Steer-by-wire torque compensation system and method based on intelligent prediction

Abstract

The invention discloses a steer-by-wire torque compensation system and method based on intelligent prediction, and belongs to the technical field of vehicle steer-by-wire. Comprising the following steps: acquiring a current state parameter of a vehicle; performing feature processing and time sequence information modeling on the current state parameter to generate a current torque signal; based on the dynamic characteristics of the road feeling motor and in combination with predefined input information and interference information, a vehicle enhanced control system is constructed; predicting dynamic information of the vehicle at a future moment based on the vehicle enhanced control system; determining an optimal control law corresponding to the current state parameter through online retrieval based on a linear function of offline optimization storage; and performing nonlinear friction force compensation on the optimal control law to obtain an optimized torque signal. According to the method, through offline optimization and online search, the calculation burden is relieved, the prediction mode is introduced, and the steering torque compensation precision and the real-time self-adaptive capacity of the vehicle under the complex driving condition are improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of vehicle steer-by-wire, and in particular relates to a steer-by-wire torque compensation system and method based on intelligent prediction. Background Art

[0002] The steer-by-wire system is an important innovation in modern vehicle steering technology. It eliminates the mechanical connection between the steering wheel and the steering actuator and achieves precise steering control through electronic signals. Wire steer not only expands the design space of steering force and angle transmission characteristics, but also makes it easier to integrate with the autonomous driving system, significantly improving the active safety performance and handling of the car.

[0003] However, the lack of mechanical connection means that the driver cannot directly feel the road feedback, and the road feedback torque needs to be estimated, corrected, simulated, and transmitted to the steering wheel. Among them, the accuracy and real-time performance of road torque compensation are crucial, which can enhance the realism and safety of driving and optimize the overall driving experience.

[0004] Existing road feel torque compensation methods are mainly divided into two categories: model-based and data-driven methods: (I) Model-based methods simulate the mechanical characteristics and road feel feedback of the steering system by establishing a mathematical model, but their adaptability is limited under complex driving conditions. CN 110606121 B The invention patent (a wire-controlled steering road feel simulation control method) combines the dynamic model with the load observer, calculates the steering torque and outputs it comprehensively, and feeds back the road conditions through the electric power steering system, but the accuracy of extracting complex road information still needs to be optimized.

[0005] (II) Data-driven methods use machine learning to learn from data and turn to perception. Although they are highly flexible, they are still insufficient in processing spatiotemporal information and nonlinear factors. CN 115783039 A The invention patent (Torque compensation method and related device for vehicle wire-controlled steering system) uses the vehicle speed and steering wheel angle combined with a preset torque compensation table to control the required torque of the steering system, but it is highly dependent on the preset torque compensation table and lacks an update and recovery mechanism, making it difficult to optimize and adjust in real time. Summary of the invention

[0006] Purpose of the invention: In order to solve the above problems, the present invention provides a wire-controlled steering torque compensation system and method based on intelligent prediction.

[0007] Technical solution: A torque compensation method for wire-controlled steering based on intelligent prediction, comprising the following steps: Step 1: obtaining current state parameters of the vehicle; the current state parameters at least include: steering parameters, driving parameters, and speed parameters; Step 2: Using a deep neural network to perform feature processing and time series information modeling on the current state parameters to generate a current torque signal; Step 3: Based on the dynamic characteristics of the road sensor motor and in combination with predefined input information and interference information, a vehicle enhancement control system is constructed; based on the vehicle enhancement control system and the current torque signal, dynamic prediction information of the vehicle at a future moment is predicted and obtained; Step 4: Based on the linear function and dynamic prediction information stored in offline optimization, determine the optimal control law corresponding to the current state parameters through online retrieval; Step 5: Perform nonlinear friction compensation on the optimal control law and output the compensated optimized torque signal.

[0008] In a further embodiment, the steering parameters in step 1 include at least: steering angle, steering angular velocity, and rack force; The driving parameters at least include: a reverse gear flag; The speed parameters include at least: longitudinal speed, longitudinal acceleration, lateral acceleration, and yaw angular velocity.

[0009] In a further embodiment, generating the current torque signal in step 2 comprises the following steps: The current state parameter is represented by multivariate time series data; wherein the multivariate time series data is defined as a combination of multiple univariate time series composed of the same time index; Extracting features from the multivariate time series data through a convolutional layer; and normalizing the output of the convolutional layer using a batch normalization layer; application LeakyReLU The activation function processes negative information; the gating mechanism effectively controls the flow of information and the update of state; and the fully connected layer is used to output the current torque signal.

