A high-speed intelligent automobile transverse active anti-interference control method
By combining an active disturbance rejection controller and an RBFNN feedforward compensator, the problem of unknown interference in high-speed intelligent vehicles under nonlinear strong coupling conditions is solved, achieving precise path tracking and stability control, and improving the safety of high-speed driving.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2023-07-11
- Publication Date
- 2026-03-17
AI Technical Summary
In high-speed driving environments, the tires of intelligent vehicles are subjected to nonlinear and strongly coupled forces, facing unknown disturbances caused by changes in internal parameters and external environmental inputs. Traditional control methods cannot guarantee safe and reliable tracking of the reference path, posing a risk of traffic accidents.
A composite transverse active anti-disturbance control system is constructed by employing an active disturbance rejection controller and a feedforward compensator based on an adaptive radial basis neural network, combined with an extended state observer and a proportional-differentiator. Through adaptive estimation and compensation of unknown disturbances, accurate path tracking and stability control are achieved.
It enhances the system's ability to suppress unknown disturbances, ensures the accuracy and robustness of path tracking, avoids the risk of vehicles deviating from the center of the lane or hitting the road boundary, and improves the safety of high-speed driving.
Smart Images
Figure CN116719323B_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of intelligent vehicle steering control, and in particular to a method for active anti-interference control of lateral movement in high-speed intelligent vehicles. Background Technology
[0002] Lateral path tracking control is one of the key technologies for intelligent vehicles. Its core objective is to send steering wheel angle signals to enable the vehicle to autonomously track a reference path. However, in high-speed driving environments (referring to speeds exceeding 72 km / h), the tire forces of intelligent vehicles may be in a nonlinear, strongly coupled state. Furthermore, they face unknown disturbances caused by changes in internal parameters and external environmental inputs, making it difficult to establish an accurate lateral dynamics model. If the controller is not robust enough, poor tracking performance can cause the intelligent vehicle to deviate from the lane center while maintaining lane position at high speeds, and may even collide with road boundaries. Therefore, researching lateral control methods with high tracking accuracy and active disturbance suppression for high-speed driving has become a crucial aspect of the intelligent vehicle field.
[0003] Currently, high-speed lateral control methods include proportional-integral-derivative (PID) methods, pure tracking methods, linear quadratic methods, and model prediction methods. However, traditional methods often fail to adequately consider factors such as strong nonlinear coupling effects, changes in internal dynamic parameters, external environmental inputs, and inaccurate modeling. This can cause complex disturbances to the tracking system. When these disturbances and their rates of change change rapidly, traditional controllers struggle to ensure safe and reliable tracking of the reference path, potentially leading to traffic accidents and threatening life and property. Summary of the Invention
[0004] To address the aforementioned problems in the existing technology, the present invention aims to provide a high-speed intelligent vehicle lateral active anti-interference control method, which can effectively improve the ability of the lateral controller to suppress internal and external disturbances of the system, and can achieve robust and accurate tracking while ensuring stability.
[0005] To achieve the above objectives, the technical solution of the present invention is as follows:
[0006] A method for lateral active anti-interference control of a high-speed intelligent vehicle is disclosed, which utilizes a lateral active anti-interference control system for control. The lateral active anti-interference control system includes an initialization module, an active anti-interference controller, and a feedforward compensator based on an adaptive radial basis neural network.
[0007] The initialization module is responsible for checking whether the signal transmission and reception of the sensing module, positioning module, planning module, and chassis module are normal; loading the parameters of the monorail vehicle dynamics model, loading the ADRC parameters, loading the parameters of the feedforward compensator based on RBFNN, and loading the parameters of the active disturbance rejection path tracking nominal model; the RBFNN is an adaptive radial basis neural network, and the ADRC is an active disturbance rejection controller.
[0008] The active disturbance rejection controller consists of a tracking differentiator, a proportional-differentiator (PD), and an extended state observer (ESO) to achieve closed-loop feedback control of the vehicle's heading deviation state. The PD is a proportional-differentiator, and the ESO is an extended state observer.
[0009] The ESO is used to estimate the total disturbance generated during path tracking. The total disturbance consists of unmodeled uncertainties and external unknown disturbances. The unmodeled uncertainties include disturbances caused by changes in vehicle mass and tire characteristics. The PD state feedback control law output is compensated to obtain the ADRC control law.
