An autonomous commercial vehicle load and road adaptive control system and method

Abstract

The application discloses a kind of automatic driving commercial vehicle load and road self-adaptive control system and method, including vehicle dynamics model part, support vector regression whole vehicle mass identification part, nonlinear model predictive control part, radial basis function neural network part and road classifier part 5 parts.By the commercial vehicle mass identification method based on support vector regression and tire cornering stiffness online correction method, the vehicle mass under different loads is effectively identified, and the tire cornering stiffness is corrected, the ability of vehicle trajectory tracking is improved;Also propose radial basis function neural network compensation lateral control based on, for external disturbance and the parameter error of dynamic model are compensated, further reduce trajectory tracking error;The road classifier proposed simultaneously, can be adjusted in model predictive controller for different road adhesion characteristics Control amount constraint and control increment constraint, effectively improve the road adaptability and safety of commercial vehicle.

Description

TECHNICAL FIELD

[0001] The present application belongs to the technical field of intelligent commercial vehicle automatic driving, in particular to a control system and method for load and road self-adaptation of an automatic driving commercial vehicle. BACKGROUND

[0002] Intelligent vehicle is the key content of modern vehicle engineering research field and the most important part of intelligent transportation system. At present, due to the rapid development of artificial intelligence, the improvement of sensor technology and the substantial improvement of vehicle computer computing power, the research of intelligent passenger car has reached a high degree of automatic driving stage, while the research of intelligent commercial vehicle is still in the stage of partial automatic driving. Among them, the model with fixed mass is generally used in the modeling of current commercial vehicles. Since the commercial vehicle has large load changes during operation, the model with fixed mass will reduce the accuracy of the model and lead to reduced vehicle control effect and greatly reduced trajectory tracking capability. At present, some quality identification methods can accurately estimate the vehicle mass, but they do not consider the influence of mass change on the vehicle side slip characteristics, so the inaccurate tire side slip characteristics will reduce the ability of vehicle lateral control. In addition, for different road adhesion characteristics, especially on low adhesion coefficient road surface, the commercial vehicle under heavy load is prone to large trajectory tracking error or side slip when turning, so it is necessary to improve the road self-adaptation of the commercial vehicle to improve safety. SUMMARY

[0003] In view of the above problems, the present application provides a control system for load and road self-adaptation of an automatic driving commercial vehicle. It mainly includes five parts: vehicle dynamics model part, support vector regression whole vehicle mass identification part, nonlinear model predictive control part, radial basis function neural network part and road classifier part.

[0004] Among them, the vehicle dynamics model part: a vehicle physical model established through some simplifications and assumptions, which will be used as the prediction model in the nonlinear model predictive controller, and the control quantity is obtained by solving this model.

[0005] The support vector regression whole vehicle mass identification part: since the mass of the commercial vehicle will change greatly in actual application scenarios according to the different mass of the goods carried, the support vector regression technology is used to collect part of the state quantity of the current vehicle to estimate the actual mass of the vehicle, and accurate vehicle mass can be obtained to calculate more accurate vehicle tire side stiffness. Accurate vehicle mass and tire stiffness will help to improve the accuracy of parameters in the nonlinear model predictive controller, improve the performance of the nonlinear model predictive controller, reduce the trajectory tracking error, etc.

[0006] Nonlinear model predictive controller part: the role of the nonlinear model predictive controller is to collect the state quantity of the vehicle, and based on the established vehicle dynamics model, while meeting the corresponding constraints, to solve an optimal control quantity acting on the actual vehicle, so as to realize the error between the actual vehicle driving trajectory and the reference trajectory as small as possible.

[0007] Radial basis function neural network part: the nonlinear model predictive controller outputs the control quantity to control the vehicle, due to the existence of external disturbances and other factors, and the model error between the vehicle dynamics model in the nonlinear model predictive controller and the actual vehicle, resulting in the actual vehicle in the trajectory tracking and the reference trajectory exist error, therefore the radial basis function neural network can input the lateral trajectory tracking error and the yaw angle error, and then output a steering angle for compensation, which can further reduce the trajectory tracking error.

