Mechanism-fused vehicle dynamics data driven modeling and path tracking control method

By establishing a data-driven prediction model of the three-degree of freedom mechanism model of vehicle dynamics and the fusion mechanism, the problems of insufficient nonlinear processing of vehicle dynamics modeling and lack of physical significance in the data-driven method in the prior art are solved, and a higher level of Modeling accuracy and adaptability improve the control performance and safety of smart cars.

CN118938660APending Publication Date: 2024-11-12CHONGQING UNIV

Patent Information

Application Number
CN202410982643.4
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-07-22
Publication Date
2024-11-12

AI Technical Summary

Technical Problem

The existing vehicle dynamics modeling methods lack accuracy when dealing with nonlinearity and complexity, and the data-driven methods lack physical meaning explanations, making them difficult to generalize, and the generalization ability and interpretability of pure neural network modeling are poor, and they cannot be fully considered. The timing correlation of automobile dynamics samples and the coupling of transverse and longitudinal state feature quantities.

Method used

A vehicle dynamics data-driven modeling and path tracking control method with fusion mechanism was designed. By establishing a three-degree of freedom mechanism model of vehicle dynamics, a data-driven prediction model of fusion mechanism was constructed and embedded in the model prediction control algorithm. A path tracking error model is constructed, constraints and cost functions are established, and the dynamic characteristics of the horizontal and vertical coupling of the vehicle are considered, and the GRU-FNN backbone network model is used for prediction and control.

Benefits of technology

It improves the accuracy and adaptability of the model, improves the accuracy and response characteristics of speed tracking control, enhances the accuracy and stability of path tracking control, and improves the control performance and safety of smart cars in complex traffic environments.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118938660A_ABST
    Figure CN118938660A_ABST
Patent Text Reader

Abstract

The invention belongs to the field of intelligent automobile modeling, and discloses a mechanism-fused vehicle dynamics data-driven modeling and path tracking control method, which comprises the following steps of: S1, establishing a vehicle dynamics three-degree-of-freedom mechanism model; s2, constructing a vehicle dynamics data driving prediction model of a fusion mechanism; s3, embedding the vehicle dynamics data driving prediction model of the fusion mechanism into a model prediction control algorithm; s4, constructing a vehicle path tracking error model; and S5, establishing a constraint condition and a cost function. The precision and adaptability of the model, the accuracy and response characteristics of speed tracking control and the accuracy and stability of path tracking control can be improved, and then the control performance and safety of an intelligent automobile in a complex traffic environment are improved.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention belongs to the field of intelligent automobile modeling, and in particular relates to a vehicle dynamics data-driven modeling and path tracking control method with a fusion mechanism. Background Art

[0002] With the rapid development of intelligent driving technology, the importance of vehicle dynamics modeling and path tracking control methods in autonomous driving systems has become increasingly prominent. Traditional vehicle dynamics modeling methods mainly rely on physical models, which are based on the structure and dynamic characteristics of the vehicle and predict the vehicle behavior through precise mathematical descriptions. However, this method has certain limitations in practical applications. For example, the accuracy of the physical model will be affected due to the nonlinearity and complexity of the vehicle. In addition, the determination of model parameters usually requires a large amount of experimental data and expertise, which increases the complexity and cost of model construction.

[0003] In recent years, with the development of big data and artificial intelligence technology, data-driven methods have gradually become an effective alternative. Data-driven modeling methods use machine learning and deep learning algorithms to establish vehicle dynamics models by utilizing a large amount of historical data. This method can effectively handle nonlinear and high-dimensional data and improve the accuracy and robustness of the model. However, pure data-driven methods also have some problems, such as high dependence on data quality and quantity, lack of physical meaning, and difficulty in generalizing to different vehicles and working conditions.

