A method for cooperative control of a steering angle actuator and a feel simulator in an SBW system

Abstract

The application discloses a kind of cooperative control method of steering angle actuator and hand feeling simulator in SBW system;S1, sensor group gathers steering data information;S2, the steering data information handled is collected;S3, steering data information is transmitted to control system;S4, steering angle actuator control;S5, feedback data information;S6, hand feeling simulator control;S7, cooperative control and optimization adjustment;The application is analyzed and handled to steering data information by model predictive control, so that steering angle actuator carries out steering control adjustment, model predictive control introduces uncertainty item, improves the precision control of steering angle actuator;Feedback information is dynamically generated by adaptive control, simulates the hand feeling of driver in traditional steering system;And the change of feedback force of hand feeling simulator and wheel angle is synchronous, optimizes driving experience, through machine learning autonomous learning driving habit of driver, so that the control of vehicle is more comfortable.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of intelligent control, and more particularly to a cooperative control method for a steering angle actuator and a feel simulator in an SBW system. BACKGROUND

[0002] With the continuous development of automobile electronic technology and the integration of automobile systems, people can drive a car without the need for traditional mechanical mechanisms to transmit control signals, but through electronic means.

[0003] The SBW system is developed from Fly-By-Wire which is applied to the driving control of an airplane. The technology uses sensors to transmit the driver's input signals to a central processor, and sends signals to the corresponding actuator through the control logic of the central processor to complete the relevant operation of the driver. In this way, the traditional mechanical structure can be replaced to realize electronic SBW for various movements of the car.

[0004] The SBW system mainly consists of a steering wheel module, a steering gear module and an electronic control unit (ECU). In the SBW system, the driver transmits the steering signal to the electronic control unit through the sensor on the steering wheel, and the electronic control unit analyzes and processes the collected signal and transmits the control signal to the steering motor, so as to control the steering torque required by the steering motor to drive the wheels to steer and realize the steering intention of the driver. At the same time, the sensor on the steering wheel feeds back the steering angle and steering acceleration to the electronic control unit, and the electronic control unit sends signals to the steering wheel return torque motor to generate a steering wheel return torque to provide the driver with corresponding sensory information.

[0005] The advantages of the SBW system are: improving the freedom of vehicle design, facilitating the arrangement of the control system; high steering efficiency, fast response and sensitive control; eliminating steering interference, providing a prerequisite for the integration of automatic control, vehicle dynamic control system and vehicle comfort control system; realizing arbitrary setting of transmission ratio to improve the maneuverability of the vehicle; and canceling the mechanical steering column to improve the crash safety and active safety of the vehicle.

[0006] However, the existing technology has some problems: the existing SBW system realizes control operation through a steering angle actuator and a feel simulator, but the cooperation between the steering angle actuator and the feel simulator is particularly important to improve the operation feel of the vehicle. The existing steering angle actuator and feel simulator use a single algorithm for control, which causes high operation delay of the feel simulator and is not convenient for synchronous cooperative operation. Therefore, we propose a cooperative control method for a steering angle actuator and a feel simulator in an SBW system. SUMMARY

[0007] In view of the problems existing in the prior art, the purpose of the present application is to provide a cooperative control method of a steering angle actuator and a feel simulator in an SBW system, which realizes the separate calculation and processing of the SBW system through various algorithms, thereby improving the cooperation between the steering angle actuator and the feel simulator, and improving the synchronization of the steering angle actuator and the feel simulator.

[0008] To achieve the above-mentioned purpose, the present application provides the following technical scheme: a cooperative control method of a steering angle actuator and a feel simulator in an SBW system, comprising the following steps:

[0009] S1, the sensor group collects steering data information: the rotation of the steering wheel generates steering data information, the steering data information includes the real-time collection of the steering angle of the steering wheel through the sensor group, and includes the collection of vehicle speed, driving direction and road conditions, and converts the steering data information into a digital signal;

[0010] S2, processing the collected steering data information: processing the steering data information collected by the sensor group, denoising, cleaning and normalizing the steering data information;

[0011] S3, the steering data information is transmitted to the control system: the processed steering data information is transmitted to the control system through the CAN bus, the control system calculates the steering angle based on the model predictive control, the control system generates the control signal, and the control signal is transmitted to the steering angle actuator through the CAN bus;

[0012] S4, steering angle actuator control: after receiving the control signal generated by the control system, the steering angle actuator adjusts the steering angle of the wheel according to the control signal to ensure that the wheel rotates according to the angle of the control signal;

[0013] S5, feedback data information: the control system receives the steering angle calculated by the model predictive control, the control system combines the vehicle state and the road information, and dynamically generates the feedback information of the steering force and the feel feedback through the adaptive control, and transmits the feedback information to the feel simulator through the CAN bus;

[0014] S6, feel simulator control: the feel simulator simulates the force feedback characteristics in the traditional mechanical steering system according to the feedback information, and the feel simulator adjusts the torque of the steering wheel through the electro-hydraulic driving device to simulate the feel that the driver can feel in the traditional steering system.