[0010] In a further embodiment, the dynamic characteristics of the road sensor motor in step 3 are expressed as follows: ; In the formula, is the electromagnetic torque of the motor; is the motor moment of inertia; is the mechanical angle of the road sensing motor; is the damping coefficient; is the total friction torque of the steering wheel assembly; is the torque measured by the torque sensor; is the reduction ratio; choose As system status; is the input information; is interference information; among them, is the quadrature axis current; is the steering wheel rotation angle.

[0011] In a further embodiment, the vehicle enhancement control system in step 3 is defined by the following formula: ; ; ; In the formula, The system is k The state vector at the moment; is the system output vector; is the system state increment; is the increment of the control input; is the increment of external disturbance; yes k System status at the moment; exist k Control input at all times; exist k External disturbances at any moment; is the system matrix; is the input matrix; is the perturbation matrix; is the output matrix; Then the expected system output is predicted sequentially : ; ; ; ; ; In the formula, is the system output vector in the prediction time domain; is the output mapping matrix; is the input influence matrix; It is the control increment; is the disturbance influence matrix; It is the scope of control; is the disturbance-invariant prediction horizon; is the state transition matrix for the remaining prediction span.

[0012] In a further embodiment, the step 4 further comprises the following steps: Based on the vehicle enhanced control system, the target value within the given prediction range , EMPC The cost function expression is as follows: ; In the formula, is the system output tracking error diagonal weight matrix; is the diagonal weight matrix that controls the input variation; The cost function is used to measure the output of the current prediction With target output The difference between the control input Optimization goal; minimize the cost function and find the optimal control law.

[0013] In a further embodiment, the Brownian friction model is used in step 5 to perform nonlinear friction compensation, wherein the Brownian friction model is expressed as: ; In the formula, represents the Coulomb friction torque; Indicates the magnitude of static friction; stands for Stribeck Speed; Represents the motor angular velocity.

[0014] In a further embodiment, the following steps are also included: The real-time output torque of the road sensing motor is obtained; the real-time output torque is compared with the torque signal after compensation optimization and the dynamic prediction information to obtain a comparison result. If the comparison result exceeds the fault threshold, it is determined that a fault has occurred and recovery signal processing is performed.

[0015] In another technical solution, a wire-controlled steering torque compensation system based on intelligent prediction is provided, which is used to implement the above-mentioned wire-controlled steering torque compensation method based on intelligent prediction, including: A vehicle state acquisition module, used to obtain current state parameters of the vehicle; the current state parameters at least include: steering parameters, driving parameters, and speed parameters; A torque feedback module, used to perform feature processing and time series information modeling on the current state parameters using a deep neural network to generate a current torque signal; A system prediction module, for constructing a vehicle enhancement control system based on the dynamic characteristics of the road sensor motor and in combination with predefined input information and interference information; and predicting the dynamic information of the vehicle at a future moment based on the vehicle enhancement control system; Display model predictive controller, which is used to determine the optimal control law corresponding to the current state parameters through online retrieval based on the linear function stored in offline optimization; The friction compensation module is used to perform nonlinear friction compensation on the optimal control law to obtain an optimized torque signal.

[0016] In a further embodiment, it also includes: a detection recovery module, which is used to obtain the control signal and real-time status of the road sensing motor, compare the system predicted status, automatically detect system failures and input the recovery signal to the torque feedback module to restore the normal operation of the system.

[0017] Beneficial effects: The present invention has good scalability and realizes high-precision steering torque compensation through deep neural networks and data-driven technology. The vehicle status data is collected in real time through various sensors, and the torque feedback module processes the spatiotemporal characteristics of the vehicle status data, and combines the nonlinear friction compensation model and the explicit model predictive controller to effectively optimize the torque feedback signal.

[0018] Compared with the existing methods, the present invention not only reduces the computational burden through offline optimization and online search, but also introduces a prediction model, which significantly improves the accuracy and real-time adaptive capability of steering torque compensation of the vehicle under complex driving conditions. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] Figure 1 It is a structural block diagram of the system in the present invention; Figure 2 is a flow chart of the method of the present invention; Figure 3 It is the structural block diagram of the torque feedback module; Figure 4 It is the structural block diagram of the convolutional layer; Figure 5 This is a block diagram of the gated recurrent unit layer; Figure 6 It is reflected by the Brownian friction model Stribeck Characteristic diagram. DETAILED DESCRIPTION

[0020] Example 1 like Figures 1 to 6 As shown, this embodiment provides a wire control steering torque compensation method based on intelligent prediction (hereinafter referred to as this method), comprising the following steps: Step 1: Obtain current state parameters of the vehicle; the current state parameters at least include: steering parameters, driving parameters, and speed parameters.