[0004] After consulting relevant patents, it was found that patent CN113657036 discloses a vehicle dynamics simulation method combining a neural network and a physical model, in which the neural network is used to fit some parameters. However, this method essentially relies on a vehicle lateral dynamics mechanism model with low degrees of freedom, and fails to fully consider the nonlinear characteristics of the actual vehicle model and the coupling of lateral and longitudinal dynamics. Patent CN116992573 discloses a data-driven modeling method combining a gated recurrent unit and a feedforward neural network (GRU-FNN), which takes into account the temporal correlation of vehicle dynamics samples and the coupling characteristics of lateral and longitudinal state feature quantities, but it does not consider the mechanism factors and does not apply it to the control system. Summary of the invention

[0005] In view of the shortcomings of simple neural network modeling, such as poor generalization ability and interpretability, the present invention fully considers the temporal correlation of vehicle dynamics samples and the coupling characteristics of lateral and longitudinal state feature quantities, and designs a vehicle dynamics data-driven modeling and path tracking control method with a fusion mechanism, aiming to improve the accuracy and adaptability of the model, the accuracy and response characteristics of speed tracking control, and the precision and stability of path tracking control, thereby enhancing the control performance and safety of smart cars in complex traffic environments.

[0006] The present invention provides a vehicle dynamics data driven modeling and path tracking control method with a fusion mechanism, comprising the following steps:

[0007] S1. Establish a three-degree-of-freedom mechanism model of vehicle dynamics;

[0008] S2. Construct a vehicle dynamics data-driven prediction model with fusion mechanism;

[0009] S3. Embed the vehicle dynamics data-driven prediction model of the fusion mechanism into the model predictive control algorithm;

[0010] S4. Construct a vehicle path tracking error model;

[0011] S5. Establish constraints and cost functions.

[0012] Further, the step S1 includes the following sub-steps:

[0013] S1.1 Make vehicle dynamics assumptions, add nonlinear tire models, and obtain the vehicle lateral and longitudinal coupled dynamics expressions;

[0014] The vehicle dynamics assumptions include: assuming that the vehicle moves only on a horizontal plane, ignoring the longitudinal tilt and lateral tilt of the vehicle; only considering the longitudinal movement of the vehicle along the longitudinal x-axis of the vehicle body, the lateral movement along the lateral y-axis of the vehicle body, and the yaw movement around the vertical z-axis of the chassis; assuming that the relationship between the lateral force and longitudinal force generated by the tire and the tire side slip angle and slip rate is linear; assuming that the vehicle is a rigid body, that is, ignoring the elastic deformation of the vehicle structure and the dynamic response of the suspension system; ignoring the influence of air resistance and side wind force; assuming that the road surface on which the vehicle is traveling is uniform and has consistent friction characteristics, ignoring the influence of the unevenness of the road surface on the vehicle dynamics performance;

[0015] The vehicle lateral and longitudinal coupling dynamics expressions are as follows:

[0016]

[0017]

[0018] F y =Dsin[Carctan(Bα i -E(Bα i -arctan(Bα i )))]

[0019] F yi =C ai α i

[0020] In the formula, F x1,F x2 are the longitudinal forces acting on the front and rear wheels respectively; F y1 ,F y2 are the lateral forces on the front and rear wheels respectively; V y ,V x ,Ψ are the lateral velocity, longitudinal velocity and yaw rate of the vehicle respectively; θ f and θ r are the front and rear wheel steering angles respectively; and are the longitudinal and lateral velocities of the vehicle in the global coordinate system; (X, Y) are the position coordinates of the vehicle; is the yaw angle; F res is the rolling resistance; F y is the tire lateral force; B is the stiffness coefficient; C is the shape coefficient; D is the peak coefficient; E is the curvature coefficient; F yi Indicates F y1 or F y2 ; C ai Indicates the cornering stiffness of the front wheel C af Or the cornering stiffness of the rear wheel C ar ; α 1 and α 2 are the sideslip angles of the front and rear wheels, α i Represents α 1 or α 2 , α i is a small angle; l a Indicates the distance from the front axle to the center of mass of the vehicle; l b represents the distance from the rear axle to the center of mass of the vehicle; m represents the mass of the vehicle; I z represents the moment of inertia;

[0021] S1.2 Determine the input and output of the vehicle dynamics model, i.e., the control quantity and state quantity;

[0022] Control quantity u k =[F x (t),θ f (t)]

[0023] State quantity x k =[V x (t),V y (t),Ψ(t)]

[0024] Where t is the current time; F x (t) is the longitudinal driving force of the vehicle at the current moment; θ f (t) is the front wheel turning angle of the vehicle at the current moment.

