model, designs and optimizes the objective function and constraints to ensure the path tracking accuracy and control continuity of the vehicle. In order to improve the path tracking accuracy and consider the stability of the vehicle, an analysis of the influence of the prediction time domain on the MPC controller is adopted, and a fuzzy controller is used to realize the adaptive adjustment of the prediction time domain in the controller, and a variable time domain MPC path tracking control model is designed. Finally, the path tracking problem is converted into a constrained quadratic programming optimization problem, and the front wheel angle control quantity is solved to realize the path tracking requirements of the vehicle. The above process can be described as the following three points:

[0038] (1) Model prediction

[0039] Based on the control objective of the MPC controller, the state and control quantities of the system are defined, and the prediction equation is derived. The system control quantity at the future moment and the system state quantity at the current moment are substituted into the prediction equation and solved within a certain finite prediction time domain. Finally, the state information of the vehicle at the future moment can be obtained.

[0040] (2) Scrolling Optimization

[0041] Generally, the first item in the system control increment sequence is selected as the actual output of the control system. As the system state is updated, the controller will recalculate a new control increment in the next sampling period. This control increment is superimposed with the control amount at the previous moment to obtain the new control amount of the current system. This new control amount is the optimal solution that can minimize the control error of the current system. As subsequent operations continue, the above process will be repeated continuously to form a rolling optimization, thereby improving the control accuracy and robustness of the controller.

[0042] (3) Feedback correction

[0043] This part is the key link for the controller to reduce the control error. In the actual control process, the actual driving state of the vehicle is often affected by factors such as the external environment, which may cause a deviation between the actual control result and the expected result. Therefore, it is necessary to set up a feedback correction part to correct these errors. At the new sampling moment, this part will calculate the control error, and then correct the model's predicted value based on this error, and update the system state for the next round of optimization solution, so as to achieve the goal of the controller to continuously reduce the control error.

[0044] Based on the above ideas, the present invention proposes a distributed drive vehicle path tracking method based on time-varying model predictive control. Figure 2 As shown in the figure, firstly, the traditional UKF-based vehicle parameter estimation module is used to estimate the vehicle's center of mass slip angle β and yaw rate ω according to the collected vehicle driving state information. The estimation results of β and ω provide an important reference for vehicle stability state judgment and dynamic control. Secondly, the stability index calculation module is based on the estimated center of mass slip angle β and road adhesion coefficient μ, and according to the division results of the traditional phase diagram stability domain, calculates the quantitative index PPS_region that characterizes the current stability state of the vehicle. The path tracking control module is based on the estimation results of PPS_region and β, ω, combined with the path tracking lateral error E y , the fuzzy control algorithm is used to control the time domain parameter N in the MPC path tracking controller p Make real-time adjustments to optimize the path tracking effect and finally output the front wheel turning angle δ f Realize the path tracking control of the vehicle. The present invention mainly includes the following parts:

[0045] (I) Design of path tracking controller based on MPC

[0046] Assume that the system state is ξ and the control quantity is u. The state quantity includes the lateral velocity Longitudinal speed Yaw angle Yaw rate The horizontal coordinate Y in the geodetic coordinate system, the vertical coordinate X in the geodetic coordinate system, and the control amount u is the front wheel steering angle δ f :

[0047]

[0048] u={δ f} (2)

[0049] To facilitate calculation, the coordinates in the vehicle coordinate system and the earth coordinate system need to be converted:

[0050]

[0051] Among them, x and y represent the longitudinal coordinate and lateral coordinate of the vehicle in the vehicle coordinate system, respectively.

[0052] In order to improve the computing speed of the controller and ensure the real-time control, the present invention simplifies the vehicle dynamics differential equation. Assuming that the front wheel angle is small, cosδ f ≈1 and sinδ f ≈0 holds true. Therefore, based on the five-degree-of-freedom vehicle dynamics model and the tire model, the state equation, i.e., the nonlinear prediction model, can be established as shown in formula (4):

[0053]

[0054] in, is the lateral acceleration in the vehicle coordinate system, is the longitudinal acceleration of the vehicle, is the yaw angular velocity, is the yaw angular acceleration, s f Indicates the side slip angle of the vehicle's front wheels, s r represents the side slip angle of the vehicle's rear wheels, is the lateral velocity of the vehicle in the geodetic coordinate system, is the longitudinal velocity of the vehicle in the geodetic coordinate system, a is the distance from the center of mass of the vehicle to the front axle, b is the distance from the center of mass of the vehicle to the rear axle, C cf is the front wheel cornering stiffness, C cr is the rear wheel cornering stiffness, I z is the moment of inertia about the vertical axis of the vehicle’s center of mass, δ f is the front wheel turning angle, f Y 、f X Both represent derivative operations.

[0055] Since the nonlinear model is complex and requires large amounts of computation, it often cannot meet the real-time requirements of the control algorithm. Therefore, it needs to be linearized to improve real-time performance.

