A variable weight MPC vehicle chassis control method

Abstract

The present application belongs to the field of intelligent automobile chassis control, and particularly relates to a variable weight MPC vehicle chassis control method, which comprises the following steps: S1. Vehicle reference model establishment and MPC optimization problem construction; S2. Variable weight QR design; the weight matrix of the MPC is dynamically adjusted in real time according to the system control mode, so as to realize the best performance of the vehicle under different working conditions; S3. Mode switching based on state determination; the dynamic working condition of the vehicle is identified to support weight switching and control strategy selection. The present application shows significant advantages under high dynamic working conditions such as extreme bends and race tracks, and can realize smaller speed deviation and smoother acceleration and deceleration response. At the same time, thanks to the weight switching strategy, the calculation efficiency and the lightweight control method are equivalent, so as to ensure the real-time performance of the algorithm.

Description

Technical Field

[0001] This invention belongs to the field of intelligent vehicle chassis control, and in particular to a variable weight MPC vehicle chassis control method, which is a vehicle chassis cooperative control based on Model Predictive Control (MPC) and adopts a variable weight mechanism. Background Technology

[0002] Distributed drive electric vehicles, by independently controlling the output torque of each drive wheel, can actively generate yaw moment in the vehicle chassis control, thus offering significant advantages in improving handling stability. For complex conditions such as high-speed steering, emergency obstacle avoidance, and low-traction road surfaces, the vehicle control system often needs to output a stable yaw moment in a timely manner based on the vehicle's motion state (such as sideslip angle and yaw rate) to keep the vehicle in the desired driving state. Therefore, higher demands are placed on the vehicle chassis motion control algorithm.

[0003] Model predictive control (MPC) is widely used for vehicle chassis stability control due to its excellent multi-objective optimization capabilities, which can handle system dynamic constraints, input constraints, and tire nonlinear constraints within the optimization framework. Traditional MPC typically uses a fixed state weight matrix Q and control weight matrix R. However, the control objectives of vehicles vary significantly under different operating conditions, and a fixed matrix cannot simultaneously meet these diverse control requirements.

[0004] While existing research has proposed various methods for determining vehicle state (such as those based on phase diagrams and sideslip angle thresholds), they often lack a mechanism to integrate the state determination results with MPC weight design. In summary, current technologies struggle to achieve real-time adaptive adjustment of MPC weights according to vehicle state, thus limiting control performance under complex and extreme conditions. Summary of the Invention

[0005] This invention aims to overcome the key problem of insufficient adaptability to operating conditions caused by fixed weights in existing MPC control methods. It proposes a variable-weight model predictive control (MPC) distributed drive vehicle chassis control method employing an adaptive weight adjustment strategy. This method selects and switches multiple sets of QR weight matrices designed for different operating conditions based on the vehicle's operating state. This allows the MPC to automatically adjust the control objective under different conditions such as normal driving, understeer, and drift tendencies, thereby significantly enhancing control performance and system robustness.

[0006] To achieve the desired performance, the specific usage scheme of this invention is as follows:

[0007] A variable weighted MPC vehicle chassis control method includes the following steps:

[0008] S1. Vehicle reference model establishment and MPC optimization problem construction;

[0009] A vehicle dynamics model is established, including changes in lateral motion, yaw motion, and sideslip angle. The selected state variables are yaw rate and sideslip angle, and the control input is yaw moment, which is generated by the difference in driving torque between the left and right wheels of the distributed drive system.