[0015] S7, cooperative control and optimization adjustment: the feedback force of the feel simulator and the change of the wheel steering angle are synchronous, and the steering parameters are adjusted through the feedback loop to optimize the driving experience, and the driving habits of the driver are learned through machine learning, and the feedback strategy is gradually optimized.

[0016] Specifically, the sensor group in S1 includes an angle sensor, a force sensor, a speed sensor, a gyroscope, and a vibration sensor.

[0017] The angle sensor is used to detect the rotation angle of the steering wheel.

[0018] The force sensor is used to detect the force of the steering wheel rotation.

[0019] The speed sensor is used to detect the speed of the vehicle.

[0020] The gyroscope is used to detect the direction and inclination of the vehicle.

[0021] The vibration sensor is used to detect the vibration data generated by the road surface conditions when the vehicle is driving.

[0022] Specifically, the control system in S5 transmits the steering angle to the steering angle actuator after calculating the steering angle, and simultaneously generates steering force and haptic feedback information through adaptive control using the steering angle, vehicle state, and road surface information, and transmits the feedback information to the haptic simulator through the CAN bus.

[0023] Specifically, the steering data information and the steering angle information in S2 and S3 are processed in the same way of denoising, cleaning, and normalization.

[0024] The denoising calculation is as follows:

[0025]

[0026] where w t is the weight of each data point of the steering data information and the steering angle information, x i is the input steering data information and steering angle information, N is the window size, and y t is the filtered steering data information and steering angle information.

[0027] And,

[0028] where σ t is the standard deviation of the data in a certain time window. The larger the standard deviation, the more intense the fluctuations of the steering data information and the steering angle information. When the data changes rapidly, more weight is given, and when the data changes smoothly, less weight is given.

[0029] Specifically, the cleaning calculation formula is as follows:

[0030]

[0031] Wherein, x is the value of the data point, μ is the mean of the data set, σ is the standard deviation of the data set, the points with the value of Z greater than 3 or less than-3 are considered as outliers, and the outliers are removed;

[0032] The removed outliers or missing values are mean filled, and the calculation formula of the mean is as follows:

[0033]

[0034] Wherein, x i is the value of non-missing data, and n is the number of non-missing data;

[0035] The calculation formula of the normalization processing is as follows:

[0036]

[0037] Wherein, x i is the input steering data information and steering angle information, x min is the minimum value in the data set, x max is the maximum value in the data set, and the steering data information and steering angle information are scaled to the range of [0, 1].

[0038] Specifically, the calculation formula of the model predictive control in S3 is as follows:

[0039] x(k+1)=Ax(k)+Bu(k)+Δx(k),

[0040] Wherein, x(k) is the state of the system, u(k) is the control input, i.e., the steering data information, A and B are the state transition matrix and control input matrix of the system, and Δx(k) is an uncertainty term, i.e., the vibration influencing factor of the road surface on the vehicle during the vehicle driving process.

[0041] Specifically, the objective of the model predictive control is to minimize the performance index in a limited prediction horizon:

[0042]

[0043] Wherein, x(k|t) is the prediction of the future state at time t, x ref (k|t) is the expected reference trajectory or target state, is the weighted quadratic error term of the state error, Q is the weight matrix, is the weighted quadratic error term of the control input, R is the weight matrix of the control input, u(k|t) is the control input at time t, i.e., the steering data information, and L is the number of steps in the prediction horizon, i.e., the future L steps of the state and control input are considered when optimizing.

[0044] Specifically, the calculation of the adaptive control in S5 is as follows:

[0045] Suppose the reference model is:

[0046]

[0047] where y m is the output of the reference model, u m is the control input of the reference model, A m and B m are parameters of the reference model, is the derivative of the reference model output, A m y m is the interaction term between the state and its own dynamics in the reference model, B m u m is the influence term of the control input on the model state in the reference model.

[0048] The state equation of the actual system is:

[0049]

[0050] where y is the output of the actual system, is the derivative of the system state, u is the actual control input, and A and B are parameters of the actual system.