[0021] The steering parameters include at least: steering angle, steering angular velocity, and rack force; the driving parameters include at least: reverse gear sign; the speed parameters include at least: longitudinal velocity, longitudinal acceleration, lateral acceleration, and yaw angular velocity.

[0022] Step 2: Use a deep neural network to perform feature processing and time series information modeling on the current state parameters to generate a current torque signal.

[0023] Further, generating a current torque signal comprises the following steps: The current state parameter is represented by multivariate time series data; wherein the multivariate time series data is defined as a combination of multiple univariate time series composed of the same time index; Extracting features from the multivariate time series data through a convolutional layer; and normalizing the output of the convolutional layer using a batch normalization layer; application LeakyReLU The activation function processes negative information; the gating mechanism effectively controls the flow of information and the update of state; and the fully connected layer is used to output the current torque signal.

[0024] The current torque signal is generated using a deep neural network. The generation process is described in detail below: First, the current state parameters are represented by multivariate time series data. Multivariate time series data is considered as a combination of multiple univariate time series with the same time index. Multivariate time series data is considered as an image whose time length is the only spatial dimension. For example, in time k In time series data, the variables are arranged in chronological order: ; In the formula, m Indicates the length of the time series; n =8 indicates the dimension of the feature.

[0025] The above convolutional layer is composed of basic blocks, each of which contains two one-dimensional convolutional layers, where the one-dimensional kernel of size 1×5 has two paddings and is scanned along the time axis. The scanning formula is: ; In the formula, It is Layer and The state of the feature, It is Layer and The state of a feature; and are the trainable kernel weights and bias matrices respectively.

[0026] Since a kernel is only abstracted as a feature map, the above basic blocks have 16, 32, and 32 different kernels to obtain high-level spatial information. The convolutional layer is followed by a batch normalization layer to speed up training by reducing internal covariate shift. The normalization formula is as follows: ; In the formula, and is a learnable parameter; and They are the mean and variance calculated along the feature dimension respectively; is a constant, For numerical stability; After normalization Layer and The state of a feature; Next, after normalization, LeakyReLU Activation function can handle negative information, keep gradient flowing and updating, and accelerate network training.

[0027] The kernel size and no pooling operation of the convolutional layer design mentioned above maintain the length of the multivariate time series data, so that the subsequent gated recurrent unit layer can have the ability to process variable-length data.

[0028] Then, the gated recurrent unit layer effectively controls the flow of information and the update of state through the gating mechanism, ensuring that the model can maintain efficient and stable performance when processing long time series data. Skip connections are added before the gated recurrent unit layer to provide shallow features and speed up the training process.

[0029] The gated recurrent unit layer has a simple structure, consisting only of a reset gate. r ( k ) and update gate z ( k ): ; ; ; ; In the formula, is a candidate state; is a hidden state; and is the weight matrix; is the state eigenvector; is the bias matrix; express sigmoid function; Represents element-wise product.

[0030] Finally, through the fully connected layer, we can get , whose final value is the current feedback torque (i.e., the current torque signal). During the training process, the root mean square error ( RMSE ) as the loss function to make the feedback torque as close to the real torque as possible .

[0031] .

[0032] Step three: construct a vehicle enhancement control system based on the dynamic characteristics of the road sensing motor and in combination with predefined input information and interference information; predict and obtain dynamic prediction information of the vehicle at a future moment based on the vehicle enhancement control system and the current torque signal.

[0033] The dynamic characteristics of the above-mentioned road sensor motor are expressed as follows: ; In the formula, is the electromagnetic torque of the motor; is the motor moment of inertia; is the mechanical angle of the road sensing motor; is the damping coefficient; is the total friction torque of the steering wheel assembly; is the torque measured by the torque sensor; is the reduction ratio; choose As system status; is the input information; It is interference information. is used as the control input. Among them, is the quadrature axis current; is the steering wheel rotation angle.

[0034] Since it can be directly measured And through the differentiator to obtain , thus achieving a comprehensive measurement of the vehicle status.