[0010] Vehicle slip angle The dynamics are: (1);

[0011] yaw rate The dynamics are: (2);

[0012] The parameters in the formula have the following meanings: The sideslip angle is the angle between the vehicle's center of gravity and its body. The yaw rate of the vehicle body. The yaw acceleration of the vehicle body. For vehicle quality, For longitudinal velocity, These are the distances from the center of mass to the front and rear axles, respectively. Let be the moment of inertia of the vehicle about its vertical axis. These are the front and rear wheel end lateral stiffness, respectively. The yaw torque is generated by the torque difference between the left and right wheels;

[0013] Based on the above dynamic differential equations, the classical state-space expression for a continuous system is established: (3); For state vectors, To control the input vector; where, ;

[0014] System Matrix With input matrix They are respectively: ;

[0015] Next, the sampling period is adopted. Discretizing the system yields: (4); and These are the discretized state vectors for the current time step and the next time step, respectively. Similarly; where the discretized system matrix With input matrix They are respectively: , ;

[0016] This discrete model serves as a prediction model for MPC.

[0017] Next, a standard MPC optimization problem is constructed for the above discrete model, including the cost function, prediction domain, control domain, and physical constraints.

[0018] The cost function is in the following form: (5);

[0019] In the formula, Let be the cost function of the current optimization process, which aims to minimize this cost function. As mentioned earlier, The state weight matrix is... This is the input weight matrix. To predict the length of the sequence.

[0020] At the same time, due to the limitations of the physical model, it is often necessary to add some physical constraints to the solution process, such as input amplitude constraints (limiting the yaw moment range). ), etc. Below are some commonly used distributed drive tire torque constraints in stability control:

[0021] 1) The yaw moment is generated by the difference in driving torque between the left and right sides: (6);

[0022] 2) The torque of the left and right wheels must meet the adhesion restrictions: (7);

[0023] In the formula, The track width is the lateral distance between the center lines of the left and right wheels. and These are the driving and braking forces for the left and right wheels, respectively; The road surface adhesion coefficient, and These represent the vertical loads on the left and right wheels, respectively. This is the rolling radius of the wheel.

[0024] S2. Variable Weight QR Design;

[0025] The strategy of this invention is based on the system control mode in real time. (Agile mode, normal mode, stable mode) Dynamically adjust the weight matrix of MPC. and To achieve optimal vehicle performance under different operating conditions.

[0026] This invention dynamically adjusts the state cost matrix based on mode switching: (8);

[0027] The control input weights are: (9);

[0028] In the formula, Cost matrix Zhongyu Related weights, Cost matrix Zhongyu Related weights, To control the input weight matrix Zhongyu Relevant weights;

[0029] As shown in Table 1, this invention employs three control modes. :

[0030] The first type, In instability mode, the primary control objective is to improve stability. When the vehicle is in a limit state, approaching oversteer or about to become unstable (e.g., the vehicle has a tendency to fishtail), this mode is activated to significantly suppress sideslip angle and yaw rate. At this time, the output is a direct yaw torque that is opposite to the direction of the vehicle's yaw rate. (10);

[0031] At the same time, the input weights are controlled to be relatively large to avoid violent fluctuations in the yaw torque: (11);

[0032] Its design logic is as follows: the sideslip angle must be strongly suppressed to prevent divergence, and the yaw moment control must be gentle to avoid exciting tail-flip instability.

[0033] The second type, In normal mode, the vehicle operates under non-extreme conditions, and performs routine trajectory planning and tracking control based on the driver's or controller's objectives. When the vehicle is in a normal, stable region, lateral and longitudinal coupling performance is prioritized. In this mode, the system does not employ aggressive stabilization control to maintain comfort and energy efficiency.

[0034] The third type, In Agile Mode, the primary goal is to improve the vehicle's cornering agility. When understeer (pushing) is detected, the design weight is: (12);

[0035] At this point, it is necessary to quickly increase the yaw rate to allow the vehicle to recover its target steering trend as soon as possible in order to improve handling agility. At this time, the output is a direct yaw torque in the same direction as the vehicle's yaw rate.

[0036] Table 1. Weighted Qualitative Design Table

[0037]

[0038] S3. Mode switching based on state determination;

[0039] It is necessary to identify the dynamic operating conditions of the vehicle in a reasonable manner to support weight switching and control strategy selection. Typically, vehicle state determination can be based on:

[0040] Yaw rate error : (13);

[0041] In the formula These represent the actual and reference values ​​of the yaw rate, respectively.