[0008] Road classifier: because the adhesion coefficient of different roads will change during the driving of the commercial vehicle, especially on low adhesion coefficient road surface, the vehicle should avoid too large front wheel steering angle. Therefore, by using the road classifier, the road surface with different adhesion coefficients can be identified, and the upper and lower limits of the front wheel steering angle constraint can be adjusted online, that is, the constraint in the nonlinear model predictive controller is adjusted, so that the solved control quantity is consistent with the safety of the current road, which helps to improve the safety of trajectory tracking.

[0009] Further, the dynamics model proposed by the present application is a two-degree-of-freedom single-track dynamics model, which does not consider the roll and pitch motion of the vehicle, and the vehicle only moves in the x-o-y plane. The vehicle is front-wheel steering, and the vehicle body coordinate system is in the left-right symmetry plane of the vehicle. The origin of the vehicle mass center is o, the x-axis is the vehicle longitudinal axis, the positive direction is the vehicle head direction, the z-axis positive direction is perpendicular to upward, and the y-axis points to the vehicle side direction, and the positive direction satisfies the right-hand rule. According to Newton's law, the rotational balance and force balance equation at the vehicle mass center is established, and the following expression is obtained:

[0010]

[0011]

[0012]

[0013] In the formula, m represents the mass of the vehicle, x and y represent the coordinates of the vehicle mass center in the vehicle body coordinate system, respectively represent the longitudinal and lateral velocities of the vehicle, respectively represent the longitudinal and lateral accelerations of the vehicle, represents the yaw angular velocity, represents the yaw angular acceleration, I zIz represents the moment of inertia of the vehicle about the z axis, a, b represent the distance from the center of mass to the front and rear axles, respectively, F xf yf Fx, Fy represent the x and y axis forces of the front axle tires, respectively xr xr Fx, Fy represent the x and y axis forces of the rear axle tires, respectively.

[0014] The motion equation of the vehicle center of mass in the inertial coordinate system is:

[0015]

[0016] where, ψ represents the yaw angle, X, Y represent the coordinates in the inertial coordinate system fixed to the ground, V represents the velocity of the vehicle center of mass in the inertial coordinate system.

[0017] The relationship between the longitudinal and lateral forces of the vehicle tires and the forces in the x and y axis directions of the tires is:

[0018]

[0019] where, F lf lr Fx, Fy represent the longitudinal forces on the front and rear wheels, respectively cf cr Fx, Fy represent the lateral forces on the front and rear wheels, respectively, δ f r δf, δr represent the steering angles of the front and rear wheels, respectively, δ r is always equal to 0.

[0020] The calculation methods of the lateral and longitudinal forces on the front and rear wheels are as follows:

[0021]

[0022] where, C cr , C cf Cf, Cr represent the cornering stiffness of the front and rear wheels, respectively, C lf , C lr Cf, Cr represent the longitudinal stiffness of the front and rear wheels, respectively, s f , s r sf, sr represent the slip ratios of the front and rear wheels, respectively.

[0023] The nonlinear model of the vehicle is obtained as:

[0024]

[0025] ​​​​​The support vector regression is used for the commercial vehicle mass identification. The vehicle tire longitudinal driving force can be expressed by the longitudinal acceleration, the whole vehicle mass, the air resistance, the tire rolling resistance and the road slope, as shown in the formula:

[0026]

[0027] In the formula, F res represents the tire longitudinal driving force, ρ represents the air density, represents the vehicle longitudinal acceleration, v x represents the vehicle longitudinal speed, C d represents the air resistance coefficient, A represents the air area, g represents the gravity acceleration, and θ represents the road slope.

[0028] The longitudinal driving force F res , the longitudinal acceleration , the longitudinal speed v x , the rolling resistance coefficient f and the air resistance are extracted in the automobile CAN bus. res , v x , f, F w are taken as the mass identification algorithm input variables.