[0051] Specifically, the controller of the adaptive control is:

[0052]

[0053] where K is the controller gain, is the estimated value of the system parameter, θ is the weight of the expected control input, u is the output of the adaptive control controller, i.e. the signal acting on the controlled system, for adjusting the system state, y is the output of the system, is the quadratic term of the adaptive parameter;

[0054] The error is defined as:

[0055] e = y - y m ,

[0056] Based on the error dynamics, the adaptive controller adjusts the control input of the system to make the error e converge, and the parameter update rule of the adaptive controller is usually based on the parameter estimator:

[0057]

[0058] where represents the adaptive controller parameter, and Γ is the adaptive gain matrix, used to control the rate of parameter update;

[0059] The adaptive control controls the hand feel simulator through the output y of the system.

[0060] Specifically, the machine learning calculation formula is as follows:

[0061]

[0062] Where s t is the current state, a t is the current action, r t+1 is the immediate reward obtained after taking action a t from state s t , γ is the discount factor, indicating the weight of future rewards, and α is the learning rate, controlling the learning step size;

[0063] DQN uses a deep neural network to approximate the Q function, and the input of the network is the current state s t , and the output is the Q value corresponding to each action a, that is, where is the parameter of the neural network, that is, the weight and bias;

[0064] Therefore, the approximation of the Q function is:

[0065]

[0066] The target Q value is calculated by the following formula:

[0067]

[0068] Where: is the parameter of the target network, which is a copy of the Q network and is updated only after every few steps to avoid instability in the training process; is the output of the target network, indicating the Q value of all future actions in the next state s t+1 ;

[0069] Q network update:

[0070] The training goal of DQN is to make the Q value output by the network close to the target Q value, and the mean square error is usually used as the loss function:

[0071]

[0072] The loss function represents the gap between the predicted Q value and the target Q value, and then the parameters of the Q network are updated through the backpropagation algorithm to optimize the error between the target Q value and the predicted Q value;

[0073] Realize the learning of the driver's driving habits and gradually optimize the feedback strategy.

[0074] Technical effects and advantages of the present application:

[0075] When the present application is used, the steering data information of the steering wheel rotation is collected through the sensor group, and the steering data information is calculated, analyzed and processed through model predictive control, so that the control system calculates the steering angle, the control system generates a control signal according to the steering angle, and the control signal is transmitted to the steering angle actuator through the CAN bus, so that the steering angle actuator can complete the steering control and adjustment of the vehicle.

[0076] And the steering angle information is fed back to the control system, combined with the vehicle state and road surface information, and the feedback information of the steering force and the feeling feedback is dynamically generated through adaptive control, and the feedback information is transmitted to the feeling simulator through the CAN bus, so that the feeling simulator can adjust the torque of the steering wheel through the electro-hydraulic driving device to simulate the feeling that the driver can feel in the traditional steering system.

[0077] And the feedback force of the feeling simulator and the change of the wheel angle are synchronous, the steering parameters are adjusted through the feedback loop, the driving experience is optimized, and the driving habits of the driver are learned through machine learning, and the feedback strategy is gradually optimized, so that the control of the vehicle is more comfortable and more in line with the operation feeling of the vehicle.

[0078] And the uncertainty term is introduced in the model predictive control, that is, the vibration influencing factor of the road surface on the vehicle during the vehicle driving process, so that the model predictive control can eliminate the influence of the road surface on the vehicle and reduce the influence of the road surface vibration on the sensor detection of the vehicle, thereby improving the precision control of the steering angle actuator of the vehicle.

[0079] Other features and advantages of the present application will become apparent from the following detailed description of exemplary embodiments thereof, which description should be taken in conjunction with the accompanying drawings. BRIEF DESCRIPTION OF DRAWINGS

[0080] Figure 1 is a step flowchart provided by the present application. DETAILED DESCRIPTION

[0081] In order to make the purpose, technical scheme and advantages of the present application more clear, the present application will be further described in detail in combination with specific embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and do not limit the present application. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor fall within the scope of the present application.

[0082] As Figure 1As shown, the steering angle actuator and the feel simulator in the SBW system are cooperatively controlled by the method provided by the embodiment of the application, which comprises the following steps:

[0083] S1, the sensor group collects steering data information: the rotation of the steering wheel generates steering data information, which includes the steering angle of the steering wheel collected in real time by the sensor group, and includes the collected vehicle speed, driving direction and road conditions, and converts the steering data information into a digital signal;

[0084] S2, processing the collected steering data information: processing the steering data information collected by the sensor group, denoising, cleaning and normalizing the steering data information;

[0085] S3, the steering data information is transmitted to the control system: the processed steering data information is transmitted to the control system through the CAN bus, the control system calculates the steering angle based on the model predictive control, the control system generates the control signal, and the control signal is transmitted to the steering angle actuator through the CAN bus;