[0035] The above vehicle enhancement control system is defined by the following formula: ; ; ; In the formula, The system is k The state vector at the moment; is the system output vector; is the system state increment; is the increment of the control input; is the increment of external disturbance; yes k System status at the moment; exist k Control input at all times; exist k External disturbances at any moment; is the system matrix; is the input matrix; is the perturbation matrix; is the output matrix.

[0036] Then the expected system output is predicted sequentially : ; ; ; ; ; In the formula, is the system output vector in the prediction time domain; is the output mapping matrix; is the input influence matrix; It is the control increment; is the disturbance influence matrix; It is the scope of control; is the disturbance-invariant prediction horizon; is the state transition matrix for the remaining prediction span.

[0037] Step 4: Based on the linear function and dynamic prediction information stored in offline optimization, determine the optimal control law corresponding to the current state parameters through online retrieval.

[0038] Further, based on the vehicle enhanced control system, a target value within a given prediction range is , EMPC The cost function expression is as follows: ; In the formula, ,and ; is the system output tracking error diagonal weight matrix; is the diagonal weight matrix that controls the input variation; and It is an adjustable positive performance parameter; is the Kronecker product; The cost function is used to measure the output of the current prediction With target output The difference between the control input Optimization goal; minimize the cost function and find the optimal control law.

[0039] The above cost function optimization problem is regarded as a multi-parameter quadratic programming problem, and the model predictive control is applied Karush - KuhnnTuckerThe method solves the control law offline and divides the state space into m convex subspaces, the explicit solution is expressed as an affine function, which is a linear function in each region: ; In the formula, , , and is a constant matrix, stored in hardware memory in advance; The cost function optimization problem can also be expressed as DeltaU ( k ) is a quadratic programming problem with variables, and the optimal solution is obtained through online solution of the system prediction module. DeltaU ( k ), where the first element DeltaU ( k ) is the actual control increment, thus obtaining the optimal control solution: ; ; ; ; In the formula, is the control input; is the control input; is the actual control increment; is the error vector; is the output mapping matrix; The system is k The state vector at the moment; is the disturbance influence matrix; is the expected output vector; It is the control increment; is the upper limit of the input; yes" subjectto " is an abbreviation for "restricted by; is the lower limit of the input; is the upper limit of output, is the lower limit of the output; the symmetric matrix ;vector .

[0040] Step 5: Perform nonlinear friction compensation on the optimal control law and output the compensated optimized torque signal.

[0041] Furthermore, the Brownian friction model is used to compensate for nonlinear friction. Compared with the dynamic friction model, the Brownian friction model has higher calculation efficiency and is more suitable for practical use. The expression of the Brownian friction model is: ; In the formula, represents the Coulomb friction torque; Indicates the magnitude of static friction; stands for Stribeck Speed; Represents the motor angular velocity.

[0042] In addition, the method further comprises the following steps: Fault identification and recovery signal flow: The real-time output torque of the road sensing motor is obtained; the real-time output torque is compared with the torque signal after compensation optimization and the dynamic prediction information to obtain a comparison result. If the comparison result exceeds the fault threshold, it is determined that a fault has occurred and recovery signal processing is performed.

[0043] The details are as follows: (1) Sensor failure: When a sensor on the vehicle fails, the current status parameters are incomplete.

[0044] The current state signal is obtained and judged. If the current state signal lacks a corresponding signal, the current state signal is judged to be an abnormal signal. The detection recovery module will issue a corresponding instruction, which is to switch the backup sensor to ensure that the torque feedback module receives an accurate and complete input signal. The above-mentioned lack of corresponding signals includes the lack of longitudinal speed signals.

[0045] (2) Control unit failure: If the display model predictive controller or other control unit fails, it will be unable to output the correct control law.

[0046] The real-time output torque of the road sensing motor is obtained; the real-time output torque is compared with the torque signal after compensation optimization and the dynamic prediction information to obtain a comparison result. If the comparison result exceeds the fault threshold, it is determined that a fault has occurred and recovery signal processing is performed.

[0047] That is to say, if the real-time output torque of the road sense motor collected by the rotating dynamic torque sensor differs from the current torque signal and dynamic prediction information by more than the fault threshold, the detection and recovery module detects a system fault and restores the signal input to the torque feedback module to restore the normal operation of the system. If the real-time output torque of the road sense motor still exceeds the fault threshold, emergency measures are taken immediately to enable the road sense control strategy of the wire-controlled steering system to maintain the basic controllability of the vehicle, and the fault location interface is displayed on the on-board screen to warn the driver.