[0029] The obtained data are standardized, and the method is shown as follows:

[0030]

[0031] In the formula, Z represents the data standardization output, p represents the input data, i.e. F res , v x , f, F w , μ represents the mean value of all input data samples, σ represents the standard deviation of all input sample data,

[0032] F res , v x , f, F w are taken as the mass identification algorithm input variables, and the identification algorithm is the support vector regression, and the identification function is expressed as:

[0033] f(p) = (ω T p) + b

[0034] In the formula, ω represents the normal vector, b represents the displacement term, p represents the input data, and f(p) represents the identification output. The cost function R reg (f) is defined as:

[0035]

[0036] where p i represents the i-th group of input data, r i represents the actual value corresponding to the i-th group of data samples, represents the insensitive loss function, C represents the regularization constant, and M represents the total number of data samples in the training set.

[0037] where, represents:

[0038]

[0039] Introducing slack variables The cost function is represented as:

[0040]

[0041] s.t.f(x i )-r i ≤ε+ξ i ,

[0042]

[0043]

[0044] Introducing Lagrange multipliers μ i ≥0, α i ≥0, The Lagrange function is obtained as follows:

[0045]

[0046] where, is greater than zero, and ε is an error constant that can be adjusted according to requirements.

[0047] Let The partial derivatives of ω, b, ξ, are zero, and the following equation is obtained:

[0048]

[0049] Therefore, the dual problem of support vector regression is obtained, as shown in the following equation:

[0050]

[0051]

[0052]

[0053] The Karush-Kuhn-Trucker condition is used, i.e. the following formula is required:

[0054]

[0055] The identification function of support vector regression is finally obtained as follows:

[0056]

[0057] The quality identification is performed by using support vector regression, and the online obtained vehicle quality is used to correct the tire cornering stiffness, as shown in the following formula:

[0058]

[0059] In the formula, m0 represents the empty weight of the commercial vehicle, m f represents the quality of the commercial vehicle identified by support vector regression, C0 represents the tire cornering stiffness when the commercial vehicle is empty, C new represents the corrected tire cornering stiffness, and the corrected cornering stiffness is used to update the tire cornering stiffness parameter of the prediction model in the model predictive controller.

[0060] The quality identification is performed by using support vector regression, and the online obtained vehicle quality is used to update the quality parameter of the prediction model in the model predictive controller. The state variable ψ and the control variable u in the model predictive control algorithm proposed in the present application are as follows:

[0061]

[0062] u = δ f

[0063] According to the nonlinear dynamics model of the vehicle, the state variable and the control variable, the following vehicle state equation is obtained:

[0064]

[0065] The Euler discretization is performed on the state equation, and the following formula is obtained:

[0066] ψ (t + 1) = F (ψ (t), △u (t) )

[0067] The cost function J is established, as shown in the following formula:

[0068]

[0069] In the formula, △u represents the control increment, l represents the loss function, k represents the current time, t represents the sampling time, N p represents the prediction time domain, and P represents the terminal constraint.

[0070] The model predictive control solves the following formula at each control period:

[0071]

[0072] stψ k+1,t =F(ψ) k,t ,△u k,t ), k=t,…,t+N p -1

[0073] △umin≤△u k,t ≤△umax,k=t,…,t+N C -1

[0074] u k,t =u k-1,t +△u k-1,t k = t, ..., t+N C -1

[0075] δmin≤u k,t ≤δmax,k=t,…,t+N C -1

[0076] In the formula, △umin and △umax represent the minimum and maximum values ​​of the control increment, respectively, and N c This represents the control time domain, where δmin and δmax represent the minimum and maximum steering angles of the front wheels, respectively.

[0077] In this invention, Δumin, Δumax, δmin, and δmax can be varied according to different road adhesion characteristics. After obtaining the optimal solution, a series of input control increments are obtained:

[0078]

[0079] And The first control increment As the actual control input increment, we get:

[0080]

[0081] In this invention, the sampling time t of the model prediction controller is set to 0.02s, and the control time domain N c The value is 1, and the prediction time domain is N. p It is 50.