[0025] Further, the step S2 includes the following sub-steps:

[0026] S2.1 constructing a vehicle dynamics data-driven prediction model integrating the mechanism model and the gated-feedforward backbone network model according to the vehicle dynamics three-degree-of-freedom mechanism model established in step S1;

[0027] The main structure of the prediction model is divided into two halves: the front and back halves of the backbone network. The front half contains a GRU layer and two fully connected layers, and the back half contains a GRU layer and a regression layer.

[0028] S2.2 constructs the forward calculation process expression of the vehicle dynamics data-driven prediction model at the current moment;

[0029] S2.3 initializes the state of the second half of the backbone network;

[0030] The initialization process satisfies the following formula:

[0031] h 0 =f GRU,init ([x k-l ,u k-l ],...,[x k-1 ,u k-1 ])

[0032] In the formula, [x k-l ,u k-l ],...,[x k-1 ,u k-1 ] is the historical control information and status information of step l; f GRU,init (·) is the GRU initializer function; h 0 is the final hidden state output by the initializer;

[0033] S2.4 divides the training process into two stages. The first stage is the training of the first half of the backbone network, and the second stage is the training of the entire vehicle dynamics data-driven prediction model.

[0034] Among them, the second stage is divided into pre-training and feedback training.

[0035] Further, the step S3 includes the following sub-steps:

[0036] S3.1 build an MPC framework based on data-driven prediction model;

[0037] S3.2 According to the MPC framework, an MPC prediction process of a vehicle dynamics data-driven prediction model based on a fusion mechanism is established;

[0038] S3.3 generates state predictions for a period of time in the future through a data-driven prediction model, and then performs optimization based on the prediction results to generate control signals.

[0039] Further, the step S4 includes the sub-steps of:

[0040] S4.1 Establish path tracking error expression;

[0041]

[0042]

[0043] Where, e is the lateral error, which is the shortest distance from the vehicle's center of mass to the reference path; is the heading error, which is the angle between the center line of the vehicle and the tangent line of the nearest point on the reference line; ref is the desired yaw rate of the vehicle;

[0044]

[0045] Where s is the distance the car travels along the reference path; κ is the continuous curvature of each point in the path;

[0046] S4.2 substitutes the output of the vehicle dynamics data-driven prediction model into the path tracking error model expression to calculate the lateral error and heading error; then, a cost function is established based on the lateral error and heading error, and a penalty term is added to minimize the path tracking error to ensure that the vehicle accurately tracks the predetermined path.

[0047] Further, the step S5 includes the following sub-steps:

[0048] S5.1 Construct control quantity and increment constraints, center of mass sideslip angle constraints, and vehicle dynamics constraints;

[0049] S5.2 Considering the change in the longitudinal velocity of the vehicle and the path tracking error model state output target value of zero, a finite time domain optimization problem expression is established at any discrete time step.

[0050] Beneficial effects:

[0051] The present invention proposes a vehicle dynamics data-driven modeling and path tracking control method with a fusion mechanism, which can improve the accuracy and adaptability of the model, the accuracy and response characteristics of speed tracking control, and the precision and stability of path tracking control, thereby improving the control performance and safety of smart cars in complex traffic environments.