[0042] Lateral acceleration : (14);

[0043] Classic indicators such as tire slip angle or its corresponding threshold model have been completed.

[0044] For example, by comparing the deviation between the actual yaw rate and the reference model, understeer tendencies can be identified; by monitoring whether the sideslip angle is close to the empirical limit, it can be determined whether the vehicle may enter the oversteer or instability region; if the vehicle's lateral acceleration is still within a safe range, the vehicle can be considered to be in a normal stable operating condition. The above-mentioned judgment methods are commonly used operating condition identification methods in the field of vehicle dynamics control. They can achieve basic classification of vehicle dynamic states without relying on complex sensors or external models, providing the necessary logical basis for subsequent control mode switching.

[0045] To avoid control jitter caused by frequent switching, this invention introduces two enhancement measures:

[0046] Hysteresis band and average dwell time constraints.

[0047] Hysteresis band design: when the system from Switch to When Agile mode is activated, insufficient steering conditions are often accompanied by noise and disturbances, which may lead to frequent switching back and forth. Therefore, a dead zone term is introduced. The switching logic is as follows: (15);

[0048] In the formula, represent The pattern of time, This is a self-designed mode switching threshold. This hysteresis band effectively avoids high-frequency switching.

[0049] Average Dwell Time Constraint: To further ensure the stability of the control system, this invention employs an average dwell time constraint to limit the number of mode switching operations. (16);

[0050] In the formula, Represents any two moments before or after. Indicates the time interval during which it is activated. The number of mode switches that occur within the time frame. This represents the jitter compensation item, here. To allow for a limited number of switching events within a short period of time, an upper bound for jitter is defined. This represents the average dwell time, used to limit the switching frequency of the system over a longer timescale.

[0051] This constraint ensures that the mode does not switch too frequently, the yaw moment command is smooth, and the MPC solution is more stable.

[0052] Compared to conventional control strategies with fixed weights, this invention demonstrates significant advantages in high-dynamic conditions such as extreme curves and racing tracks. Simulation results show that this method effectively reduces vehicle sideslip error in lateral path tracking, keeping the root mean square value of heading error within 10% of that of traditional nonlinear MPC. In longitudinal speed tracking, this invention achieves smaller speed deviations and smoother acceleration / deceleration responses. Furthermore, thanks to the weight switching strategy employed in this invention, its computational efficiency is comparable to lightweight control methods (linear constant coefficient MPC), with average and peak computation times controlled within 3% and 1% of the control method, respectively, thus ensuring the algorithm's real-time performance. Attached Figure Description

[0053] Figure 1 This is a flowchart of the present invention;

[0054] Figure 2 This is a schematic diagram of a three-degree-of-freedom monorail model of a vehicle, as shown in the embodiment.

[0055] Figure 3 The vehicle yaw stability phase diagram is shown in the example. Detailed Implementation

[0056] The specific technical solutions of the present invention will be described with reference to the embodiments. For example... Figure 1 As shown, it includes the following steps:

[0057] Step 1: Vehicle reference model establishment and MPC optimization problem construction;

[0058] like Figure 2 As shown, this embodiment uses a three-degree-of-freedom single-track model, including longitudinal, lateral, and yaw degrees of freedom. The vehicle dynamics model is... Figure 2 The differential equations for vehicle dynamics are given below: (17);

[0059] The parameters in the formula have the following meanings: For vehicle quality, These are the longitudinal and lateral velocities of the vehicle body, respectively. The yaw rate is angular velocity. These are the distances from the center of mass to the front and rear axles, respectively. Let be the moment of inertia of the vehicle about its vertical axis. Let xy represent the longitudinal and lateral forces of the tire, and fr represent the front and rear axles, respectively. The yaw torque is generated by the torque difference between the left and right wheels. The trajectory tracking deviation is also defined. (18);