[0056] At the system reference operating point (ξ ref ,u ref ), the state equation of the system is Taylor expanded. Subtracting the expanded state equation from the original state equation can obtain the linear time-varying equation, that is, the continuous time error prediction model:

[0057]

[0058] In the formula, a(t) and B(t) are the Jacobian matrices of the system state vector ξ(t) and the input vector u(t), respectively, which describe the instantaneous dynamic characteristics of the system error. Specifically, A(t) is the state transfer matrix, which describes the dynamic characteristics of the system state error, that is, the change of the system state error over time, and B(t) is the input matrix, which describes the influence of the input error on the system state error. Δξ=ξ-ξ ref ; u(t)=uu ref .

[0059] The vehicle state parameters at the future time cannot be directly calculated by the above linear time-varying equation. Therefore, the present invention discretizes the linear time-varying equation. The processed discrete system state space equation and output equation are:

[0060] ξ(k+1)=A(k)ξ(k)+B(k)u(k) (10)

[0061] η(k)=Cξ(k) (11)

[0062] A(k)=I+TA(t) (12)

[0063] B(k)=TB(t) (13)

[0064] In the formula, η is the output of the system state; I is the unit matrix; A(k) and B(k) are the Jacobian matrices with respect to the sampling period T. Specifically, A(k) is the state transfer matrix, which describes the dynamic characteristics of the system state between discrete time steps and reflects the changes of the system state between discrete time steps. B(k) is the input matrix, which describes the influence of the input on the changes of the system state between discrete time steps.

[0065] Assume that the predicted output in the MPC controller is the yaw angle Combined with the horizontal coordinate Y, formula (1) and formula (11), the C matrix can be determined as:

[0066]

[0067] In the continuous time domain, a new system state space equation is constructed to characterize the update iteration of the system state quantity:

[0068]

[0069] Δu(k|t)=u(t|t)-u(k-1|t) (19)

[0070] In the formula, u is the control dimension, and its value is 1; n is the state dimension, and its value is 6. ξ(k|t) is the state at time t in the kth step, is the continuous state quantity at time t in the k-th step, u(k-1|t) is the control quantity at time t in the k-th step, and Δu(k|t) is the control increment at time t in the k-th step.

[0071] According to formula (15), using the control increment Δu(k|t)Δu(t) and the state quantity ξ(k|t) at time t, the system output η(k+1|t) and the state quantity ξ(k+1|t) of the next step length k+1 can be obtained.

[0072] The present invention sets the prediction time domain of the controller to be N p , the control time domain is N u , stipulate N u <N p After continuous iterations, we can get p The system state and output at the time. After integration, the prediction equation required by the controller can be obtained:

[0073]

[0074]

[0075] Where Z(t) is the prediction time domain N p η(t) sequence in the control time domain N u The front wheel turning angle δ f The increment sequence of ψ and Θ are coefficient matrices. By substituting Δu(k|t)Δu(t) into formula (20) for calculation, the state output of the system at the future moment can be predicted, thus realizing the “prediction function”.

[0076] The expected lateral displacement ΔY and expected yaw angle of the vehicle can be calculated from the preset expected path, and the difference between the expected value and the actual value of the two parameters of lateral displacement and yaw angle is often used as the tracking error of the path. The primary goal of the MPC controller is to make the vehicle follow the expected path with the smallest possible tracking error. Therefore, in order to further ensure the control accuracy of the MPC controller, the present invention designs an optimization objective function to achieve this goal.

[0077] In this embodiment, the optimization objective function includes: a path tracking accuracy term J1 and a control amount penalty term J2, as shown in formula (25).

[0078]

[0079] (1) Path tracking accuracy item J1

[0080] First, the predicted vehicle lateral displacement ΔY is penalized against the expected lateral displacement ΔY of the set path. ref Secondly, the vehicle yaw angle should be penalized and the desired yaw angle The deviation value between them is used to ensure that the vehicle heading and the tangent angle of the path are as consistent as possible.

[0081]

[0082] In the formula, q y , are the weight matrices of path tracking error and yaw angle error, respectively. y reflects the sensitivity to path tracking errors, Reflects the sensitivity to yaw angle error, q y and The larger the weight is, the more sensitive it is to the error, and the greater the impact of the error on J1.

[0083] (2) Control quantity penalty term J2

[0084] Since the research object of the present invention is a distributed drive electric vehicle with a steering function on the front wheels, it is necessary to consider whether the steering mechanism can meet the limitations of the mechanical structure when outputting the control quantity, which requires that the steering angle of the front wheels should not be too large. At the same time, in some high-speed working conditions, in order to avoid the vehicle instability caused by excessive steering when the vehicle is tracking the path, it is also necessary to penalize the front wheel angle increment:

[0085] J2=q δ δ f 2 +q Δδ Δδ f 2 (27)

[0086] In the formula, q δ ,q Δδ are the weight matrix of the control amount and the weight matrix of the control increment respectively. δ With q Δδ The larger the weight is, the more sensitive it is to the control quantity and control increment, and the greater the impact on J2.