[0086] S4, steering angle actuator control: after receiving the control signal generated by the control system, the steering angle actuator adjusts the steering angle of the wheel according to the control signal to ensure that the wheel rotates at the angle of the control signal;

[0087] S5, feedback data information: the control system receives the steering angle calculated by the model predictive control, the control system combines the vehicle state and the road information, and dynamically generates the feedback information of the steering force and the feel feedback through the adaptive control, and transmits the feedback information to the feel simulator through the CAN bus;

[0088] S6, feel simulator control: the feel simulator simulates the force feedback characteristics in the traditional mechanical steering system according to the feedback information, and adjusts the torque of the steering wheel through the electro-hydraulic driving device to simulate the feel that the driver can feel in the traditional steering system;

[0089] S7, cooperative control and optimization adjustment: the feedback force of the feel simulator and the change of the wheel steering angle are synchronous, and the steering parameters are adjusted through the feedback loop to optimize the driving experience, and the driver's operation habit is learned through machine learning, and the feedback strategy is gradually optimized.

[0090] In this embodiment, preferably, the sensor group in S1 includes an angle sensor, a force sensor, a speed sensor, a gyroscope and a vibration sensor;

[0091] The angle sensor is used to detect the rotation angle of the steering wheel;

[0092] The force sensor is used to detect the force of the steering wheel rotation;

[0093] The speed sensor is used for detecting the speed of the vehicle;

[0094] The gyroscope is used for detecting the direction and inclination degree of the vehicle;

[0095] The vibration sensor is used for detecting the vibration data generated by the road surface condition when the vehicle is running;

[0096] It should be noted that the rotation of the steering wheel is detected by the plurality of sensors of the sensor group, and various information of the vehicle is detected, so as to facilitate the control and adjustment of the steering of the vehicle.

[0097] In this embodiment, preferably, the control system in S5 transmits the steering angle to the steering angle actuator after calculating the steering angle, and the control system directly utilizes the steering angle, the state of the vehicle and the road surface information to dynamically generate the feedback information of the steering force and the feedback of the hand feeling through adaptive control, and transmits the feedback information to the hand feeling simulator through the CAN bus;

[0098] It should be noted that the steering angle of the wheel is calculated, and the speed, direction and inclination degree of the vehicle and the vibration data are used for reverse control, so as to simulate the operation feeling of the mechanical vehicle, and after the steering angle is transmitted to the steering angle actuator, the feedback information of the steering force and the feedback of the hand feeling are generated through adaptive control at the same time, so that the steering and the feedback are kept synchronous.

[0099] In this embodiment, preferably, the denoising, cleaning and normalization processing of the steering data information and the steering angle information in S2 and S3 are in the same way;

[0100] The denoising calculation is as follows:

[0101]

[0102] Wherein, w t is the weight of each data point of the steering data information and the steering angle information, x i is the input steering data information and steering angle information, N is the window size, y t is the filtered steering data information and steering angle information;

[0103] And,

[0104] Wherein, σ t is the standard deviation of the data in a certain time window, the larger the standard deviation, the more intense the fluctuation of the steering data information and the steering angle information, and more weight is given when the data changes sharply, and less weight is given when the data changes smoothly;

[0105] It should be noted that the collected steering data information and steering angle information are denoised to improve the accuracy of the steering data information and the steering angle information, and according to the distribution weight, more weight distribution is facilitated during steering, and less weight distribution is facilitated during vehicle vibration, thereby facilitating precise control adjustment.

[0106] In this embodiment, preferably, the calculation formula of the cleaning is as follows:

[0107]

[0108] Wherein, x is the value of the data point, μ is the mean of the data set, σ is the standard deviation of the data set, the points with a value of Z greater than 3 or less than -3 are considered as abnormal values, and the abnormal values are removed;

[0109] The removed abnormal values or missing values are filled with the mean, and the calculation formula of the mean is as follows:

[0110]

[0111] Wherein, x i is the value of the non-missing data, and n is the number of non-missing data;

[0112] The calculation formula of the normalization processing is as follows:

[0113]

[0114] Wherein, x i is the input steering data information and steering angle information, x min is the minimum value in the data set, x max is the maximum value in the data set, and the steering data information and the steering angle information are scaled to the range of [0, 1];

[0115] It should be noted that the abnormal points in the data information are detected and removed by the mean and the standard deviation, and the missing values and the removed values are compensated by the mean, so as to maintain the integrity of the steering data information and the steering angle information, and the steering data information and the steering angle information can be scaled to the range of [0, 1] by the normalization processing, thereby facilitating the reduction of the calculation pressure of the steering data information and the steering angle information.