[0010] The feedforward compensator based on RBFNN estimates the steady-state disturbances that ESO cannot estimate and directly compensates for them. Finally, the front wheel steering angle control law is obtained by superimposing the ADRC control law and the estimated value of the feedforward compensator based on RBFNN. This realizes the composite lateral active anti-interference control method by combining the ADRC control law and the feedforward compensator based on RBFNN, and completes closed-loop feedback and feedforward regulation.
[0011] The control method includes the following steps:
[0012] Step 1: Initialization
[0013] The initialization module is responsible for checking whether the signal transmission and reception of the sensing module, positioning module, planning module and chassis module are normal, loading the parameters of the monorail vehicle dynamics model, loading the parameters of the active disturbance rejection path tracking nominal model, loading the parameters of the active disturbance rejection controller, and loading the parameters of the feedforward compensator based on RBFNN.
[0014] The dynamic model of the monorail vehicle is established in the form of mathematical equations as follows:
[0015]
[0016]
[0017] In the formula: For vehicle quality; Lateral speed; This refers to the longitudinal speed of the vehicle. Angular velocity of heading; This refers to the angular acceleration of the heading. It is the moment of inertia; and These are the distances from the front and rear axles to the center of gravity, respectively. and These are the lateral stiffness of the front and rear wheels, respectively; This refers to the steering angle of the front wheels.
[0018] The nominal model for active disturbance rejection path tracking is:
[0019]
[0020] In the formula: This is due to an unknown external disturbance. For the total disturbance, The controlled input gain coefficient.
[0021] Step 2: Obtain the reference heading and its rate of change
[0022] The differential tracker receives reference path information from the planning module, which includes reference heading information. The reference heading information is then input into the differential tracker to obtain the reference heading change rate information. Next, the closest matching point to the intelligent vehicle on the reference path is found, and the heading error and heading error change rate between the vehicle's actual heading and its rate of change and the reference heading and its rate of change are calculated.
[0023] The mathematical equation for the differential tracker is as follows:
[0024]
[0025] In the formula: Sampling time; and These are all output state quantities of the differential tracker, namely, the filtered reference heading and its rate of change information, respectively. and All of these are the output state quantities of the differential tracker at the next moment; This is the input information for the differential tracker, namely the desired heading angle; The speed factor determines the tracking speed; It is a filtering factor and has a filtering function. , , , and These are all intermediate variables generated during the calculation process of the differential tracker. Function Defined as:
[0026]
[0027] The formulas for calculating the heading error and the rate of change of heading error are as follows:
[0028]
[0029] In the formula: This refers to the vehicle's real-time heading angle in the geodetic coordinate system. For heading error; This represents the rate of change of heading error.
[0030] Step 3: State estimation based on ESO
[0031] The real-time heading information of the intelligent vehicle and the front wheel steering angle information of the intelligent vehicle at the previous moment are input into the ESO. The ESO outputs the estimated heading, the estimated rate of change of heading, and the estimated total disturbance of the intelligent vehicle. The state-space equation of the ESO is:
[0032]
[0033] In the formula: , ,
[0034] Here is the state variable matrix of the ESO. , and They are respectively , and The estimated value, This is the coefficient matrix of the ESO state variable matrix. Input the coefficient matrix of the control quantity for ESO. Let be the gain coefficient matrix of the ESO. Choose the ESO gain such that the characteristic polynomial... It is a Herwitz polynomial. For ease of tuning, the ESO poles are configured at... ,and The aforementioned Define the bandwidth of ESO. .
[0035] Step 4: Determine the PD state feedback control law
[0036] The real-time heading estimate, heading rate of change estimate, and total disturbance estimate of the intelligent vehicle obtained by the ESO in step 3, along with the output information of the tracking differentiator in step 2, are input into the PD. A PD state feedback control law is then established through the PD. The PD state feedback control law is as follows:
[0037]
[0038] In the formula: This is a PD state feedback control law. and All are PD gain coefficients and satisfy , Choosing the PD gain coefficient to make the characteristic polynomial It is a Herwitz polynomial; the bandwidth of PD is taken as... , , .