[0087] In order to ensure the control accuracy of vehicle path tracking and avoid oscillation in the output of the vehicle's front wheel angle, the adaptability of each weight matrix to the algorithm program should be fully considered in the adjustment, and corresponding optimization adjustments should be made in combination with the simulation results, so that the MPC path tracking controller can reach an optimal state as much as possible.

[0088] In order to ensure the control continuity and driving stability of the vehicle under certain extreme conditions, the front wheel steering angle δ f , front wheel steering angle increment Δδ f , lateral displacement ΔY, yaw angle The center of mass sideslip angle β is constrained by upper and lower limits:

[0089]

[0090] To facilitate the solution, the objective function and constraints are integrated to transform the path tracking problem into a constrained quadratic programming optimization problem:

[0091]

[0092] G = 2θ T QE (31)

[0093]

[0094] Z ref (t) = [η ref (k+1|t),…,η ref (k+N p |t)] (33)

[0095] Where ε is the relaxation factor, ρ is the weight coefficient of the relaxation factor, H is the positive definite Hessian matrix, G is the matrix of the path tracking error E, and Z ref (t) is the expected lateral displacement and expected yaw angle of the trajectory in the prediction time domain at the current time t.

[0096] If the objective function is to be solved while satisfying all constraints, no solution may occur. The role of the relaxation factor ε is to avoid the situation where there is no optimal solution within the constraints, thereby ensuring control continuity. At the same time, ρ should be as small as possible, in order to ensure that each constraint does not exceed the boundary. Based on formula (29), the optimization toolbox can be used to solve the front wheel angle increment at the current moment, and superimpose it with the front wheel angle at the previous solution moment to obtain the front wheel angle at the current moment. Then, it is continuously iterated to finally achieve the path tracking control of the vehicle.

[0097] In a specific implementation of the present invention, the five-degree-of-freedom vehicle dynamics model and the tire model are described as follows. The model itself belongs to the conventional technology in the field. (1) Five-degree-of-freedom vehicle dynamics model

[0098] The present invention uses a five-degree-of-freedom vehicle dynamics model, which includes lateral, longitudinal, yaw and rotational motions of the left and right rear wheels. The longitudinal, lateral and yaw dynamic differential equations of the vehicle can be obtained by force analysis:

[0099]

[0100] In the formula, F yf 、F yr is the lateral force of the front and rear wheels; F xf 、F xr is the longitudinal force of the front and rear wheels; δ f is the front wheel turning angle; x, y are the center of mass coordinates in the vehicle coordinate system.

[0101] (2) Tire model

[0102] In terms of tire model selection, in order to reduce the computational complexity, it is assumed that the tire is in a small turning angle and linear working area and simplified into a linear tire model. At this time, the lateral force and longitudinal force of the tire can be calculated by the following formula:

[0103]

[0104] Where s is the tire slip rate; α is the tire side slip angle; C l is the longitudinal stiffness of the tire; C c is the lateral stiffness of the tire.

[0105] The side slip angle α can be calculated by the following formula:

[0106]

[0107] In the formula, v ty and v tx are the lateral and longitudinal speeds of the tire, respectively.

[0108] In the vehicle coordinate system, the longitudinal and lateral velocities of the wheels can be converted as follows:

[0109]

[0110] Under the small angle assumption, the tangent value of the tire slip angle is approximately equal to its own value, that is, tanα≈α. Therefore, combining equations (38) to (39) yields:

[0111]

[0112] Combining formula (37) and formula (40), we can get the lateral force and longitudinal force of the front and rear wheels as follows:

[0113]

[0114] In the formula, parameter C cf , C cr , C lf , C lr The stiffness coefficients of the left and right wheels are taken into account during the calculation, so in formula (41), the calculated forces of the front and rear wheels do not need to be multiplied by 2.

[0115] (II) Design of path tracking controller based on variable time domain MPC

[0116] Since the control parameters in the controller often have an important influence on the control effect. Therefore, realizing the adaptive adjustment of the control parameters in the controller will play a very important role in optimizing the control effect. The present invention can be known from the simulation results that the changes in working conditions and vehicle speed have a certain influence on the control effect of the path tracking controller. Especially when the path tracking is performed under a high-speed and relatively extreme path, the path tracking accuracy and driving stability of the vehicle will be significantly reduced. However, when the vehicle is traveling at a lower speed or tracking a relatively simple and stable path, the controller often has a good control effect. One of the important reasons for this difference in results is that the designed controller adopts fixed parameter control, and the vehicle often encounters some complex and changeable working conditions during actual driving, and the speed is also constantly changing, which requires the designed controller to adjust the control parameters in the controller in real time according to the specific motion state of the vehicle at the current moment. Fixed control parameters can often only adapt to specific vehicle driving states, so realizing adaptive adjustment of control parameters and allowing the controller to achieve optimal control under the current driving state as much as possible is of great significance to improving the control effect.