[0052] Other advantages, objectives and features of the present invention will be described in the following description to some extent, and to some extent, will be obvious to those skilled in the art based on the following examination and study, or can be taught from the practice of the present invention. The objectives and other advantages of the present invention can be realized and obtained through the following description. BRIEF DESCRIPTION OF THE DRAWINGS

[0053] Figure 1It is a structural schematic diagram of the main body of a vehicle dynamics data driven prediction model of a fusion mechanism of the present invention;

[0054] Figure 2 It is a schematic diagram of vehicle dynamics analysis of the present invention;

[0055] Figure 3 This is a schematic diagram of network status initialization of the present invention;

[0056] Figure 4 This is a schematic diagram of the feedback training process of the present invention;

[0057] Figure 5 It is a schematic diagram of the model predictive control of the vehicle dynamics data driven prediction model based on the fusion mechanism of the present invention;

[0058] Figure 6 A schematic diagram of vehicle path tracking error of the present invention;

[0059] Figure 7 It is a schematic diagram of the prediction process of VDMGF in the model predictive control of the present invention. DETAILED DESCRIPTION

[0060] In order to make the technical solutions, advantages and purposes of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the described embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work belong to the protection scope of this application.

[0061] like Figure 1 As shown, the present invention provides a vehicle dynamics data-driven modeling and path tracking control method with a fusion mechanism, comprising the following steps:

[0062] S1. Establish a three-degree-of-freedom mechanism model of vehicle dynamics;

[0063] S1.1 Make vehicle dynamics assumptions, add nonlinear tire models, and obtain the vehicle lateral and longitudinal coupled dynamics expressions;

[0064] like Figure 2 Hint, F x1 ,F x2 are the longitudinal forces (along the x-axis) on the front and rear wheels, F y1 ,F y2 are the lateral forces (along the y-axis) on the front and rear wheels, V y ,V x ,Ψ are the lateral velocity, longitudinal velocity and yaw rate of the vehicle, θ f and θ ris the steering angle of the front and rear wheels. The complexity of the vehicle's dynamic model is positively correlated with its degree of freedom. It is advisable to make the following idealized assumptions:

[0065] Assume that the vehicle moves only on a horizontal plane, and ignore the longitudinal tilt (pitch) and lateral tilt (roll) movements of the vehicle. Only consider the longitudinal movement of the vehicle along the longitudinal x-axis of the vehicle body, the lateral movement along the lateral y-axis of the vehicle body, and the yaw movement around the vertical z-axis of the chassis. Assume that the relationship between the lateral force and longitudinal force generated by the tire and the tire sideslip angle and slip rate is linear. Assume that the vehicle is a rigid body, that is, ignore the elastic deformation of the vehicle structure and the dynamic response of the suspension system. Ignore changes in air resistance and wind force, and ignore the effects of air resistance and side wind force. Assume that the road surface on which the vehicle is traveling is uniform and has consistent friction characteristics, and ignore the effects of road surface unevenness such as potholes and changes in the friction coefficient on the vehicle's dynamic performance.

[0066] Supplementing the HBPacejke tire model and ignoring the vehicle's steering system and transmission system, the vehicle's lateral and longitudinal coupled dynamics expressions are as follows:

[0067]

[0068] F y =Dsin[Carctan(Bα i -E(Bα i -arctan(Bα i )))]

[0069] F yi =C ai α i

[0070] In the formula, and are the longitudinal and lateral velocities of the vehicle in the global coordinate system; (X, Y) are the position coordinates of the vehicle; is the yaw angle; F res is the rolling resistance; F y is the tire lateral force; B is the stiffness coefficient; C is the shape coefficient; D is the peak coefficient; E is the curvature coefficient; F yi Indicates F y1 or F y2 ; C ai Indicates the cornering stiffness of the front wheel C af Or the cornering stiffness of the rear wheel C ar ; α 1 and α 2 are the sideslip angles of the front and rear wheels, α i Represents α 1 or α 2 , α iis a small angle; l a Indicates the distance from the front axle to the center of mass of the vehicle; l b represents the distance from the rear axle to the center of mass of the vehicle; m represents the mass of the vehicle; I z represents the moment of inertia;

[0071] S1.2 According to the established mechanism model, the control quantity information is composed of the vehicle longitudinal driving force and the vehicle front wheel angle, and the state quantity information is composed of the longitudinal speed, lateral speed and yaw angular velocity;

[0072] Control quantity u k =[F x (t),θ f (t)]

[0073] State quantity x k =[V x (t),V y (t),Ψ(t)]

[0074] Where t is the current time; F x (t) is the longitudinal driving force of the vehicle at the current moment; θ f (t) is the front wheel turning angle of the vehicle at the current moment.