[0039] Step 5: Determine the ADRC control law
[0040] The estimated value of the total disturbance from the ESO in step 3 is compensated for in the output of the PD state feedback control law in step 4, thereby determining the ADRC control law as follows:
[0041]
[0042] Step 6: Calculate the front wheel steering angle control law
[0043] The heading error and its rate of change calculated in step 2 are input into the RBFNN-based feedforward compensator. The RBFNN-based feedforward compensator updates the weights in real time through the RBFNN weight adaptive law, thereby estimating the steady-state disturbances that the ESO cannot estimate in real time and directly compensating for them. Finally, the front wheel steering angle control law is obtained by superimposing the ADRC control law and the estimated value of the RBFNN-based feedforward compensator. This realizes the composite lateral active anti-interference control method by combining the ADRC control law and the RBFNN-based feedforward compensator, thus completing the closed-loop feedback and feedforward adjustment of the high-speed intelligent vehicle lateral active anti-interference control system.
[0044] The RBFNN-based feedforward compensator is:
[0045]
[0046] In the formula: ; For RBFNN-based weights, the activation function of the hidden layer Represented as:
[0047]
[0048] In the formula: The input vector; For the first The center value vector of each node; For the first The basis width vector of the Gaussian function for each node. The ideal output of RBFNN. and actual output They are respectively:
[0049]
[0050] In the formula: and The weights are, in order, the ideal and the actual RBFNN weights. Assume... If the value is a very small positive real number, then it satisfies the following condition: ,in The steady-state disturbance that the RBFNN-based feedforward compensator needs to eliminate is defined; the RBFNN approximation error is defined. ,because It is bounded; let's assume the upper bound of the error is... Actual output value of RBFNN network and satisfy:
[0051]
[0052] In the formula: .
[0053] The adaptive law for the RBFNN weights is as follows:
[0054]
[0055] In the formula: , , and These are all set parameters; Here is the input matrix of the RBFNN. For matrix The norm; The coefficient matrix is set and satisfies and ;matrix It is symmetric positive definite and satisfies ,in , ,and For a parameter greater than 0, For the given matrix and .
[0056] The front wheel steering angle control law, which is the sum of the ADRC control law and the estimated value of the feedforward compensator based on RBFNN, is as follows:
[0057]
[0058] In the formula: This is the estimated value for the feedforward compensator based on RBFNN.
[0059] Step 7: Perform lateral active anti-interference control
[0060] The final front wheel steering angle determined in step 6 is subject to control constraints and filtering. This angle is then sent to the intelligent vehicle as a control signal at a fixed frequency. The intelligent vehicle performs active lateral anti-interference control based on the final front wheel steering angle. Further, it is determined whether the intelligent vehicle has reached its destination. If so, the vehicle completes its control task; otherwise, proceed to step 1.
[0061] The present invention has the following beneficial effects:
[0062] 1. This invention establishes a nominal model for self-disturbance rejection path tracking based on a monorail vehicle dynamics model. The designed ESO can autonomously estimate unmodeled uncertainties and unknown external disturbances in the system, improving the anti-interference capability of the path tracking control algorithm and ensuring lateral and heading tracking accuracy.
[0063] 2. The feedforward compensator based on RBFNN in this invention achieves real-time weight updates through the RBFNN weight adaptive law, which can accurately estimate the part of the total disturbance that ESO cannot estimate. Specifically, it refers to the steady-state disturbance in the real control model of active disturbance rejection path tracking. When facing different disturbances (including unmodeled uncertain disturbances and external unknown disturbances, where unmodeled uncertain disturbances include disturbances caused by changes in vehicle mass and tire characteristics, etc.), both ESO and the feedforward compensator based on RBFNN can jointly estimate and compensate for the disturbances in real time, ultimately improving the robustness of the path tracking control algorithm and the stability of the vehicle. Attached Figure Description
[0064] Figure 1 This is a schematic diagram of the lateral active anti-interference control method of the present invention.
[0065] Figure 2 This is a schematic diagram of the transverse active anti-interference control system of the present invention.
[0066] Figure 3 This is a schematic diagram of the nominal model of the self-disturbance rejection path tracking of the present invention.