[0048] Example 2 This embodiment provides a wire-controlled steering torque compensation system based on intelligent prediction (hereinafter referred to as this system), which is used to implement a wire-controlled steering torque compensation method based on intelligent prediction as described in Example 1, including: A vehicle state acquisition module, used to obtain current state parameters of the vehicle; the current state parameters at least include: steering parameters, driving parameters, and speed parameters; A torque feedback module, used to perform feature processing and time series information modeling on the current state parameters using a deep neural network to generate a current torque signal; A system prediction module, for constructing a vehicle enhancement control system based on the dynamic characteristics of the road sensor motor and in combination with predefined input information and interference information; and predicting the dynamic information of the vehicle at a future moment based on the vehicle enhancement control system; Display model predictive controller, which is used to determine the optimal control law corresponding to the current state parameters through online retrieval based on the linear function stored in offline optimization; The friction compensation module is used to perform nonlinear friction compensation on the optimal control law to obtain an optimized torque signal.

[0049] In addition, the system also includes: a detection and recovery module. The detection and recovery module is used to obtain the control signal and real-time status of the road sensing motor, compare the system prediction status, automatically detect system failures and input the recovery signal to the torque feedback module to restore the normal operation of the system.

[0050] The vehicle state acquisition module obtains the vehicle state parameters through sensors and inputs the state parameters into the torque feedback module. The sensors include speed sensors, steering wheel angle sensors, yaw rate sensors, acceleration sensors, wheel speed sensors, torque sensors, etc.

[0051] The torque feedback module takes the vehicle state parameters as input, extracts features using the convolutional layer, and models the timing information through the gated recurrent unit to generate the current torque signal. The current torque signal is also called the reference feedback torque. The system prediction module uses the vehicle state parameters (including the torque signal measured by the sensor) for prediction, but the current torque signal itself is not directly used as the output of the system prediction module, but as one of the inputs to participate in the prediction process.

[0052] The system prediction module predicts the future system state based on the dynamic characteristics of the road sense motor, combined with predefined inputs and disturbances. The predicted state of the vehicle at the future moment output by the system prediction module is the basis and input for the display model predictive controller to calculate the optimal control law, which can improve the feedback torque tracking accuracy and ensure real-time performance.

[0053] The explicit model predictive controller quickly finds the optimal control law for the current state based on the linear function stored in the offline optimization. The offline optimization and online search of the explicit model predictive control eliminates the need to solve the optimization problem at each time step, significantly reducing the computational burden compared to traditional model predictive control.

[0054] The friction compensation module uses the Brownian friction model to perform nonlinear friction compensation on the optimal control law to obtain an optimized feedback torque signal. The friction compensation module accurately calculates the additional torque demand caused by friction and corrects the optimal control law of the explicit model predictive controller, effectively reducing the impact of friction on system performance and ensuring the accurate implementation of the control law to drive the vehicle.

[0055] The detection and recovery module is used to monitor the real-time status of the road sense motor, compare it with the system predicted status, automatically detect faults and input the recovery signal to the torque feedback module to restore the normal operation of the system. The detection and recovery module contains a rotating dynamic torque sensor to collect the real-time output torque of the road sense motor.

Claims

1. A torque compensation method for steer-by-wire based on intelligent prediction, characterized in that: The following steps are involved: Step 1: Obtain the current state parameters of the vehicle; The current state parameters include at least: steering parameters, driving parameters, and speed parameters; Step 2: Using a deep neural network to perform feature processing and time series information modeling on the current state parameters to generate a current torque signal; Step 3: Based on the dynamic characteristics of the road sensor motor and in combination with predefined input information and interference information, a vehicle enhancement control system is constructed; based on the vehicle enhancement control system and the current torque signal, dynamic prediction information of the vehicle at a future moment is predicted and obtained; Step 4: Based on the linear function and dynamic prediction information stored in offline optimization, determine the optimal control law corresponding to the current state parameters through online retrieval; Step 5: Perform nonlinear friction compensation on the optimal control law and output the compensated optimized torque signal.

2. The method for torque compensation of steer-by-wire based on intelligent prediction according to claim 1, characterized in that: The steering parameters in step 1 at least include: steering angle, steering angular velocity, and rack force; The driving parameters at least include: a reverse gear flag; The speed parameters include at least: longitudinal speed, longitudinal acceleration, lateral acceleration, and yaw angular velocity.