[0082] This invention addresses the issue of reduced control performance and decreased trajectory tracking capability caused by model accuracy and external disturbances. It proposes a radial basis function (RBF) neural network to compensate for lateral control. The RBF neural network employs a 2-5-1 structure: two neurons in the input layer, five neurons in the hidden layer, and one neuron in the output layer. The input to the neural network is the lateral trajectory tracking error e. cgand yaw angle error Output is front wheel steering angle compensation δ m , the network weight training adopts online adjustment mode.h j is the jth neuron of the hidden layer, and the Gaussian basis function is selected:

[0083]

[0084] In the formula, c j = [c j1 .c j2 ] is the center vector value of the jth hidden layer neuron, b = [b1, …, b5] T is the base width vector of the neural network, w = [w1, …, w5] T is the weight of the neural network output layer.

[0085] The RBF neural network output is:

[0086] δ m (t) = w1h1 + … + w5h5

[0087] The error index of network approximation is:

[0088]

[0089] The online adjustment of parameters b, c j , w j uses the gradient descent method:

[0090]

[0091] In the formula, η represents the learning rate, λ represents the momentum factor, ρ represents the adjustment parameter, and △ρ represents the change amount of the adjustment parameter. △ρ(t) = ρ(t) - ρ(t-1).

[0092] In order to improve the road adaptability of commercial vehicles, the present application proposes a road classifier. The road classifier first detects the braking event by monitoring the pressure of the brake master cylinder to determine whether braking occurs. When braking occurs, the vertical force F y , the longitudinal force F x and the rear wheel longitudinal slip rate s r of the tire are recorded. After braking, a friction curve composed of normalized longitudinal force μ and slip rate is drawn. The normalized longitudinal force is represented as:

[0093] μ = F x / F y

[0094] The road classifier obtains a series of uniformly distributed μ values from the friction curve, μ = [μ1, μ2, …, μ nThe μ value will be used as the neural network (N) in the road classifier. f The input of N is N. f The input layer has n neurons, the hidden layer has 5 neurons, and the output layer has 3 neurons. The output of the neural network is dry road surface [1,0,0], wet road surface [0,1,0], and icy / snowy road surface [0,0,1]. When the neural network output is dry road surface, the control constraints are umin,umax, and the control increment constraints are Δumin,Δumax. When the road surface is wet road surface, the control constraints are updated to 0.8umin,0.8umax, and the control increment constraints are updated to 0.8Δumin,0.8Δumax. When the road surface is icy / snowy, the control constraints are updated to 0.5umin,0.5umax, and the control increment constraints are updated to 0.5Δumin,0.5Δumax.

[0095] Based on the above system, this invention also proposes a load and road adaptive control method for autonomous commercial vehicles. The method involves a nonlinear model-based predictive controller acquiring vehicle state parameters in real time and obtaining control variables by optimally solving the vehicle dynamics model. The tire lateral stiffness and mass in the predictive model are calculated from the support vector regression vehicle mass identification part, i.e., C. new ,m f The constraint part of the model predictive controller is updated online after the road classifier obtains the road surface type. The model predictive controller outputs the front wheel steering angle δ. f The front wheel steering compensation value δ is output by the RBF radial basis function neural network. m The actual front wheel steering angle δ is obtained by summing the results.

[0096] The beneficial effects of this invention are as follows:

[0097] 1. The online mass identification method and tire lateral stiffness correction method for commercial vehicles based on support vector regression proposed in this invention address the problem of reduced accuracy of vehicle dynamics models caused by large load variations in commercial vehicles. The mass identification method effectively identifies the vehicle mass under different loads, and the tire lateral stiffness correction method adjusts the tire lateral stiffness based on the identified mass. The accurate identification of mass and effective correction of lateral stiffness significantly improve the accuracy of vehicle dynamics models and enhance the vehicle trajectory tracking capability.

[0098] 2. The radial basis function (RBF) neural network compensation lateral control proposed in this invention can adaptively and self-adjust nonlinearly compensate for errors caused by external disturbances, parameter errors in the dynamic model, and assumptions when establishing the dynamic model. Its output is combined with the output of the model predictive controller, which significantly improves the trajectory tracking capability of commercial vehicles under complex dynamic model conditions.