[0060] In the formula This refers to the deviation between the actual centroid position and the reference path. The heading angle of the vehicle at the current moment. This refers to the deviation between the actual vehicle heading angle and the reference heading angle. It is the slope of the reference trajectory at the current position, which is related to the specific position of the vehicle. Based on the above dynamic differential equation, the classical expression of the state space of the continuous system is established according to equation (3). (19);

[0061] in, For the front wheel steering angle, The total output torque of the vehicle, combined with the yaw torque. The output torque of each wheel end can be obtained by distributing the torque.

[0062] This embodiment uses a cost function of the form (5). The objectives of this control strategy include trajectory tracking ( (Keep the value as small as possible) and speed tracking ( The weights are designed to be as close as possible to the reference values, so the weight matrix Q is a diagonal matrix with four diagonal elements. Matrix R, on the other hand, is a diagonal matrix with three elements. The weight values ​​of matrix R can affect the output matrix. The degree of change. The following is in matrix form: (20); (twenty one);

[0063] In the formula, for The maximum value, and the same applies to the others; for The maximum value of the range of variation (the difference between the maximum and minimum values) is the denominator value, which is generally a constant obtained based on experience in actual use.

[0064] In this calculation, the values ​​in the numerators of each term are the nominal weights of that quantity, and the values ​​in the denominators are the normalization coefficients of that term, which are often derived from the actual range of change of the control target.

[0065] Step 2: Variable Weight QR Design;

[0066] This invention strategy works by adjusting the system control mode in real time. (Agile mode, normal mode, stable mode) Dynamically adjust the weight matrix of MPC. and To achieve optimal vehicle performance under different operating conditions. This embodiment, based on matrices (20) and (21), specifically utilizes the design logic of the first part:

[0067] weight matrix and It will dynamically adjust based on the current mode. When the system is in stable mode ( When the control objective prioritizes vehicle stability, the longitudinal velocity tracking is assigned zero weight in this mode, while the weight of the lateral velocity-related terms is increased to enhance the vehicle's lateral stability.

[0068] When the system is not in a potentially over-direction state (Si≠1), the algorithm further distinguishes between normal mode and agile mode. In normal mode (typically...) The vehicle is not at its limits, therefore both longitudinal and lateral control objectives are considered simultaneously; at the same time, a high penalty is applied to the direct yaw moment to effectively disable this control channel. Conversely, when a potential understeer tendency of the vehicle is detected ( When the front wheel angle and the direct yaw moment are engaged, the direct yaw moment will be activated to improve the vehicle's agility. At this time, both the front wheel angle and the direct yaw moment will affect the vehicle's directional response.

[0069] It should be emphasized that the direction of the direct yaw moment is determined by the switching parameters. Decision, and Having opposite values ​​in different modes (e.g.) Corresponding to stable mode, (Corresponding to Agile mode).

[0070] Table 2 shows the nominal weights used in the specific implementation process.

[0071] Table 2. Nominal weight design table for the example scenario.

[0072]

[0073] Step 3: Mode switching based on state determination;

[0074] This embodiment introduces a rule-based state determination and mode switching logic based on the centroid sideslip angle β phase diagram at a longitudinal vehicle speed of 10 m / s. The centroid sideslip angle phase diagram is as follows: Figure 3 As shown.

[0075] The red trajectories in the central region of the phase diagram converge densely to the equilibrium point. This indicates that the vehicle possesses strong self-stability within this region. Therefore, this invention defines the region within the envelope of the two central dashed lines as the "normal stability zone." Within this region, the vehicle's yaw response essentially follows a linearized dynamic model and does not require additional yaw moment assistance.