[0117] Prediction time domain N p It is one of the important parameters of MPC controller, N p The size setting often has a great impact on the performance of the controller. p It can express the degree of prediction of the future state of the controlled system by the MPC controller. When other conditions remain unchanged, N p The larger the N is, the farther the MPC controller can predict. Therefore, choosing an appropriate prediction time domain has an important impact on improving the control effect of the MPC controller. p Too large or too small will weaken the control effect. For example, when the vehicle is traveling at a higher speed, the controller is required to predict a farther position and obtain more vehicle status information at future moments. At this time, the prediction time domain N should be appropriately increased. p, so that the vehicle can make steering operations in time. When the car is driving at a lower speed, the prediction time domain N is appropriately reduced. p , which can not only improve the real-time performance of the MPC controller, but also optimize the control accuracy of the controller.

[0118] This embodiment specifically analyzes the impact of too large or too small prediction time domain on the control effect of the controller:

[0119] (1)N p Too large

[0120] On the one hand, the prediction time domain N p If N is too large, the calculation time of the MPC controller will increase, making the real-time performance of the controller worse. p If the value is too large, the weight of the output deviation farther away from the vehicle will increase, while the weight closer to the vehicle will decrease. This will cause the optimal control quantity obtained by the controller to not be the optimal control quantity closer to the current vehicle position. Therefore, the path tracking error at a closer location will increase. The controller will quickly enter the next sampling cycle, re-solve the optimal control quantity for the next cycle, and continue to increase the tracking error at the next closer location. In this way, as the path tracking error continues to accumulate, the vehicle's driving trajectory will eventually deviate from the expected trajectory, such as Figure 3 (a) shown.

[0121] (2)N p Too small

[0122] Although the prediction time domain N p The reduction of N will improve the real-time performance of the MPC controller to a certain extent. p If the value is too small, the controller will predict too little information about the vehicle's future state. At the same time, due to the constraints of the control amount and control increment, the front wheels may not be able to turn in time, which will eventually increase the vehicle's path tracking error. At this time, the controller will try to complete the optimization solution in a shorter time, which often causes the vehicle to move violently or even become unstable, such as Figure 3 (b) as shown.

[0123] Based on the above analysis, in order to improve the adaptability of the MPC path tracking controller, the present invention proposes an adaptive prediction time domain adjustment method based on fuzzy logic algorithm. The basic principle of fuzzy logic control theory is to describe the relationship between various variables in the system according to the fuzzy rules set inside the controller. The advantages of fuzzy control are simple methods, no need to establish an accurate mathematical model, and strong anti-interference ability. At the same time, fuzzy control can be solved offline in advance, and can be converted into a table lookup form during actual application, which has the characteristics of fast response speed.

[0124] In the present invention, the vehicle stability state index (PPS_region), the lateral error of path tracking (E y ), sideslip angle and yaw rate as input to predict the change in time domain (ΔN p ) as output, design as Figure 4 The fuzzy controller shown in the figure is used to adjust the MPC prediction time domain N in real time. p , to ensure the driving stability of the vehicle, and thus improve the path tracking accuracy of the vehicle. Among them, the vehicle stability state index (PPS_region) represents the stable region on the phase plane, which is used to evaluate the stability of the vehicle and can be calculated by existing technologies. Generally, the boundary of the stable region on the phase plane is first determined by theoretical analysis, experimental data or simulation methods, and then the vehicle state is collected in real time by sensors to determine whether it is in the stable region, which is used for control decisions in active safety systems, driver assistance systems and automatic driving systems. It is specifically implemented by those skilled in the art according to the existing technology and will not be repeated here.

[0125] like Figure 4 As shown in the figure, the variable time domain fuzzy controller firstly transforms the input vehicle stability state index PPS_regionPPS_region and the lateral error E y Fuzzification is performed and the corresponding membership function is used to convert it into a variable value in natural language. Then, fuzzy reasoning is performed according to fuzzy rules. Finally, the result obtained after fuzzy reasoning is clarified (i.e., defuzzified) to obtain the change in the predicted time domain ΔN. p .

[0126] The domain of each variable is shown in Table 1. Figure 5-Figure 7 They are the membership functions of the input parameters and output parameters of the fuzzy control algorithm, respectively. Among them, PB, PM, PS are positive large, positive medium and positive small, ZO is zero, NB, NM, NS are negative large, negative medium and negative small, respectively.

[0127] Table 1 Input and output variable domains

[0128]

[0129] Combined with the prediction time domain N p The influence of size on the control effect is analyzed and the fuzzy rules are designed as shown in Table 2. When the PPS_region value is closer to 1, it means that the stability of the vehicle is worse and closer to the unstable state. At this time, it is necessary to increase N appropriately. p When PPS_region is closer to 0, it means that the vehicle is more stable and closer to a stable state. At this time, N needs to be appropriately reduced. pWhen the vehicle is in some extreme situations (such as turning at high speed), the vehicle may produce large tracking errors when tracking the path. However, the first priority at this time is to ensure the stability of the vehicle. The tracking accuracy should be improved as much as possible under the premise of ensuring the stability of the vehicle. At this time, a larger prediction time domain needs to be selected.