[0116] In this embodiment, preferably, the calculation formula of the model predictive control in S3 is as follows:

[0117] x(k+1)=Ax(k)+Bu(k)+Δx(k),

[0118] Wherein, x(k) is the state of the system, u(k) is the control input, that is, steering data information, A and B are the state transition matrix and control input matrix of the system, and Δx(k) is an uncertainty term, that is, the vibration influence factor of the road surface on the vehicle during the driving of the vehicle.

[0119] It should be noted that the introduction of the uncertainty term in the model predictive control facilitates the elimination of the vibration influence factor of the road surface on the vehicle, and facilitates the improvement of the precision of the model predictive control.

[0120] In the embodiment, preferably, the target of the model predictive control is to minimize the performance index in a limited prediction time domain:

[0121]

[0122] Wherein, x(k|t) is the prediction of the future state at time t, x ref (k|t) is the expected reference trajectory or target state, is a weighted quadratic error term of the state error, Q is a weight matrix, is a weighted quadratic error term of the control input, R is a weight matrix of the control input, u(k|t) is the control input at time t, that is, steering data information, and L is the number of steps of the prediction time domain, that is, the future L steps of the state and the control input are considered when optimization is performed.

[0123] It should be noted that the model predictive control predicts the future trajectory through multiple steps, facilitates the minimization of the performance index in the prediction time domain, facilitates the calculation and processing of the future N steps of the state and the control input, and enables the model predictive control to calculate the steering angle according to the collected steering data information, and realize the steering processing of the steering angle actuator.

[0124] In the embodiment, preferably, the calculation of the adaptive control in S5 is as follows:

[0125] Suppose the reference model is:

[0126]

[0127] Wherein, y m is the output of the reference model, u m is the control input of the reference model, A m and B m are parameters of the reference model, is the derivative of the output of the reference model, A m y m is the interaction term between the state and its own dynamics in the reference model, B m u m is the influence term of the control input on the model state in the reference model.

[0128] The state equation of the actual system is:

[0129]

[0130] where y is the output of the actual system, is the derivative of the system state, u is the actual control input, and A and B are parameters of the actual system;

[0131] It should be noted that the goal of adaptive control is to adjust the parameters of the controller so that the output of the system can track an ideal reference model output as much as possible. The basic idea is to update the parameters of the controller through error dynamics, and the reference model is used to define the desired behavior or performance of the system. The goal of the controller is to make the output of the actual system track the output of the reference model as much as possible. The state equation of the actual system is usually used to describe the dynamic behavior of the system and reflects the relationship between the input, output and state variables of the system.

[0132] In this embodiment, the controller of the adaptive control is preferably:

[0133]

[0134] where K is the gain of the controller, is the estimated value of the system parameter, θ is the weight of the desired control input, u is the control input or output, i.e. the signal acting on the controlled system, for adjusting the system state, y is the output of the system, is the quadratic term of the adaptive parameter;

[0135] The error is defined as:

[0136] e=y-y m ,

[0137] Based on the error dynamics, the adaptive controller adjusts the control input of the system to make the error e converge. The parameter update rule of the adaptive controller is usually based on the parameter estimator:

[0138]

[0139] where is the adaptive controller parameter, and Γ is the adaptive gain matrix, which controls the rate of parameter update;

[0140] The adaptive control is realized by controlling the haptic simulator through the output y of the system;

[0141] It should be noted that the controller will adjust its control strategy in real time according to the error between the actual output of the system and the output of the reference model. The error is the difference between the actual system output and the reference model output, which is usually used to evaluate the performance of the control system.

[0142] In this embodiment, preferably, the machine learning calculation formula is as follows:

[0143]

[0144] where s t is the current state, a t is the current action, r t+1 is the immediate reward obtained after taking action a t from state s t , γ is the discount factor, indicating the weight of future rewards, and α is the learning rate, controlling the learning step size;

[0145] DQN uses a deep neural network to approximate the Q function, and the input of the network is the current state s t , and the output is the Q value corresponding to each action a , that is, where is the parameter of the neural network, that is, the weight and bias;

[0146] Therefore, the approximation of the Q function is:

[0147]

[0148] The target Q value is calculated by the following formula:

[0149]

[0150] where: is the parameter of the target network, which is a copy of the Q network and is updated only after a certain number of steps to avoid instability during training; is the output of the target network, indicating the Q value of all future actions in the next state s t+1 ;

[0151] Q network update:

[0152] The training goal of DQN is to make the Q value output by the network close to the target Q value, and the mean square error is usually used as the loss function:

[0153]

[0154] The loss function represents the gap between the predicted Q value and the target Q value, and then the parameters of the Q network are updated through the backpropagation algorithm to optimize the error between the target Q value and the predicted Q value;

[0155] Implement autonomous learning of driver's driving habits and gradually optimize feedback strategy;

[0156] It should be noted that the Q value is used to represent the expected cumulative reward of taking an action in a certain state, and the Q value is a state-action value function, and the goal of Q-learning is to learn an optimal Q function, so that the agent can select the optimal action to maximize the long-term return; through continuous iteration and training, DQN can learn an efficient strategy to adapt to complex environments.