[0067] Figure 4 This is a structural diagram of the radial basis function neural network model of the present invention.
[0068] Figure 5 This is a flowchart of an embodiment of the present invention.
[0069] Figure 6 This is a schematic diagram of an existing intelligent vehicle deviating from the center of the lane while maintaining its position in a highway lane.
[0070] Figure 7 This is a diagram illustrating an existing intelligent vehicle colliding with a road boundary while maintaining its position in a highway lane. Detailed Implementation
[0071] The following is for reference only. Figure 1-7 The present invention provides a further explanation of a high-speed intelligent vehicle lateral active anti-interference control method.
[0072] like Figure 1 The diagram shown is a schematic of a high-speed intelligent vehicle lateral active anti-interference control method according to the present invention. The method mainly consists of two parts:
[0073] (1) ADRC controller composed of a tracking differentiator, PD, and ESO: The filtered reference heading and its rate of change information are obtained through the differentiator tracker. The PD is designed to calculate the tracking error between the reference heading and the actual heading and establish the PD state feedback control law. At the same time, the real-time heading of the vehicle and the front wheel steering angle information of the previous moment are input to the ESO, so that the ESO estimates the estimated values of the vehicle's real heading, heading rate, and total disturbance in real time. Then, the disturbance estimated by the ESO is used to compensate the output of the PD state feedback control law to obtain the ADRC control law. If the ESO's estimation of the total disturbance is inaccurate, it will lead to insufficient robustness of the controller and poor tracking performance, causing the intelligent vehicle to deviate from the lane center while maintaining driving in a high-speed lane, such as... Figure 6 As shown; it may even hit the road boundary, such as Figure 7 As shown. Therefore, it is necessary to introduce a feedforward compensator based on RBFNN for feedforward compensation to eliminate steady-state disturbances during the tracking process.
[0074] (2) Feedforward compensator based on RBFNN: The heading error and heading error change rate between the vehicle's actual heading and its rate of change and the reference heading and its rate of change are calculated and input into RBFNN. Steady-state disturbances that ESO cannot estimate are estimated and directly compensated. Finally, the front wheel steering angle control law is obtained by superimposing the ADRC control law and the estimated value of the feedforward compensator based on RBFNN. The ADRC control law and the feedforward compensator based on RBFNN are combined to form a composite lateral active anti-interference control method, and the closed-loop feedback and feedforward adjustment of the high-speed intelligent vehicle lateral active anti-interference control system are completed.
[0075] like Figure 2 As shown, this invention provides a high-speed intelligent vehicle lateral active anti-interference control system, which includes an initialization module, ADRC, and a feedforward compensator based on RBFNN, among other control modules.
[0076] according to Figure 1 and Figure 2 This invention provides a method for lateral active anti-interference control of high-speed intelligent vehicles, utilizing a lateral active anti-interference control system for control. The specific design steps of the lateral active anti-interference control system for high-speed intelligent vehicles are as follows:
[0077] Step 1: Design the Differential Tracker
[0078] The differential tracker receives reference heading information from the planning module and inputs it into the differential tracker to obtain the reference heading change rate information. Then, it calculates the heading error and the heading error change rate between the vehicle's actual heading and its rate of change and the reference heading and its rate of change.
[0079] Step 2: Design the nominal model for active disturbance rejection path tracking
[0080] Designed based on the dynamics model of a monorail vehicle, as follows: Figure 3 The nominal model for active disturbance rejection path tracking shown primarily comprises a reference road and an intelligent vehicle dynamics model. It establishes the path tracking control relationship between the reference road and the intelligent vehicle's single-track dynamics model. This model also includes vehicle attitude information, Frenet coordinate system Geodetic coordinate system and vehicle coordinate system The dynamic model of the monorail vehicle is established in the form of mathematical equations as follows:
[0081]
[0082]
[0083] In the formula: For vehicle quality; Lateral speed; This refers to the longitudinal speed of the vehicle. Angular velocity of heading; This refers to the angular acceleration of the heading. It is the moment of inertia; and These are the distances from the front and rear axles to the center of gravity, respectively. and These are the lateral stiffness of the front and rear wheels, respectively; This refers to the steering angle of the front wheels.