[0130] Table 2 Fuzzy reasoning rules for variable time domain adjustment

[0131]

[0132] Use the centroid method to perform the defuzzification operation. p ) is rounded and finally assigned to the MPC controller to achieve adaptive adjustment in the prediction time domain.

[0133] N p =N p0 +Round(ΔN p ) (34)

[0134] The present invention further performs simulation verification of an MPC path tracking controller and simulation verification of a variable time domain MPC path tracking controller.

[0135] (1) Lane-changing and serpentine conditions simulation

[0136] The present invention uses CarSim and Matlab / Simulink software to build a joint simulation platform to verify the effectiveness of the control algorithm. The simulation conditions are shown in Table 3, and the simulation results are shown in Table 3. Figure 9-10 shown.

[0137] Table 3 Path tracking algorithm simulation condition settings

[0138]

[0139] like Fig. 9 As shown in (a), the expected lane change trajectory of the vehicle is set as a smooth trajectory curve generated by a fifth-order polynomial. When the vehicle is tracking the path at a speed of 45 km / h, the actual driving trajectory curve of the vehicle is highly consistent with the expected trajectory curve. Fig. 9 (b) It can be seen that during the entire path tracking process, the maximum tracking error does not exceed 0.07m. Fig. 9 (c) and Fig. 9 (d) It can be seen that the actual yaw rate curve of the vehicle can well follow the expected yaw rate curve, and the side slip angle of the vehicle's center of mass does not exceed 0.15 rad during the whole process.

[0140] like Fig.10As shown in (a), the set serpentine driving trajectory of the vehicle is a smooth trajectory curve generated by a sine wave function. During the vehicle's path tracking process, the actual driving trajectory curve of the vehicle can be highly consistent with the expected trajectory curve with a small tracking error. Fig.10 (b) It can be seen that during the entire path tracking process, the maximum tracking error does not exceed 0.05m. Fig.10 (c) and Fig.10 (d) It can be seen that the actual yaw rate curve of the vehicle can also follow the expected yaw rate curve very well, and the side slip angle of the vehicle's center of mass does not exceed 0.15 rad during the whole process.

[0141] The simulation results of lane-changing and serpentine conditions show that under the control of the MPC path tracking controller, the vehicle can achieve accurate path tracking, which preliminarily verifies the feasibility of the controller. However, considering that the vehicle does not always follow a simple driving path during actual driving. At certain special moments (such as emergency collision avoidance), the vehicle may be required to follow some more complex and extreme paths, which puts higher requirements on the reliability of the path tracking controller.

[0142] (2) Simulation of double lane shifting conditions

[0143] The double lane change path has the characteristic of sudden change in curvature and is the most common vehicle emergency obstacle avoidance driving path. In order to further verify the reliability of the control algorithm, the present invention selects this path for simulation verification again. The double lane change path is drawn with reference to the ISO 3888-1:2018 standard, as shown in Figure 2. Fig.11 The discrete points in the ISO 3888 double lane shift path standard are collected and sorted, and then the curve fitting toolbox is used to fit the double lane shift path using the Gauss fitting method to obtain the expected reference path. The expected reference path result of the double lane shift path fitting is shown in Fig.12 shown.

[0144] In order to quantitatively evaluate the actual path tracking of the vehicle, the maximum forward tracking deviation of the path is used. Maximum backtracking deviation of the path The maximum tracking error E of the path dmax , path tracking variance E dm to conduct evaluation and analysis.

[0145] Path tracking control should not only focus on the tracking accuracy, but also on ensuring the driving stability of the vehicle. This requires that the various stability state parameters of the vehicle should always be within a reasonable range during the tracking process. The yaw rate and the sideslip angle of the center of mass are often used to measure the driving stability of the vehicle. Generally, the instability of the vehicle can be preliminarily determined by judging whether they are within a reasonable range. The upper limit values ​​and allowable ranges of the relevant parameters defined in this embodiment are shown in Table 4. In the table, β lim ,ω lim ,δ flim They are the upper limits of the sideslip angle, yaw rate, and front wheel turning angle, respectively.

[0146] Table 4 Stability parameters and upper limits of control parameters

[0147]

[0148] In this embodiment, the vehicle speed is set to 36km / h, 54km / h and 72km / h respectively to cover low, medium and high speed conditions, and the vehicle speed is kept constant during driving. Figure 13-Figure 14 , as shown in Tables 5 and 6.

[0149] Table 5 Path tracking deviation table

[0150]

[0151] Table 6 Stability parameters and control quantity parameters

[0152]

[0153] Depend on Fig.13 It can be seen that when the vehicle is traveling at a low speed of 36km / h, the controller has a small tracking error, and the actual trajectory curve of the vehicle has a good fit with the reference trajectory curve. As the vehicle speed continues to increase, the deviation between the actual vehicle driving trajectory curve and the reference trajectory curve gradually increases. When the vehicle is traveling at a high speed of 72km / h, the high-speed sharp turn causes a significant deviation between the actual driving trajectory and the reference path.