[0099] 3、The road classifier proposed in the application can adjust the front wheel rotation angle constraint and the front wheel rotation angle change constraint in the model predictive controller online according to different road adhesion characteristics, compared with the fixed constraint, the road classifier improves the trajectory tracking ability of the commercial vehicle under different road adhesion characteristics, effectively improves the road adaptability of the commercial vehicle, and significantly improves the safety of the commercial vehicle under large load and high speed. BRIEF DESCRIPTION OF DRAWINGS

[0100] Figure 1 The vehicle single-track dynamics model is provided;

[0101] Figure 2 The RBF radial basis function neural network structure is provided;

[0102] Figure 3 The road classifier is provided;

[0103] Figure 4 The control principle diagram of the load and road adaptability of the autonomous driving commercial vehicle is provided. DETAILED DESCRIPTION

[0104] The application provides an automatic driving commercial vehicle load and road adaptability control system and method, which includes a vehicle dynamics model part, a support vector regression whole vehicle mass identification part, a nonlinear model predictive control part, a radial basis function neural network part and a road classifier part. Through the support vector regression based commercial vehicle mass identification method and the tire cornering stiffness online correction method, the vehicle mass under different loads and the tire cornering stiffness are effectively identified and corrected, and the vehicle trajectory tracking ability is improved. The radial basis function neural network based compensation lateral control is also provided, which compensates for external disturbances and parameter errors of the dynamics model, further reduces the trajectory tracking error, and the road classifier is provided, which can adjust the control amount constraint and the control increment constraint in the model predictive controller online according to different road adhesion characteristics, effectively improves the road adaptability and safety of the commercial vehicle.

[0105] The application will be further described below with reference to the drawings.

[0106] Figure 1 The vehicle single-track dynamics model is provided, without considering the roll and pitch motion of the vehicle, the vehicle only has motion in the x-o-y plane. The vehicle is front wheel steering, the vehicle body coordinate system is in the vehicle left-right symmetry plane, the origin of the vehicle mass center is o, the x-axis is the vehicle longitudinal axis, the positive direction is the vehicle head direction, the z-axis positive direction is perpendicular to upward, the y-axis points to the vehicle side direction, and the positive direction satisfies the right-hand rule. According to Newton's law, the rotational balance and force balance equation is established at the vehicle mass center, and the following expression is obtained:

[0107]

[0108]

[0109]

[0110] where m represents the mass of the vehicle, x, y represent the coordinates of the mass center of the vehicle in the body coordinate system, represent the longitudinal and lateral velocities of the vehicle, respectively, represent the longitudinal and lateral accelerations of the vehicle, respectively, represents the yaw rate, represents the yaw angular acceleration, I z represents the moment of inertia of the vehicle about the z axis, a, b represent the distances from the mass center to the front and rear axles, respectively, F xf ,F yf represent the forces in the x and y directions of the front axle tires, respectively, F xr ,F xr represent the forces in the x and y directions of the rear axle tires, respectively.

[0111] The motion equation of the mass center of the vehicle in the inertial coordinate system is:

[0112]

[0113] where represents the yaw angle, X, Y represent the coordinates in the inertial coordinate system fixed to the ground, represents the relationship between the longitudinal force, the lateral force of the vehicle tires and the forces in the x and y directions of the tires of the mass center of the vehicle in the inertial coordinate system,

[0114]

[0115] where F lf ,F lr represent the longitudinal forces received by the front and rear wheels, respectively, F cf ,F cr represent the lateral forces received by the front and rear wheels, respectively, δ f ,δ r represent the steering angles of the front and rear wheels, respectively, δ r is always equal to 0.

[0116] The calculation methods of the lateral forces and the longitudinal forces received by the front and rear wheels are shown as follows:

[0117]

[0118] where C cr , C cf represent the cornering stiffness of the front and rear wheels, respectively, C lf , C lrrespectively denote the longitudinal stiffness of the front and rear wheels, s f r respectively denote the slip ratio of the front and rear wheels.