3. The method for torque compensation of steer-by-wire based on intelligent prediction according to claim 1, characterized in that: The step 2 of generating the current torque signal includes the following steps: The current state parameter is represented by multivariate time series data; wherein the multivariate time series data is defined as a combination of multiple univariate time series composed of the same time index; Extracting features from the multivariate time series data through a convolutional layer; and normalizing the output of the convolutional layer using a batch normalization layer; application LeakyReLU The activation function processes negative information; the gating mechanism effectively controls the flow of information and the update of state; and the fully connected layer is used to output the current torque signal.

4. The method for torque compensation of steer-by-wire based on intelligent prediction according to claim 1, characterized in that: The dynamic characteristics of the road induction motor in step 3 are expressed as follows: ； In the formula, is the electromagnetic torque of the motor; is the motor moment of inertia; is the mechanical angle of the road sensing motor; is the damping coefficient; is the total friction torque of the steering wheel assembly; is the torque measured by the torque sensor; is the reduction ratio; choose As system status; is the input information; is interference information; among them, is the quadrature axis current; is the steering wheel rotation angle.

5. The method for torque compensation of steer-by-wire based on intelligent prediction according to claim 1, characterized in that: The vehicle enhancement control system in step 3 is defined by the following formula: ； ； ； In the formula, The system is k The state vector at the moment; is the system output vector; is the system state increment; is the increment of the control input; is the increment of external disturbance; yes k System status at the moment; exist k Control input at all times; exist k External disturbances at any moment; is the system matrix; is the input matrix; is the perturbation matrix; is the output matrix; Then the sequential prediction expected system output : ； ； ； ； ； In the formula, is the system output vector in the prediction time domain; is the output mapping matrix; is the input influence matrix; It is the control increment; is the disturbance influence matrix; It is the scope of control; is the disturbance-invariant prediction horizon; is the state transition matrix for the remaining prediction span.

6. The method for torque compensation of steer-by-wire based on intelligent prediction according to claim 1, characterized in that: The step 4 further comprises the following steps: Based on the vehicle enhanced control system, the target value within the given prediction range , EMPC The cost function expression is as follows: ； In the formula, is the system output tracking error diagonal weight matrix; is the diagonal weight matrix that controls the input variation; The cost function is used to measure the output of the current prediction With target output The difference between Optimization goal; minimize the cost function and find the optimal control law.

7. The method for torque compensation of steer-by-wire based on intelligent prediction according to claim 1, characterized in that: In step 5, the Brownian friction model is used to perform nonlinear friction compensation, wherein the Brownian friction model expression is: ； In the formula, represents the Coulomb friction torque; Indicates the magnitude of static friction; stands for Stribeck Speed; Represents the motor angular velocity.

8. The method for torque compensation of steer-by-wire based on intelligent prediction according to claim 1, characterized in that: The following steps are also included: The real-time output torque of the road sensing motor is obtained; the real-time output torque is compared with the torque signal after compensation optimization and the dynamic prediction information to obtain a comparison result. If the comparison result exceeds the fault threshold, it is determined that a fault has occurred and recovery signal processing is performed.

9. A wire-controlled steering torque compensation system based on intelligent prediction, used to implement a wire-controlled steering torque compensation method based on intelligent prediction as claimed in any one of claims 1 to 8, characterized in that: include: The vehicle status acquisition module is used to obtain the current status parameters of the vehicle; The current state parameters include at least: steering parameters, driving parameters, and speed parameters; A torque feedback module, used to perform feature processing and time series information modeling on the current state parameters using a deep neural network to generate a current torque signal; A system prediction module, for constructing a vehicle enhancement control system based on the dynamic characteristics of the road sensor motor and in combination with predefined input information and interference information; and predicting the dynamic information of the vehicle at a future moment based on the vehicle enhancement control system; Display model predictive controller, which is used to determine the optimal control law corresponding to the current state parameters through online retrieval based on the linear function stored in offline optimization; The friction compensation module is used to perform nonlinear friction compensation on the optimal control law to obtain an optimized torque signal.

10. The intelligent prediction-based steer-by-wire torque compensation system according to claim 9, characterized in that: Also includes: The detection and recovery module is used to obtain the control signal and real-time status of the road sensing motor, compare it with the system prediction status, automatically detect system failures and input the recovery signal to the torque feedback module to restore the normal operation of the system.

Images