[0084] The nominal model for active disturbance rejection path tracking is:
[0085]
[0086] In the formula: This is due to an unknown external disturbance. For the total disturbance, The controlled input gain coefficient.
[0087] Step 3: Design ESO
[0088] An extended state observer (ESO) is designed based on the nominal model of active disturbance rejection path tracking (ADRP) to estimate the total disturbance in the nominal model of ADRP in step 2. The ESO is designed in state-space equation form as follows:
[0089]
[0090] In the formula: , ,
[0091] Here is the state variable matrix of the ESO. , and They are respectively , and The estimated value, This is the coefficient matrix of the ESO state variable matrix. Input the coefficient matrix of the control quantity for ESO. Let be the gain coefficient matrix of the ESO. Choose the ESO gain such that the characteristic polynomial... It is a Herwitz polynomial. For ease of tuning, the ESO poles are configured at... ,and The aforementioned Define the bandwidth of ESO. .
[0092] Step 4: Design PD
[0093] The real-time heading estimate, heading rate of change estimate, and total disturbance estimate of the intelligent vehicle obtained by the ESO in step 3, along with the output information of the tracking differentiator in step 2, are input into the PD. The PD state feedback control law is then established through the PD as follows:
[0094]
[0095] In the formula: This is a PD state feedback control law. and All are PD gain coefficients and satisfy , Choosing the PD gain coefficient to make the characteristic polynomial It is a Herwitz polynomial; the bandwidth of PD is taken as... , , .
[0096] Step 5: Design ADRC control law
[0097] The estimated value of the total disturbance from the ESO in step 3 is compensated for in the output of the PD state feedback control law in step 4, and then the ADRC control law is designed as follows:
[0098]
[0099] Furthermore, in step 3, it is assumed that... ,in The portion accurately estimated by ESO. This represents the portion of the total disturbance that is difficult for ESO to estimate. After being eliminated, the nominal model for active disturbance rejection path tracking in step 2 is updated to... Its control law is designed as
[0100]
[0101] In the formula: , ,matrix for and The coefficient matrix; matrix State quantity matrix of the nominal model for active disturbance rejection path tracking The coefficient matrix, The state matrix is the expected state quantity matrix of the active disturbance rejection path tracking nominal model. This is the error state quantity matrix of the nominal model for active disturbance rejection path tracking. for The coefficient matrix, ,and It is a parameter greater than 0.
[0102] Furthermore, in step 2, it is assumed that the real control model for active disturbance rejection path tracking is as follows: The active disturbance rejection path tracking (ADRP) real control model is the real model corresponding to the ADRP nominal model, and the error system based on the ADRP real control model is determined as follows:
[0103]
[0104] Where: matrix In the real control model for active disturbance rejection path tracking and The coefficient matrix, and Both are the difference matrices between the coefficient matrices in the nominal model of active disturbance rejection path tracking and the actual control model of active disturbance rejection path tracking. This is the differential matrix of the error state matrix in the real control model for active disturbance rejection path tracking. This is to track steady-state disturbances in the real control model using the active disturbance rejection path.
[0105] Step 6: Design a feedforward compensator based on RBFNN
[0106] Design as Figure 4 The feedforward compensator based on RBFNN shown is, where The weights of the RBFNN are given by the input layer, which contains the tracking error and its rate of change between the vehicle's actual heading state and the reference system's heading, respectively; the hidden layer consists of 7 neurons. The Gaussian function is used; the output layer contains steady-state perturbations that ESO cannot estimate.
[0107] The feedforward compensator based on RBFNN is designed as follows:
[0108]
[0109] Where: the activation function of the hidden layer Represented as:
[0110]
[0111] In the formula: The input vector; For the first The center value vector of each node; For the first The basis width vector of the Gaussian function for each node. The ideal output of RBFNN. and actual output They are respectively:
[0112]
[0113] In the formula: and The weights are, in order, the ideal and the actual RBFNN weights. Assume... If the value is a very small positive real number, then it satisfies the following condition: ,in The steady-state disturbance that the RBFNN-based feedforward compensator needs to eliminate is defined; the RBFNN approximation error is defined. ,because It is bounded; let's assume the upper bound of the error is... Actual output value of RBFNN network and satisfy:
[0114]
[0115] In the formula: .