[0119] The nonlinear model of the vehicle is obtained as:

[0120]

[0121] Figure 2 The RBF neural network structure is adopted, the radial basis function (RBF) neural network is used for compensating the lateral control, the RBF neural network adopts a 2-5-1 structure, namely, the input layer has 2 neurons, the hidden layer has 5 neurons, and the output layer has 1 neuron, the input of the neural network is the lateral trajectory tracking error e cg and the yaw angle error the output is the front wheel steering angle compensation δ m , and the network weight training adopts an online adjustment mode.h j is the jth neuron of the hidden layer, and a Gaussian function is selected:

[0122]

[0123] In the formula, c j =[c j1 .c j2 is the center vector value of the jth hidden layer neuron, b = [b1, …, b5] T is the base width vector of the neural network, w = [w1, …, w5] T is the weight of the output layer of the neural network.

[0124] The output of the RBF neural network is:

[0125] δ m (t) = w1h1 + … + w5h5

[0126] The error index of the network approximation is:

[0127]

[0128] The online adjustment of the parameters b, c j , w j adopts a gradient descent method:

[0129]

[0130] In the formula, η represents the learning rate, λ represents the momentum factor, ρ represents the adjustment parameter, △ρ represents the change amount of the adjustment parameter, and △ρ(t) = ρ(t) - ρ(t-1).

[0131] Figure 3 ​For the road classifier, the road classifier first detects the braking event by monitoring the pressure of the brake master cylinder to determine whether braking occurs, and records the vertical force F y , longitudinal force F x and rear wheel longitudinal slip ratio s r of the tire when braking occurs. After braking, a friction curve composed of normalized longitudinal force μ and slip ratio is drawn. The normalized longitudinal force is expressed as:

[0132] μ = F x / F y

[0133] The road classifier obtains a series of uniformly distributed μ values, μ = [μ1, μ2, …, μ n ] from the friction curve. The μ values will be input to the neural network (N f ) in the road classifier, the number of input layer neurons of N f is n, the number of hidden layer neurons is 5, the number of output layer neurons is 3, and the output of the neural network is dry road surface [1, 0, 0], wet road surface [0, 1, 0] and icy road surface [0, 0, 1]. When the output of the neural network is dry road surface, the control amount constraint is umin, umax, and the control amount increment constraint is Δumin, Δumax. When the road characteristic is wet road surface, the control amount constraint is updated to 0.8umin, 0.8umax, and the control increment constraint is updated to 0.8Δumin, 0.8Δumax. When the road characteristic is icy road surface, the control amount constraint is updated to 0.5umin, 0.5umax, and the control increment constraint is updated to 0.5Δumin, 0.5Δumax.

[0134] Figure 4 The schematic diagram of the control principle of the automatic driving commercial vehicle load and road self-adaption is shown in FIG. 1. When the commercial vehicle tracks the trajectory, the model predictive controller obtains the vehicle state parameters in real time, and obtains the control amount through optimal solution, wherein the tire cornering stiffness and mass in the prediction model are calculated by the tire stiffness correction module and the vehicle mass identification module, respectively, that is, C new , m f . The constraint part in the model predictive controller is updated online after the road classifier obtains the road type. The front wheel steering angle δ f output by the model predictive controller and the front wheel steering compensation value δ m output by the RBF radial basis function neural network are summed to obtain the actual front wheel steering angle δ.

[0135] The above series of detailed descriptions are only specific descriptions of the feasible embodiments of the present application, and are not used to limit the protection scope of the present application. Any equivalent means or changes without departing from the technology of the present application shall be included in the protection scope of the present application.