[0157] The specific operation process of the application is as follows:

[0158] Step one, the sensor group collects steering data information: the rotation of the steering wheel generates steering data information, the steering data information includes the real-time collection of the steering angle of the steering wheel through the sensor group, and includes the collection of vehicle speed, driving direction and road conditions, and converts the steering data information into digital signals;

[0159] Step two, processing the collected steering data information: processing the steering data information collected by the sensor group, denoising, cleaning and normalizing the steering data information;

[0160] Step three, the steering data information is transmitted to the control system: the processed steering data information is transmitted to the control system through the CAN bus, the control system calculates the steering angle based on the model predictive control, the control system generates the control signal, and the control signal is transmitted to the steering angle actuator through the CAN bus;

[0161] The calculation formula of model predictive control is as follows:

[0162] x(k+1)=Ax(k)+Bu(k)+Δx(k),

[0163] Wherein, x(k) is the state of the system, u(k) is the control input, that is, the steering data information, A and B are the state transition matrix and control input matrix of the system, and Δx(k) is the uncertainty term, that is, the influence factor of the vibration of the vehicle caused by the road condition during the driving of the vehicle;

[0164] The goal of model predictive control is to minimize the performance index in a limited prediction time domain:

[0165]

[0166] Wherein, x(k|t) is the prediction of future state at time t, x ref (k|t) is the expected reference trajectory or target state, is the weighted quadratic error term of state error, Q is the weight matrix, is the weighted quadratic error term of control input, R is the weight matrix of control input, u(k|t) is the control input at t time, i.e., steering data information, L is the number of steps of prediction time domain, i.e., when optimizing, the state and control input of future L steps are considered;

[0167] Step four, steering actuator control: the steering actuator receives the control signal generated by the control system, and adjusts the steering angle of the wheel according to the control signal to ensure that the wheel rotates at the angle of the control signal;

[0168] Step five, feedback data information: the control system receives the steering angle calculated by the model predictive control, and the control system combines the vehicle state and road surface information, and dynamically generates the feedback information of steering force and hand feeling feedback through adaptive control, and transmits the feedback information to the hand feeling simulator through the CAN bus;

[0169] The calculation of adaptive control is as follows:

[0170] Suppose the reference model is:

[0171]

[0172] Where, y m is the output of the reference model, u m is the control input of the reference model, A m and B m are parameters of the reference model, is the derivative of the reference model output, A m y m is the interaction term between the state and its own dynamics in the reference model, B m u m is the influence term of the control input on the model state in the reference model;

[0173] The state equation of the actual system is:

[0174]

[0175] Where, y is the output of the actual system, is the derivative of the system state, u is the actual control input, A and B are parameters of the actual system;

[0176] The controller of adaptive control is:

[0177]

[0178] Where, K is the controller gain, is the estimated value of the system parameter, θ is the weight of the expected control input, u is the output of the controller of adaptive control, i.e., the signal acting on the controlled system, used to adjust the system state, y is the output of the system, is a quadratic term of adaptive parameters;

[0179] The error is defined as:

[0180] e = y - y m ,

[0181] Based on the error dynamics, the adaptive controller adjusts the control input of the system to make the error e converge, and the parameter update law of the adaptive controller is usually based on the parameter estimator:

[0182]

[0183] where, is the adaptive controller parameter, and Γ is the adaptive gain matrix, which controls the rate of parameter update;

[0184] The adaptive control is achieved through the output y of the system to control the haptic simulator;

[0185] Step six, haptic simulator control: the haptic simulator simulates the force feedback characteristics in the traditional mechanical steering system according to the feedback information, and the haptic simulator adjusts the torque of the steering wheel through the electro-hydraulic driving device to simulate the haptic that the driver can feel in the traditional steering system;

[0186] Step seven, coordinated control and optimization adjustment: the feedback force of the haptic simulator and the change of the wheel angle are synchronized, and the steering parameters are adjusted through the feedback loop to optimize the driving experience, and the machine learning autonomously learns the driving habits of the driver and gradually optimizes the feedback strategy;

[0187] The calculation formula of machine learning is as follows:

[0188]

[0189] where, s t is the current state, a t is the current action, r t+1 is the immediate reward obtained after taking action a t from state s t , γ is the discount factor, indicating the weight of future rewards, and α is the learning rate, controlling the learning step size;