[0076] To enhance the controller's response capability in edge-stable states, this invention further defines the narrow region between the two black central dashed lines as a "control intervention zone." This region can be considered as an asymptotically stable envelope approximated by a linear invariant system, and its boundary satisfies the form: (twenty two);

[0077] in and All are constant coefficients, with a numerical range of approximately When the system state enters this control intervention zone, even if it has not completely left the stable region, this invention still allows MPC to intervene in advance through direct yaw moment. or front wheel steering angle Coordinated control improves the lateral stability margin of the vehicle.

[0078] When the sideslip angle increases negatively but its rate of change is insufficient to generate a corresponding yaw acceleration, and the state point crosses the left envelope of the control intervention zone, the sideslip angle will develop along a divergent trajectory to the left, resulting in insufficient yaw response. This invention defines this as an understeering state.

[0079] Conversely, when When the yaw rate is positive and its rate of change exceeds the upward trend of the right envelope, the system trajectory exhibits a rapid outward expansion along the right side, indicating that the sideslip angle is deteriorating rapidly. The state is determined to be oversteering or a potential tail-drift state. At this time, yaw instability dominates, requiring MPC to immediately increase the lateral stability control weight and apply a reverse direct yaw moment to suppress the divergence trend.

[0080] This phase diagram-based determination method not only intuitively reflects the dynamic evolution of the vehicle under different sideslip conditions, but also provides operable switching logic for the variable weight MPC proposed in this invention, enabling the controller to achieve natural and smooth condition identification between normal response, early intervention, assisted steering and inhibited steering to ensure stability.

[0081] To avoid control jitter caused by frequent switching, this invention introduces two enhancement measures: hysteresis band and average dwell time constraint. The same determination method as in the first part is used.

[0082] This invention proposes a variable weight model predictive control method based on vehicle dynamic conditions. By designing independent MPC weight matrices for different vehicle states and switching them in real time according to the operating conditions, this invention enables the controller to adaptively adjust the control objective between agility and stability, overcoming the performance limitation of traditional fixed cost function MPC under all operating conditions.

[0083] This invention constructs a stable and reliable multi-mode switching logic to ensure smooth transitions in control strategies. By combining vehicle state determination, hysteresis bands, and dwell time mechanisms, this invention achieves smooth switching between control modes, effectively avoiding control jitter caused by frequent switching and improving the robustness and engineering feasibility of the variable weight control strategy.

Claims

1. A variable weighted MPC vehicle chassis control method, characterized in that, Includes the following steps: S1. Vehicle reference model establishment and MPC optimization problem construction; A vehicle dynamics model is established, including changes in lateral motion, yaw motion, and sideslip angle. The selected state variables are yaw rate and sideslip angle, and the control input is yaw moment, which is generated by the difference in driving torque between the left and right wheels of the distributed drive system. For the standard MPC optimization problem of model building, including cost function, prediction domain, control domain and physical constraints; S2. Variable Weight QR Design; By real-time based on system control mode Dynamically adjust the weight matrix of MPC and To achieve optimal vehicle performance under different operating conditions; among them, In These are Agile Mode, Normal Mode, and Stable Mode, respectively. S3. Mode switching based on state determination; Identify the dynamic operating conditions of the vehicle to support weight switching and control strategy selection.

2. The variable weighted MPC vehicle chassis control method according to claim 1, characterized in that, The specific method for S1 is as follows: Vehicle slip angle The dynamics are: （1）； yaw rate The dynamics are: （2）； The parameters in the formula have the following meanings: The sideslip angle is the angle between the vehicle's center of gravity and its body. The yaw rate of the vehicle body. The yaw acceleration of the vehicle body. For vehicle quality, For longitudinal velocity, These are the distances from the center of mass to the front and rear axles, respectively. Let be the moment of inertia of the vehicle about its vertical axis. These are the front and rear wheel end lateral stiffness, respectively. The yaw torque is generated by the torque difference between the left and right wheels; Based on the above dynamic differential equations, the classical state-space expression for a continuous system is established: （3）； For state vectors, To control the input vector; where, ， System Matrix With input matrix They are respectively: ， Using sampling period Discretizing the system yields: （4）； and These are the discretized state vectors for the current time step and the next time step, respectively. Similarly; where the discretized system matrix With input matrix They are respectively: ， ； This discrete model serves as a prediction model for MPC; A standard MPC optimization problem is constructed for the above discrete model, including the cost function, prediction domain, control domain, and physical constraints; The cost function is in the following form: （5）； In the formula, Let be the cost function of the current optimization process, which aims to obtain the smallest possible cost function. The state weight matrix is... The input weight matrix; To predict the length of the sequence; Add physical constraints to the solution process.