[0075] S2. Construct a vehicle dynamics data-driven prediction model with fusion mechanism;

[0076] S2.1 constructing a vehicle dynamics data-driven prediction model integrating the mechanism model and the gated-feedforward backbone network model according to the vehicle dynamics three-degree-of-freedom mechanism model established in step S1;

[0077] For the convenience of description and analysis, the gated-feedforward backbone network model integrating vehicle dynamics mechanism is referred to as VDMGF prediction model. Figure 1 As shown in the figure, the GRU-FNN backbone network is divided into two halves, the front half contains a GRU layer and two fully connected layers, and the back half contains a GRU layer and a regression layer. The control quantity information and state quantity information at the current moment are connected in series and upgraded in dimension, and then input into the front half of the backbone network and the mechanism model respectively, and the state quantity is updated to the next moment after processing. The front half of the backbone network is equivalent to making up for the nonlinearity, complexity and uncertainty factors of the vehicle dynamic system lost by the mechanism model, and the mechanism model is equivalent to adding the interpretability missing from the backbone network. The updated state quantity The transition state is formed by superposition The second half of the backbone network receives the hidden state h of the previous moment k , control quantity u k and the current transition state Output after network processing The state x at the previous moment k The second half of the backbone network does not need to be input because the information it contains is already encoded in the GRU. The state prediction is obtained by superimposing each other

[0078] S2.2 constructs the forward calculation process expression of the vehicle dynamics data-driven prediction model at the current moment;

[0079]

[0080]

[0081]

[0082]

[0083] In the formula, f VD (·) represents the mechanism model function, They represent the nonlinear mapping of the first and second half of the backbone network respectively. The number of GRU units in the first and second half of the backbone network are The number of nodes in the two hidden layers of the FNN in the first half of the backbone network is N. 1 ,N 2 . b g The weight matrix and bias matrix of GRU in the first half of the backbone network are: b f The weight matrix and bias matrix of FNN in the first half of the backbone network are: b g2 ,W hx ,b hx The weight and bias matrices of the second half of the backbone network. The input layer dimensions of the first and second half of the backbone network are both 5, and the output layer dimensions are both 3;

[0084] S2.3 initializes the state of the second half of the backbone network;

[0085] Select another layer of GRU as the initializer and let it receive historical data to generate the hidden state of the prediction network. Figure 3 Schematic diagram of GRU init represents the GRU initializer, GF 1 -P represents the whole formed by the first half of the backbone network and the mechanism model in parallel, GF 2 represents the second half of the backbone network, h i is a hidden state. The initialization process satisfies the following formula:

[0086] h 0 =f GRU,init([x k-l ,u k-l ],...,[x k-1 ,u k-1 ])

[0087] In the formula, [x k-l ,u k-l ],...,[x k-1 ,u k-1 ] is the historical control information and status information of step l; f GRU,init (·) is the GRU initializer function; h 0 is the final hidden state output by the initializer;

[0088] S2.4 divides the training process into two stages, namely the training of the first half of the backbone network and the training of the entire VDMGF prediction model; for the first training stage, the loss function for the first half of the backbone network is constructed as follows:

[0089]

[0090] Where k is the discrete time step, T is the total time step of the training set samples, D is the mini-batch size, and loss i is the loss value for each mini-batch of data, is the predicted value of the first half of the backbone network at discrete time step k+1, is the predicted value of the mechanism model at discrete time step k+1, x k+1 is the true value at discrete time step k+1;

[0091] In the second training phase, the real state value of the training set is first used to input the network for training and obtain the pre-training weights, and then the previous pre-training weights are used to continue as follows Figure 4 Schematic feedback training. The loss function of the second training stage is as follows:

[0092]

[0093] In the formula, is the prediction quantity for the second half of the backbone network at discrete time step k+1. During pre-training, it satisfies:

[0094]

[0095] In the formula, x k is the true value of the state quantity of the training set at discrete time step k. During feedback training, it satisfies:

[0096]

[0097] In the formula, is the predicted value of the state quantity of the VDMGF prediction model at discrete time step k.