[0190] DQN uses a deep neural network to approximate the Q function, and the input of the network is the current state s t , and the output is the Q value corresponding to each action a, that is, where is the parameter of the neural network, that is, the weight and bias;

[0191] Therefore, the approximation of the Q function is:

[0192]

[0193] The target Q value is calculated by the following formula:

[0194]

[0195] Wherein: is the parameter of the target network, and is a copy of the Q network, which is updated only after every several steps to avoid instability in the training process; is the output of the target network, indicating the Q value of the next state s t+1 of all future actions;

[0196] The update of the Q network:

[0197] The training target of the DQN is to make the Q value output by the network close to the target Q value, and the mean square error is usually used as the loss function:

[0198]

[0199] The loss function represents the gap between the predicted Q value and the target Q value, and then the parameters of the Q network are updated through the back propagation algorithm to optimize the error between the target Q value and the predicted Q value;

[0200] To realize the driving habits of autonomous learning drivers and gradually optimize the feedback strategy.

[0201] Finally, it should be pointed out that: the above only for the preferred embodiments of the present application, and not for the purpose of limiting the present application, although the foregoing detailed description of the present application is made with reference to the foregoing embodiments, for those skilled in the art, it still can be modified to the technical solutions recorded in the foregoing embodiments, or to replace some of the technical features, within the spirit and principles of the present application, any modification, equivalent replacement, improvement, etc., should be included in the protection scope of the present application.

Claims

1. A method for coordinated control of a corner actuator and a hand-feel simulator in an SBW system, characterized in that, Includes the following steps: S1. Sensor group collects steering data information: The rotation of the steering wheel generates steering data information, which includes the steering angle of the steering wheel collected in real time by the sensor group, as well as the collected vehicle speed, driving direction and road conditions, and converts the steering data information into digital signals. S2. Process the collected steering data: Process the steering data collected by the sensor group, including noise reduction, cleaning and normalization. S3. Transmission of steering data information to the control system: The processed steering data information is transmitted to the control system via the CAN bus. The control system calculates the steering angle based on model predictive control, generates a control signal, and transmits the control signal to the steering actuator via the CAN bus. S4. Angle actuator control: After receiving the control signal generated by the control system, the angle actuator adjusts the steering angle of the wheel according to the control signal to ensure that the wheel rotates according to the angle of the control signal. S5. Feedback data information: The control system receives the steering angle calculated by the model predictive control. The control system combines the vehicle status and road information, and dynamically generates steering force and feel feedback information through adaptive control. The feedback information is then transmitted to the feel simulator via the CAN bus. The controller for adaptive control is: Where K is the controller gain, Here, θ represents the estimated system parameters, θ is the desired control input weight, u is the output of the adaptive controller (the signal applied to the controlled system to adjust its state), and y is the system output. It is a quadratic term of the adaptive parameter; Error is defined as: e=yy m , Based on error dynamics, an adaptive controller adjusts the system's control input to bring the error e to converge. m The parameter update rule of an adaptive controller is usually based on a parameter estimator, which is the output of the reference model. in, Represented as adaptive controller parameters, Γ is the adaptive gain matrix, used to control the rate of parameter update; Adaptive control achieves control of the hand-feel simulator through the system's output y; S6. Hand feel simulator control: The hand feel simulator imitates the force feedback characteristics of the traditional mechanical steering system based on feedback information. The hand feel simulator adjusts the torque of the steering wheel through an electro-hydraulic drive device to simulate the feel that the driver can feel in the traditional steering system. S7. Cooperative Control and Optimization: The feedback force of the hand-feel simulator is synchronized with the change in wheel angle, and the steering parameters are adjusted through the feedback loop to optimize the driving experience. It learns the driver's driving habits through machine learning and gradually optimizes the feedback strategy.

2. The method for coordinated control of a corner actuator and a hand-feel simulator in an SBW system according to claim 1, characterized in that: The sensor group in S1 includes an angle sensor, a force sensor, a speed sensor, a gyroscope, and a vibration sensor; The angle sensor is used to detect the rotation angle of the steering wheel; The force sensor is used to detect the force of the steering wheel rotation; The speed sensor is used to detect the vehicle's speed; The gyroscope is used to detect the vehicle's direction and tilt. The vibration sensor is used to detect vibration data generated by the vehicle's movement and road conditions.

3. The method for coordinated control of a corner actuator and a hand-feel simulator in an SBW system according to claim 1, characterized in that: After calculating the steering angle, the control system in S5 transmits the steering angle to the steering actuator. At the same time, the control system directly uses the steering angle, vehicle status and road information to dynamically generate steering force and feel feedback information through adaptive control, and transmits the feedback information to the feel simulator through the CAN bus.