3. The variable weighted MPC vehicle chassis control method according to claim 2, characterized in that, The physical constraints include distributed drive tire torque constraints in stability control: 1) The yaw moment is generated by the difference in driving torque between the left and right sides: （6）； 2) The torque of the left and right wheels must meet the adhesion restrictions: （7）； In the formula, The track width is the lateral distance between the center lines of the left and right wheels. and These are the driving and braking forces for the left and right wheels, respectively; The road surface adhesion coefficient, and These represent the vertical loads on the left and right wheels, respectively. This is the wheel's rolling radius.

4. The variable weighted MPC vehicle chassis control method according to claim 1, characterized in that, The specific method for S2 is as follows: Dynamically adjust the state cost matrix based on mode switching: （8）； The control input weights are: （9）； In the formula, Cost matrix Zhongyu Related weights, Cost matrix Zhongyu Related weights, To control the input weight matrix Zhongyu Relevant weights; Three control modes are adopted : In instability mode, the main control objective is to improve stability. When the vehicle is in extreme condition, close to oversteer or about to become unstable, this mode is activated to significantly suppress the sideslip angle and yaw rate. At this time, the output is a direct yaw torque that is opposite to the direction of the vehicle's yaw rate. （10）； Simultaneously, the input weights are controlled to avoid drastic fluctuations in the direct yaw torque. （11）； Side slip angle must be strongly suppressed to prevent divergence, and yaw moment control must be gentle to avoid triggering tail-flip instability; In normal mode, the vehicle is in non-extreme operating conditions, and the vehicle performs conventional trajectory planning and tracking control based on the driver's or controller's target. When the vehicle is in a normal stable area, lateral and longitudinal coupling performance is given priority. In this mode, the system does not use aggressive stabilization control in order to maintain comfort and energy efficiency. In Agile Mode, the primary goal is to improve the vehicle's cornering agility; when understeer (pushing) is detected, the design weight is: （12）； At this point, it is necessary to quickly increase the yaw rate to allow the vehicle to recover its target steering trend as soon as possible in order to improve handling agility. At this time, the output is a direct yaw torque in the same direction as the vehicle's yaw rate.

5. The variable weighted MPC vehicle chassis control method according to claim 1, characterized in that, The specific method for S3 is as follows: Vehicle status determination can be based on: Yaw rate error : （13）； In the formula These represent the actual and reference values ​​of the yaw rate, respectively. Lateral acceleration : （14）； Classic indicators such as tire slip angle or its corresponding threshold model have been completed; Two enhancement measures are introduced: Hysteresis band and mean dwell time constraints; Hysteresis band design: when the system from Switch to That is, when Agile mode is activated, dead zone items are introduced. The switching logic is as follows: （15）； In the formula, represent The pattern of time, The modal switching threshold is a self-designed threshold. Average Dwell Time Constraint: The average dwell time constraint limits the number of mode switches. （16）； In the formula, Represents any two moments before or after. Indicates the time interval during which it is activated. The number of mode switches that occur within the time frame; This represents the jitter compensation item, here. To allow for a limited number of switching events within a short period of time, and... This represents the average dwell time, used to limit the switching frequency of the system over a longer timescale.

Images