[0154] Depend on Fig.14 Further analysis of (a) and Table 5 shows that when the vehicle is tracking the path at three different speeds of 36km / h, 54km / h and 72km / h, the maximum tracking errors are 0.1414m, 0.338m, and 0.5667m, respectively, and the path tracking variances are 0.0026m, respectively. 2 、0.015m 2 、0.0458m 2 As the vehicle speed increases, the maximum path tracking error of the vehicle increases, and the control accuracy of the path tracking controller decreases. Fig.14 From the analysis of (b), (c), (d) and Table 6, we can see that although the controller can ensure that the vehicle's center of mass slip angle, yaw rate, and front wheel angle are within the specified range at low, medium, and high speeds, the maximum yaw rate and the maximum front wheel angle of the vehicle are constantly increasing as the vehicle speed continues to increase. When the vehicle is traveling at a high speed of 72km / h, during the entire process of path tracking, the maximum front wheel angle reaches 4.37°, and the maximum yaw rate reaches 16.54° / s, resulting in a "sudden steering" phenomenon, the vehicle's driving stability decreases, and the tracking trajectory also deviates greatly from the reference trajectory.

[0155] Based on the above analysis, in the extreme working condition of double lane change, although the MPC path tracking controller can basically meet the path tracking requirements of the vehicle at low and medium speeds, it has some shortcomings such as increased vehicle path tracking error, poor driving stability, and decreased control accuracy at high speeds. Therefore, it is of great significance to further optimize the control accuracy of the MPC path tracking controller and improve the driving stability of the vehicle.

[0156] Next, the effectiveness of the variable time domain MPC path tracking controller proposed in the present invention is verified using a double lane change condition. The simulated vehicle speeds are selected as 36km / h, 54km / h, and 72km / h to cover three different vehicle speed conditions: low, medium, and high. At the same time, in order to better analyze, this embodiment compares the simulation results of the variable time domain MPC controller with the simulation results of the fixed time domain MPC controller. The simulation results are shown in Figure 2. Figure 15-17 , and as shown in Table 7-Table 8. Fig.15 (a) It can be seen that when the vehicle is traveling at a speed of 36 km / h, both the time domain controller MPC and the variable time domain MPC controller have relatively good path tracking performance, and the actual trajectory curve of the vehicle has a good fit with the reference trajectory curve. Fig.15 (b) It can be seen that the variable time domain MPC control can further reduce the maximum error of vehicle path tracking and improve the control accuracy of the controller compared to the fixed time domain MPC controller. Fig.15 (c) and Fig.15 (d) It can be seen that under the control of the fixed-time MPC controller and the variable-time MPC controller, the center of mass sideslip angle curves of the two have a high degree of similarity, and the actual yaw rate follows the reference value well. However, the variable-time MPC controller can further reduce the curve amplitude of the center of mass sideslip angle. Fig.15 (f) It can be seen that the front wheel steering angle curves output by the two controllers are similar and both can respond well to path tracking requirements.

[0157] Depend on Fig.16(a) It can be seen that when the vehicle is traveling at a speed of 54 km / h, although the vehicle has a good tracking effect under the control of the fixed-time MPC controller and the variable-time MPC controller, the actual trajectory curve of the vehicle under the control of the variable-time MPC controller has a better fit with the reference trajectory curve as a whole. Fig.16 (b) It can be seen that the variable time domain MPC control can more significantly reduce the maximum error of vehicle path tracking than the fixed time domain MPC controller, and further improve the control accuracy of the controller. Fig.16 (c), (d), and (e) show that under the control of the time-domain MPC controller and the variable-time-domain MPC controller, the center-of-mass sideslip angle curves of the two are similar overall, but in terms of the actual yaw rate following the reference value, the control effect of the variable-time-domain MPC controller is better than that of the time-domain MPC controller. Fig.16 (f) It can be seen that compared with the fixed-domain MPC controller, the variable-time domain MPC controller can reduce the jitter of the front wheel angle output curve in the later stage, and the output curve is smoother, which can further improve the driving experience of passengers.