Claims

1. An autonomous commercial vehicle load and road adaptive control system, characterized by, The vehicle dynamics model part, the support vector regression whole vehicle mass identification part, the nonlinear model predictive control part, the radial basis function neural network part and the road classifier part are included. The vehicle dynamics model part is used as a prediction model in the nonlinear model predictive controller, and a control variable is obtained by solving the model. The support vector regression whole vehicle mass identification part is used for mass identification, and the obtained whole vehicle mass is used for correcting the tire cornering stiffness and updating the mass parameter of the prediction model in the model predictive controller. The support vector regression is used for commercial vehicle mass identification, and the vehicle tire longitudinal driving force can be represented by the longitudinal acceleration, the whole vehicle mass, the wind resistance, the tire rolling resistance and the road slope, as shown in the following formula: where m is the vehicle mass, F res represents the tire longitudinal driving force, p represents the air density, represents the vehicle longitudinal acceleration, v x represents the vehicle longitudinal speed, C d represents the air resistance coefficient, A represents the windward area, g represents the gravitational acceleration, 0 represents the road surface slope, and f represents the rolling resistance coefficient; extracting the longitudinal driving force F in the car CAN bus res , longitudinal acceleration longitudinal speed v x , rolling resistance coefficient f, air resistance F res , v x , f, F w as input variables for the mass identification algorithm The obtained data is standardized, and the method is shown in the following formula: where Z represents the output of the data normalization, p represents the input of the data normalization, μ represents the mean of all data samples, σ represents the standard deviation of all sample data, M represents the total number of data samples in the training set, p i represents the data of the i-th input, F res 、 v x 、f、F w As the input variable of the quality identification algorithm, the identification algorithm is a support vector regression, and the identification function is represented as: f(p) = (ω T p) + b where ω represents a normal vector, b represents a displacement term, p represents input data, and f(p) represents an identified output; a cost function R is defined reg (f) is represented as: In the formula, r i actual value corresponding to the first group of data samples, l ε insensitive loss function, C represents a regularization constant, and M represents the total number of data samples in the training set. The support vector regression is used for mass identification, and the obtained whole vehicle mass is used for correcting the tire cornering stiffness, as shown in the following formula: wherein m0represents the empty mass of the commercial vehicle, m f represents the tire cornering stiffness identified by the support vector regression, C0represents the tire cornering stiffness when the commercial vehicle is empty, C new represents the corrected tire cornering stiffness, which will be used to update the tire cornering stiffness parameter of the prediction model in the model predictive controller; The nonlinear model predictive control part collects the state variables of the vehicle, and based on the established vehicle dynamics model, the optimal solution is obtained under the condition of meeting the corresponding constraints, and an optimal control variable is obtained for the actual vehicle. The radial basis function neural network part inputs the lateral trajectory tracking error and the yaw angle error, and outputs the steering angle for compensating the lateral control of the nonlinear model predictive control part, as shown in the following formula: The radial basis function neural network part adopts a 2-5-1 structure, that is, the input layer has 2 neurons, the hidden layer has 5 neurons, and the output layer has 1 neuron, and the input of the neural network is the lateral trajectory tracking error e cg and the yaw angle error The output is the front wheel steering angle compensation δ m The network weight training adopts an online adjustment mode, h j is the jth neuron of the hidden layer, and a Gaussian basis function is selected: In the formula, c j =[c j1 .c j2 ] represents the center vector value of the j-th hidden layer neuron, b j =[b1,…,b5] T Let w be the basis width vector of the neural network, where w = [w1, ..., w5]. T These are the weights of the output layer of the neural network; The RBF neural network output is as follows: δ m (t) = w1h1+... + w5h5 The error index of network approximation is as follows: Online adjustment of parameters b, c j , w j is done using gradient descent method: In the formula, η represents the learning rate, α represents the momentum factor, ρ represents the adjustment parameter, and Δρ represents the change of the adjustment parameter, Δρ(t) = ρ(t) - ρ(t-1). The road classifier part identifies the road surface with different adhesion coefficients, and adjusts the upper and lower limits of the front wheel steering angle constraint, that is, adjusts the constraint in the nonlinear model predictive controller, so that the obtained control variable meets the safety of the current road, and the road self-adaptive ability of the commercial vehicle is improved.