[0116] The adaptive law for the RBFNN weights is as follows:
[0117]
[0118] In the formula: , , and These are all set parameters; Here is the input matrix of the RBFNN. For matrix The norm; The coefficient matrix is set and satisfies and ;matrix It is symmetric positive definite and satisfies ,in , For the given matrix and .
[0119] Step 7: Based on Steps 5 and 6, design the front wheel steering angle control law as follows:
[0120]
[0121] Step 8: Apply control constraints and perform filtering on the final front wheel steering angle determined in Step 7, and send it as a control signal to the intelligent vehicle at a fixed frequency. The intelligent vehicle performs lateral active anti-interference control based on the final front wheel steering angle. Further, determine whether the intelligent vehicle has reached its destination. If so, the vehicle completes the control task; otherwise, repeat steps 1 to 7 to ensure the intelligent vehicle safely and stably tracks the desired path and completes the control task.
[0122] This invention is not limited to the details of the above embodiments. Any equivalent concept or modification within the technical scope disclosed in this invention shall be included within the protection scope of this invention.
Claims
1. A method for active anti-interference control of lateral movement in high-speed intelligent vehicles, characterized in that: The high-speed intelligent automobile transverse active anti-interference control system is used for control, and the high-speed intelligent automobile transverse active anti-interference control system comprises an initialization module, a self-anti-interference controller and a feedforward compensator based on an adaptive radial basis neural network; The initialization module is responsible for checking whether signals of a perception module, a positioning module, a planning module and a chassis module are normal, loading single-track vehicle dynamics model parameters, loading ADRC parameters, loading feedforward compensator parameters based on RBFNN and loading self-anti-interference path tracking nominal model parameters; the RBFNN is an adaptive radial basis neural network, and the ADRC is a self-anti-interference controller; The self-anti-interference controller is composed of a tracking differentiator, a PD and an ESO, realizes state closed-loop feedback control of a vehicle heading deviation, the PD is a proportional differentiator, and the ESO is an extended state observer; The ESO is used for estimating total disturbance generated in a path tracking process, the total disturbance is unmodeled uncertainty disturbance and external unknown disturbance, the unmodeled uncertainty disturbance comprises disturbance caused by vehicle mass and tire characteristic change, and the ADRC control law is obtained by compensating the output of the PD state feedback control law; The feedforward compensator based on RBFNN estimates steady-state disturbance that cannot be estimated by the ESO and directly compensates the steady-state disturbance, finally, the front wheel steering angle control law is obtained by superimposing the ADRC control law and the estimated value of the feedforward compensator based on RBFNN, the ADRC control law and the feedforward compensator based on RBFNN are combined to realize a composite transverse active anti-interference control method, and closed-loop feedback and feedforward regulation are completed; The control method comprises the following steps: Step 1: initialization The initialization module is responsible for checking whether signals of a perception module, a positioning module, a planning module and a chassis module are normal, loading single-track vehicle dynamics model parameters, loading self-anti-interference path tracking nominal model parameters, loading self-anti-interference controller parameters and loading feedforward compensator parameters based on RBFNN; The single-track vehicle dynamics model is established in the form of a mathematical equation as follows: wherein: is the vehicle mass; is the lateral vehicle speed; is the longitudinal vehicle speed; is the yaw rate; is the yaw acceleration; is the moment of inertia; and are the distances from the front and rear axles to the center of mass, respectively; and are the front and rear wheel cornering stiffness, respectively; is the front wheel steering angle; The self-anti-interference path tracking nominal model is as follows: wherein: is an external unknown disturbance, is the total disturbance, is the controlled input gain coefficient; Step 2: obtaining reference heading and its rate of change A differential tracker receives reference path information issued by a planning module, the reference path information comprises reference heading information; the reference heading information is input into the differential tracker to obtain reference heading rate information; then a matching point closest to an intelligent automobile on a reference path is found, and heading error and heading error rate of change between real vehicle heading and its rate of change and reference heading and its rate of change are calculated; The mathematical equation of the differential tracker is as follows: wherein: is the sampling time; and are the output