[0158] Depend on Fig.17 (a) It can be seen that when the vehicle is traveling at a speed of 72km / h, the path tracking effect of the variable time domain MPC controller is better than that of the fixed time domain MPC controller. From the tracking curve analysis, it can be seen that in the 100m-120m section, the vehicle under the fixed time domain MPC controller begins to deviate significantly from the reference path, and the maximum tracking error is close to 0.5m. Fig.17 (b) It can be seen that the variable time domain MPC control can significantly reduce the maximum error of vehicle path tracking compared to the fixed time domain MPC controller. Especially at the end of path tracking, the variable time domain MPC controller can make the error curve converge to near 0 faster, further improving the control accuracy of the controller. Fig.17 (c), (d), and (e) show that under the control of the time-domain MPC controller and the variable-time-domain MPC controller, the center of mass slip angle curves of the two controllers have similar trends overall. In terms of the actual yaw rate following the reference value, the control effect of the variable-time-domain MPC controller is slightly better than that of the time-domain MPC controller. In addition, the variable-time-domain MPC controller can significantly reduce the jitter of the center of mass slip angle curve and the yaw rate curve at the end of the steering phase (6-8s period), thereby improving the vehicle's driving stability. Fig.17 (f) It can be seen that compared with the fixed-time domain MPC controller, the variable-time domain MPC controller also reduces the jitter of the front wheel angle output curve in the later stage and improves the passenger driving experience.

[0159] Table 7 Stability parameter comparison table

[0160]

[0161] It can be seen from Table 7 that under the two control modes, the maximum and range of the sideslip angle and yaw rate of the vehicle are similar, but Fig.15 (d) Fig.16 (d) and Fig.17 (d) Analysis shows that the variable time domain MPC controller is slightly better than the fixed time domain MPC controller in terms of the actual value of the yaw rate following the reference value, and can significantly reduce the jitter of the curve at the end of the turn under high-speed conditions. When the vehicle is traveling at speeds of 54km / h and 72km / h, the variable time domain MPC controller can reduce the curve amplitude of the center of mass sideslip angle by 6.4% and 6.6% respectively. Fig.17 (e) It can be seen that under the high-speed condition of 72 km / h, the variable time domain MPC controller can also significantly reduce the jitter of the center of mass sideslip angle curve at the end of the steering, thereby improving the vehicle driving stability.

[0162] Table 8 Path tracking deviation comparison table

[0163]

[0164] As shown in Table 8, when the vehicle is controlled by the variable time domain MPC controller, as the vehicle speed increases, the maximum path tracking errors are 0.1m, 0.26m, and 0.47m, respectively, while when the vehicle is controlled by the time domain MPC controller, the maximum path tracking errors are 0.14m, 0.34m, and 0.57m, respectively. It can be analyzed from the data in the table that compared with the time domain MPC controller, the variable time domain MPC can further reduce the maximum path tracking error of the vehicle. At three different vehicle speeds from low to high, the maximum path tracking errors can be reduced by 28.6%, 23.5%, and 17.5%, respectively, and the variance of the path tracking error value can be reduced by 38.5%, 46.7%, and 32.6%, respectively, which can improve the control accuracy.

[0165] The above examples are only specific embodiments of the present invention. Obviously, the present invention is not limited to the above examples, and many variations are possible. All variations that can be directly derived or associated with the contents disclosed by a person skilled in the art should be considered as the protection scope of the present invention.

Claims

1. A distributed drive vehicle path tracking method based on time-varying model predictive control, characterized in that: include: Based on the five-degree-of-freedom vehicle dynamics model and the linear tire model, an MPC path tracking control model is established; The prediction time domain in the MPC path tracking control model is adaptively adjusted using the fuzzy controller to obtain a variable time domain MPC path tracking control model. The path tracking problem of the variable time domain MPC path tracking control model is transformed into a constrained quadratic programming optimization problem. The front wheel angle increment at the current moment is obtained by solving it. The front wheel angle increment at the current moment is superimposed on the front wheel angle at the previous moment to obtain the front wheel angle at the current moment as the control variable to drive the vehicle to track the desired path.

2. The distributed drive vehicle path tracking method based on time-varying model predictive control according to claim 1, characterized in that: The process of establishing the MPC path tracking control model includes: (1-1) The lateral speed, longitudinal speed, yaw angle, yaw angular velocity, lateral coordinate, and longitudinal coordinate of the distributed drive vehicle are taken as state quantities, and the front wheel steering angle is taken as the control quantity. A nonlinear state equation is established based on the five-degree-of-freedom vehicle dynamics model and the linear tire model. The nonlinear state equation is linearized and discretized in turn to obtain the discrete system state space equation and output equation: ξ(k+1)=A(k)ξ(k)+B(k)u(k) η(k)=Cξ(k) Wherein, ξ(k) and ξ(k+1) represent the state quantities of the k-th step time period and the k+1-th step time period respectively, u(k) represents the control quantity of the k-th step time period, A(k) and B(k) represent the Jacobian matrix of the state quantity of the k-th step time period within the sampling period, C represents the coefficient matrix, and η(k) represents the state output quantity of the k-th step time period; (1-2) Based on the discrete system state space equation, the system state space equation and output equation in the continuous time domain are constructed: Δu(k|t)=u(k|t)-u(k-1|t) η(k+1|t)=Cξ(k+1|t) in, represents the continuous state quantity at time t of the k-th step time period, represents the continuous state quantity at the k+1th step time period at time t, Δu(k|t) represents the control increment at the kth step time period at time t, represents the Jacobian matrix of the continuous state of the k-th step time period within the sampling period, ξ(k|t) represents the state quantity at time t in the k-th step time period, u(k|t) represents the control quantity at time t in the k-th step time period, and u(k-1|t) represents the control quantity at time t in the k-1-th step time period; According to the system state space equation and output equation in the continuous time domain, the state output η(k+1|t) and state quantity ξ(k+1|t) at time t in the k-th step time period can be calculated using the control increment Δu(k|t) and state quantity ξ(k|t) at time t in the k+1-th step time period; (1-3) Set the prediction time domain N p and control time domain N u , N u <N p , the MPC path tracking control model is constructed based on the system state space equation and output equation in the continuous time domain: Among them, Z(t) represents the prediction time domain N p The state output at time t in the control domain N u The control increment at time t within ; ψ represents The influence matrix on Z(t) is defined as the state error propagation matrix, which describes the propagation of the system state quantity within the prediction time range; Θ represents the influence matrix of ΔU(t) on Z(t), which is defined as the control input sensitivity matrix, which describes the influence of the control increment within the prediction time range.