[0098] S3. Embed the vehicle dynamics data-driven prediction model of the fusion mechanism into the model predictive control algorithm;

[0099] S3.1 build an MPC framework based on data-driven prediction model;

[0100] like Figure 5 Schematically, the MPC algorithm design framework and process based on the data-driven prediction model includes prediction model construction, control target setting, optimization problem establishment, optimization solution and control implementation in step S2. The optimization problem establishment includes the establishment of cost function and constraint conditions, and the optimization solution and control implementation include rolling optimization and feedback correction;

[0101] S3.2 According to the MPC framework, an MPC prediction process of a vehicle dynamics data-driven prediction model based on a fusion mechanism is established;

[0102] Let t be the start time of the optimization process and τ be the discretized time step. The changes in the input of the VDMGF prediction model at each stage are as follows: Figure 6 Schematically, until finally only relying on the state and control information obtained through optimization;

[0103] S3.3 generates a state prediction for a period of time in the future through a data-driven prediction model, and then performs optimization based on the prediction results to generate a control signal;

[0104] During the operation of the control system in the prediction time domain, the VDMGF prediction model performs each optimization calculation step by step. When the optimization process starts, the VDMGF prediction model first considers the historical state information based on measurement, the current state information and the control actions obtained through optimization; in the next l time steps, the input of the model is expanded to include the vehicle state information and control actions obtained through measurement and optimization. From the l+2 time step onwards, the input of the model is limited to the vehicle state and control information obtained through optimization.

[0105] S4. Construct a vehicle path tracking error model;

[0106] S4.1 As Figure 7 As shown, the path tracking error expression is established;

[0107]

[0108]

[0109] Where, e is the lateral error, which is the shortest distance from the vehicle's center of mass to the reference path; is the heading error, which is the angle between the center line of the vehicle and the tangent line of the nearest point on the reference line; ref is the desired yaw rate of the vehicle; when the reference path is known, assuming that the heading angle deviation of the vehicle is small during driving, Ψ ref It can be obtained by the following formula:

[0110]

[0111] Where s is the distance the car travels along the reference path; κ is the continuous curvature of each point in the path;

[0112] S4.2 The lateral error and heading error are related to the longitudinal velocity, lateral velocity, and yaw rate state of the system. In the model predictive control algorithm, the output of the VDMGF prediction model is substituted into the path tracking error model expression, so that the lateral error and heading error can be calculated. A reasonable cost function is established based on the lateral error and heading error, and a penalty term is added to minimize the path tracking error to ensure that the vehicle follows the predetermined path as much as possible.

[0113] S5. Establish constraint conditions and cost function;

[0114] S5.1 Construct control quantity and increment constraints, center of mass sideslip angle constraints, and vehicle dynamics constraints;

[0115] Control quantity and increment constraint: vehicle longitudinal driving force F x The range should be limited to the maximum and minimum driving / braking forces that the vehicle can generate, F x,min ,F x,max ΔF is the minimum and maximum longitudinal driving force allowed. x,max is the maximum allowable longitudinal driving force increment. Front wheel steering angle θ f The range of θ should be limited to the maximum and minimum steering angles that the vehicle can produce. f,min ,θ f,max is the minimum and maximum allowed front wheel steering angle. f,max is the maximum allowed front wheel steering angle increment.