2. An automatic drive commercial vehicle load and road adaptive control system according to claim 1, wherein, The vehicle dynamics model is established based on the following assumptions: The vehicle does not consider the roll and pitch motion, and only moves in the x-o-y plane, the vehicle is front-wheel steering, the vehicle body coordinate system is in the vehicle left-right symmetry plane, the origin of the vehicle mass center is o, the x axis is the vehicle longitudinal axis, the positive direction is the vehicle head direction, the z axis positive direction is perpendicular to upward, and the y axis points to the vehicle lateral direction, and the positive direction satisfies the right-hand rule.

3. An automatic drive commercial vehicle load and road adaptive control system according to claim 2, wherein, The vehicle dynamics model is a two-degree-of-freedom single-track dynamics model, and the rotational balance and force balance equations are established at the vehicle mass center according to Newton's law, and the following expressions are obtained: where m represents the vehicle mass, x, y represent the coordinates of the vehicle mass center in the body coordinate system, respectively represent the vehicle longitudinal and lateral velocities, respectively represent the vehicle longitudinal and lateral accelerations, represents the yaw angular velocity, represents the yaw angular acceleration, I z represents the vehicle moment of inertia about the z axis, a, b respectively represent the distances from the mass center to the front and rear axles, F xf ,F yf respectively represent the x-axis and y-axis directions of the front axle tire forces, F xr ,F yr respectively represent the x-axis and y-axis directions of the rear axle tire forces; The motion equation of the vehicle mass center in the inertial coordinate system is as follows: wherein denotes the yaw angle, X, Y denote the coordinates in the inertial coordinate system fixed to the ground, denotes the velocity of the vehicle mass center in the inertial coordinate system; The relationship between the longitudinal force and the lateral force of the vehicle tire and the force in the x axis and y axis directions of the tire is as follows: where F lf ,F lr represent the longitudinal forces on the front and rear wheels, respectively, F cf ,F cr represent the lateral forces on the front and rear wheels, respectively, δ f ,δ r represent the steering angles of the front and rear wheels, respectively, δ r is identically equal to 0; The calculation methods of the lateral force and the longitudinal force of the front and rear wheels are as follows: In the formula, C cr , C cf respectively represent the side stiffness of the front and rear wheels, C lf , C lr respectively represent the longitudinal stiffness of the front and rear wheels, s f , s r respectively represent the slip ratio of the front and rear wheels; The nonlinear model of the vehicle is obtained as follows:

4. An automatic drive commercial vehicle load and road adaptive control system according to claim 1, wherein, The sampling time t of the nonlinear model predictive controller is set to 0.02 s, the control horizon N c is 1, and the prediction horizon is N p is 50.

5. An automatic drive commercial vehicle load and road adaptive control system according to claim 1, wherein, The road classifier section, firstly for braking event detection, determines whether braking occurs by monitoring the pressure of the master cylinder, and records the vertical force F y , longitudinal force F x and rear wheel longitudinal slip ratio s r of the tire when braking occurs; after braking ends, a friction curve consisting of normalized longitudinal force μ and slip ratio is drawn, and the normalized longitudinal force is expressed as: μ = F x / F y The road classifier obtains a series of uniformly distributed μ values from the friction curve, μ = [μ1, μ2, …, μ n ], which will be input into the neural network N f in the road classifier. The number of input layer neurons of N f is n, the number of hidden layer neurons is 5, the number of output layer neurons is 3, and the output of the neural network is dry road surface [1, 0, 0], wet road surface [0, 1, 0], and icy road surface [0, 0, 1].

6. An automatic drive commercial vehicle load and road adaptive control system according to claim 5, wherein, When the neural network output is a dry road, the control quantity constraint is umin, umax, the control quantity increment constraint is deltaumin, deltaumax, when the road characteristic is a wet and slippery road, the control quantity constraint is updated to 0.8umin, 0.8umax, the control increment constraint is updated to 0.8deltaumin, 0.8deltaumax, when the road characteristic is an icy and snowy road, the control quantity constraint is updated to 0.5umin, 0.5umax, the control increment constraint is updated to 0.5deltaumin, 0.5deltaumax.

Images