3. The distributed drive vehicle path tracking method based on time-varying model predictive control according to claim 1 or 2, characterized in that: The objective function of the MPC path tracking control model is to minimize the sum of the path tracking accuracy term J1 and the control amount penalty term J2; The path tracking accuracy term is as follows: The control amount penalty term J2 is as follows: J2=q δ d f 2 +q Δδ Dd f 2 Among them, ΔY, Indicates the actual lateral displacement and actual yaw angle of the vehicle, ΔY ref , represents the vehicle reference lateral displacement and reference yaw angle, q y , are the weight matrices of path tracking error and yaw angle error, respectively, f represents the vehicle's front wheel turning angle, Δδ f represents the vehicle front wheel turning angle increment, q δ ,q Δδ They are the weight matrix of the vehicle's front wheel steering angle and the weight matrix of the vehicle's front wheel steering angle increment respectively.

4. The distributed drive vehicle path tracking method based on time-varying model predictive control according to claim 3 is characterized in that: In the MPC path tracking control model, upper and lower limit constraints are set for the vehicle front wheel steering angle, the vehicle front wheel steering angle increment, the vehicle actual lateral displacement, the vehicle actual yaw angle, and the vehicle center of mass sideslip angle.

5. The distributed drive vehicle path tracking method based on time-varying model predictive control according to claim 1, characterized in that: The method of using a fuzzy controller to adaptively adjust the prediction time domain in the MPC path tracking control model to obtain a variable time domain MPC path tracking control model includes: The vehicle stability state index, lateral error of path tracking, sideslip angle of center of mass and yaw rate are taken as the input of the fuzzy controller, which are converted into variable values ​​in natural language by using the membership function. The change in the predicted time domain is taken as the output of the fuzzy controller, and the change in the predicted time domain is superimposed on the predicted time domain at the current moment as the variable time domain result. The variable time domain result is substituted into the MPC path tracking control model as the variable time domain MPC path tracking control model.

6. The distributed drive vehicle path tracking method based on time-varying model predictive control according to claim 5, characterized in that: The vehicle stability state index is calculated based on the vehicle's center of mass sideslip angle and road adhesion coefficient. The closer the vehicle stability state index is to 1, the worse the vehicle stability is, the change in the prediction time domain output by the fuzzy controller is a positive value, and the prediction time domain increases; the closer the vehicle stability state index is to 0, the better the vehicle stability is, the change in the prediction time domain output by the fuzzy controller is a negative value, and the prediction time domain decreases.

7. The distributed drive vehicle path tracking method based on time-varying model predictive control according to claim 5, characterized in that: The lateral error of path tracking is the difference between the actual lateral displacement of the vehicle and the reference lateral displacement of the vehicle.

8. The distributed drive vehicle path tracking method based on time-varying model predictive control according to claim 2, characterized in that: The path tracking problem of the variable time domain MPC path tracking control model is transformed into a constrained quadratic programming optimization problem, which is expressed as: s.t.ΔU min ≤ΔU(t)≤ΔU max U min ≤MΔU(t)+U(t)≤U max Among them, ΔU(t) and U(t) represent the control time domain N u The control increment and control amount at time t in the above example are as follows: the superscript T represents the transpose, ε represents the relaxation factor, H represents the positive definite Hessian matrix, G represents the path tracking error matrix, and ΔU represents the control increment and control amount at time t in the above example. min , ΔU max represents the upper and lower limit constraints of the control increment, M represents the change rate constraint matrix of the control increment ΔU(t), which is used to ensure that the change of the control input within the prediction time range does not exceed the minimum value U of the specified control amount. min and the maximum value U max , U min , U max Represents the upper and lower limit constraints of the control quantity, Z min , Z max Indicates the upper and lower limit constraints of the state output; According to the solved ΔU(t) as the control time domain N u The control increment at time t within the time domain can be obtained by superimposing the front wheel angle at the previous time. u The control variable at time t drives the vehicle to track the desired path.

Images