[0116] Center of mass sideslip angle constraint: In the vehicle path tracking model predictive control algorithm, adding the center of mass sideslip angle constraint can ensure the stability of the vehicle when turning, β k is the vehicle's center of mass sideslip angle at the current moment, β max is the maximum allowable sideslip angle

[0117] S5.2 Considering the change of vehicle longitudinal speed and the path tracking error model state output target value is zero, at any discrete time step, the finite time domain optimization problem can be expressed as:

[0118]

[0119] In the formula, ε is the slack variable, λ v ,λ e , is the weight coefficient of the horizontal and vertical error penalty term, λ 1 ,λ 2 ,λ 3 ,λ 4 are the weight coefficients of the control amount and incremental penalty term, which are used to balance speed tracking performance, operation smoothness and control energy saving. 5 is the penalty weight coefficient of the slack variable. is the expected longitudinal velocity at discrete time step k.

[0120] By solving the optimization problem with a nonlinear programming solver, the optimal control variables can be obtained, including the front wheel steering angle and the longitudinal driving force. x With wheel radius R e When combined, they are combined through a static process to form the total longitudinal drive torque T e =F x ·R e The calculated torque and the front wheel angle signal are input into the physical entity of the controlled vehicle. Similar to the analysis in Chapter 3, in each control cycle, the solver receives the state observation of the current vehicle. And according to step 4, it is converted into tracking error, the control strategy is recalculated based on the latest vehicle tracking error state, and an optimal control sequence is output. The first control input of the sequence is applied to the system, and then the vehicle state is updated to the next moment. The whole process is repeated to realize the vehicle path tracking model predictive control based on the VDMGF prediction model.

[0121] It is hereby declared that the above embodiments are only used to illustrate the technical solutions of the present invention and are not limiting. Although the present invention is described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention can be modified or replaced by equivalents without departing from the purpose and scope of the technical solutions, which should be included in the scope of the claims of the present invention.

Claims

1. A vehicle dynamics data-driven modeling and path tracking control method with fusion mechanism, characterized in that: The following steps are involved: S1. Establish a three-degree-of-freedom mechanism model of vehicle dynamics; S2. Construct a vehicle dynamics data-driven prediction model with fusion mechanism; S3. Embed the vehicle dynamics data-driven prediction model of the fusion mechanism into the model predictive control algorithm; S4. Construct a vehicle path tracking error model; S5. Establish constraints and cost functions.

2. The vehicle dynamics data-driven modeling and path tracking control method of claim 1 is characterized by: The step S1 comprises the following sub-steps: S1.1 Make vehicle dynamics assumptions, add nonlinear tire models, and obtain the vehicle lateral and longitudinal coupled dynamics expressions; The vehicle dynamics assumptions include: assuming that the vehicle moves only in a horizontal plane, ignoring the longitudinal tilt and lateral tilt motion of the vehicle; Only the longitudinal motion of the vehicle along the longitudinal x-axis of the vehicle body, the lateral motion along the lateral y-axis of the vehicle body, and the yaw motion around the vertical z-axis of the chassis are considered; the relationship between the lateral force and longitudinal force generated by the tire and the tire side slip angle and slip rate is assumed to be linear; the vehicle is assumed to be a rigid body, that is, the elastic deformation of the vehicle structure and the dynamic response of the suspension system are ignored; the influence of air resistance and side wind force is ignored; it is assumed that the road surface on which the vehicle is driving is uniform and has consistent friction characteristics, and the influence of the unevenness of the road surface on the vehicle dynamics performance is ignored; The vehicle lateral and longitudinal coupling dynamics expressions are as follows: F y =Dsin[Carctan(Bα i -E(Bα i -arctan(Bα i )))] F yi =C ai a i In the formula, F x1 ,F x2 are the longitudinal forces acting on the front and rear wheels respectively; F y1 ,F y2 are the lateral forces on the front and rear wheels respectively; V y ,V x ,Ψ are the lateral velocity, longitudinal velocity and yaw rate of the vehicle respectively; θ f and θ r are the front and rear wheel steering angles respectively; and are the longitudinal and lateral velocities of the vehicle in the global coordinate system; (X, Y) are the position coordinates of the vehicle; is the yaw angle; F res is the rolling resistance; F y is the tire lateral force; B is the stiffness coefficient; C is the shape coefficient; D is the peak coefficient; E is the curvature coefficient; F yi Indicates F y1 or F y2 ; C ai Indicates the cornering stiffness of the front wheel C af Or the cornering stiffness of the rear wheel C ar ; α1 and α2 are the side slip angles of the front and rear wheels respectively, α i Indicates α1 or α2, α i is a small angle; l a Indicates the distance from the front axle to the center of mass of the vehicle; l b represents the distance from the rear axle to the center of mass of the vehicle; m represents the mass of the vehicle; I z represents the moment of inertia; S1.2 Determine the input and output of the vehicle dynamics model, i.e., the control quantity and state quantity; Control quantity u k =[F x (t),θ f (t)] State quantity x k =[V x (t),V y (t),Ψ(t)] Where t is the current time; F x (t) is the longitudinal driving force of the vehicle at the current moment; θ f (t) is the front wheel turning angle of the vehicle at the current moment.