state variables of the derivative tracker, i.e. the filtered reference heading and its rate of change information respectively, and are the output state variables of the derivative tracker at the next time instant; is the input information of the derivative tracker, i.e. the desired heading angle; is the velocity factor, which determines the tracking speed; is the filter factor, which has a filtering effect; , , , and are intermediate variables produced in the calculation process of the derivative tracker; the function is defined as: The calculation formulae of the heading error and the heading error rate of change are as follows: In the formula: is the real-time heading angle of the vehicle in the earth coordinate system; is the heading error; is the heading error rate of change; Step 3: state estimation based on ESO Real-time heading information of the intelligent automobile and front wheel steering angle information of the intelligent automobile at the last moment are input into the ESO, and ESO output information is an estimated value of intelligent automobile heading, an estimated value of heading rate of change and an estimated value of total disturbance; the state space equation of the ESO is as follows: In the formulae: , , is the state matrix of the ESO, , and are the estimates of , and respectively, is the coefficient matrix of the ESO state matrix, is the coefficient matrix of the ESO input control, is the gain coefficient matrix of the ESO; the ESO gain is chosen such that the characteristic polynomial is a Hurwitz polynomial; for simple tuning, the ESO poles are configured at , and ; the is the bandwidth of the ESO, defined as ; Step 4: determining a PD state feedback control law The ESO described in step 3 is input into the PD with the real-time heading estimation value of the intelligent vehicle, the estimation value of the heading rate of change, the estimation value of the total disturbance, and the output information of the tracking differentiator described in step 2, and a PD state feedback control law is established by the PD; the PD state feedback control law is: wherein: is a PD state feedback control law, and are PD gain coefficients and satisfy , The PD gain coefficients are selected such that the characteristic polynomial is a Hurwitz polynomial; the PD bandwidth is taken as , , ; Step 5: determining the ADRC control law The estimation value of the total disturbance of the ESO in step 3 is compensated to the PD state feedback control law output in step 4, and the ADRC control law is determined as: Step 6: calculating the front wheel steering angle control law The heading error and its rate of change calculated in step 2 are input into the RBFNN-based feedforward compensator, the RBFNN-based feedforward compensator realizes real-time updating of the RBFNN weight value through the RBFNN weight value adaptive law, and then estimates and compensates the steady-state disturbance that cannot be estimated by the ESO in real time, and finally obtains the front wheel steering angle control law superimposed with the estimation value of the RBFNN-based feedforward compensator, realizes the composite lateral active anti-interference control method composed of the ADRC control law and the RBFNN-based feedforward compensator, and completes the closed-loop feedback and feedforward regulation of the high-speed intelligent vehicle lateral active anti-interference control system; The RBFNN-based feedforward compensator is: In the formula: ; is the weight based on the RBFNN, the activation function of the hidden layer is expressed as: wherein: is the input vector; is the center value vector of the th node; is the basis width value vector of the th node; and the ideal output and the actual output of the RBFNN are respectively: where and are the ideal and actual RBFNN weights, respectively; assuming is a very small positive real number, then where is the steady-state disturbance to be eliminated by the RBFNN-based feedforward compensator; define the RBFNN approximation error Since is bounded, assume the error upper bound is ; the actual output value of the RBFNN network satisfies In the formulae: ; The RBFNN weight value adaptive law is: wherein: , , and are set parameters; is the input matrix of the RBFNN, is the norm of the matrix ; is a set coefficient matrix and satisfies and ; the matrix is symmetric positive definite and satisfies wherein , , and is a parameter greater than 0, is a set matrix and ; The front wheel steering angle control law superimposed with the estimation value of the ADRC control law and the RBFNN-based feedforward compensator is: In the formula: is the estimated value of the feedforward compensator based on the RBFNN; Step 7: performing lateral active anti-interference control The final front wheel steering angle determined in step 6 is subjected to control constraints and filtering processing, and is used as a control signal to be issued to the intelligent vehicle at a fixed frequency, and the intelligent vehicle performs lateral active anti-interference control according to the final front wheel steering angle. Further, it is judged whether the intelligent vehicle has reached the destination, if yes, the vehicle completes the control task, otherwise, it goes back to step 1.