4. The method for coordinated control of a corner actuator and a hand-feel simulator in an SBW system according to claim 1, characterized in that: The steering data information in S2 is processed in the same way for noise reduction, cleaning and normalization; The noise reduction is calculated as follows: Among them, w t For the weight of each data point of steering data and steering angle information, x i The input steering data and steering angle information are N, where N is the window size and y is the steering angle. t This refers to the filtered steering data and steering angle information; and, Where, σ t The standard deviation of the data within a certain time window is given. The larger the standard deviation, the more drastic the fluctuation of the steering data and steering angle information. More weight is given when the data changes drastically, and less weight is given when the changes are stable.

5. The method for coordinated control of a corner actuator and a hand-feel simulator in an SBW system according to claim 4, characterized in that: The calculation formula for the cleaning process is as follows: Where x is the value of the data point, μ is the mean of the dataset, σ is the standard deviation of the dataset, and points with Z values ​​greater than 3 or less than -3 are considered outliers and are removed. Outliers or missing values ​​that have been removed are imputed using the mean; and the mean is calculated using the following formula: Where, x' i represents the values ​​of the non-missing data, and n is the number of non-missing data; The calculation formula for the normalization process is as follows: Where, x i For the input steering data and steering angle information, x min It is the minimum value in the dataset, x max It is the maximum value in the dataset, which scales the steering data and steering angle information to the range of [0, 1].

6. The method for coordinated control of a corner actuator and a hand-feel simulator in an SBW system according to claim 1, characterized in that: The calculation formula for model predictive control in S3 is as follows: x(k+1)=Ax(k)+Bu(k)+Δx(k), Where x(k) is the system state, u(k) is the control input, i.e., the steering data information, A and B are the system state transition matrix and control input matrix, and Δx(k) is the uncertainty term, i.e., the factors affecting the vibration of the vehicle due to road conditions during vehicle operation.

7. The method for coordinated control of a corner actuator and a hand-feel simulator in an SBW system according to claim 6, characterized in that: The objective of the model predictive control is to minimize the performance index within a finite prediction time domain: Where x(k|t) is the prediction of the future state at time t, x ref (k|t) is the desired reference trajectory or target state. It is the weighted quadratic error term of the state error, and Q is the weight matrix. R is the weighted quadratic error term of the control input, R is the weight matrix of the control input, u(k|t) is the control input at time t, which is the steering data information, and L is the number of steps in the prediction time domain, that is, when optimizing, the state and control input of the next L steps are considered.

8. The method for coordinated control of a corner actuator and a hand-feel simulator in an SBW system according to claim 1, characterized in that: The adaptive control in S5 is calculated as follows: Assume the reference model is: Among them, y m It is the output of the reference model, u m It is the control input of the reference model, A m and B m These are the parameters of the reference model. It is the derivative of the reference model output, A m y m B is the interaction term between the state and its own dynamics in the reference model. m u m It is the term in the reference model that describes the influence of control inputs on the model state; The state equation of the actual system is: Where y is the output of the actual system. It is the derivative of the system state, u is the actual control input, and A and B are the parameters of the actual system.

9. The method for coordinated control of a corner actuator and a hand-feel simulator in an SBW system according to claim 1, characterized in that: The calculation formula for the machine learning is as follows: Among them, s t This is the current state, a t It is the current action, r t+1 From state s t Take action a t The immediate reward obtained afterward, γ is the discount factor, which represents the weight of future rewards, and α is the learning rate, which controls the step size of learning; DQN uses a deep neural network to approximate the Q-function, with the current state s as the network's input. t The output is the Q-value for each action 'a', i.e., Q(s). t ,a,θ), where θ are the parameters of the neural network, namely the weights and biases; Therefore, the Q function is approximated as: Q(s t ,a,θ)≈Q DQN (s t ,a); The target Q value is calculated using the following formula: Where: θ - These are the parameters of the target network, a copy of the Q-network, updated only after a certain number of steps to avoid instability during training; Q(s) t+1 ,a';θ - ) is the output of the target network, representing the output in the next state s. t+1 Below, the Q value of all future actions; Q network update: The training objective of DQN is to make the network's output Q-value close to the target Q-value, and the mean squared error is typically used as the loss function. L(θ)=E[(y t -Q(s t ,a t ;i)) 2 ]; The loss function represents the difference between the predicted Q value and the target Q value. Then, the parameters θ of the Q network are updated through the backpropagation algorithm to optimize the error between the target Q value and the predicted Q value. It enables autonomous learning of the driver's operating habits and gradually optimizes the feedback strategy.

Images