3. The vehicle dynamics data-driven modeling and path tracking control method of the fusion mechanism according to claim 2 is characterized by: The step S2 comprises the following sub-steps: S2.1 constructing a vehicle dynamics data-driven prediction model integrating the mechanism model and the gated-feedforward backbone network model according to the vehicle dynamics three-degree-of-freedom mechanism model established in step S1; The main structure of the prediction model is divided into two halves: the front and back halves of the backbone network. The front half contains a GRU layer and two fully connected layers, and the back half contains a GRU layer and a regression layer. S2.2 constructs the forward calculation process expression of the vehicle dynamics data-driven prediction model at the current moment; S2.3 initializes the state of the second half of the backbone network; The initialization process satisfies the following formula: h0=f GRU,init ([x k-l ,u k-l ],...,[x k-1 ,u k-1 ]) In the formula, [x k-l ,u k-l ],...,[x k-1 ,u k-1 ] is the historical control information and status information of step l; f GRU,init (·) is the GRU initializer function; h0 is the final hidden state output by the initializer; S2.4 divides the training process into two stages. The first stage is the training of the first half of the backbone network, and the second stage is the training of the entire vehicle dynamics data-driven prediction model. Among them, the second stage is divided into pre-training and feedback training.

4. The vehicle dynamics data-driven modeling and path tracking control method of the fusion mechanism according to claim 3 is characterized by: The step S3 comprises the following sub-steps: S3.1 build an MPC framework based on data-driven prediction model; S3.2 According to the MPC framework, an MPC prediction process of a vehicle dynamics data-driven prediction model based on a fusion mechanism is established; S3.3 generates state predictions for a period of time in the future through a data-driven prediction model, and then performs optimization based on the prediction results to generate control signals.

5. The vehicle dynamics data-driven modeling and path tracking control method of the fusion mechanism according to claim 4 is characterized by: The step S4 comprises the following sub-steps: S4.1 Establish path tracking error expression; Where, e is the lateral error, which is the shortest distance from the vehicle's center of mass to the reference path; is the heading error, which is the angle between the center line of the vehicle and the tangent line of the nearest point on the reference line; ref is the desired yaw rate of the vehicle; Where s is the distance the car travels along the reference path; κ is the continuous curvature of each point in the path; S4.2 substitutes the output of the vehicle dynamics data-driven prediction model into the path tracking error model expression to calculate the lateral error and heading error; then, a cost function is established based on the lateral error and heading error, and a penalty term is added to minimize the path tracking error to ensure that the vehicle accurately tracks the predetermined path.

6. The vehicle dynamics data-driven modeling and path tracking control method of the fusion mechanism according to claim 5 is characterized by: The step S5 comprises the following sub-steps: S5.1 Construct control quantity and increment constraints, center of mass sideslip angle constraints, and vehicle dynamics constraints; S5.2 Considering the change in the longitudinal velocity of the vehicle and the path tracking error model state output target value of zero, a finite time domain optimization problem expression is established at any discrete time step.

Citation Information

Patent Citations

  • Vehicle dynamics data driving modeling method suitable for IVCPS

    CN116992573A

Cited By

  • Intelligent automobile path tracking control method based on dual-network iterative optimization

    CN120686638A