Event-triggered robust nonlinear model predictive control method for agricultural machinery trajectory tracking

Abstract

The present invention belongs to the field of control technology, and relates to an event-triggered robust nonlinear model predictive control method for agricultural machinery trajectory tracking. The method comprises: establishing a transverse and longitudinal dynamics model of agricultural machinery and its state space equations, and tracking the operation trajectory of agricultural machinery; constructing an improved nonlinear model predictive control system, the improved nonlinear model predictive control system comprises a nonlinear controller of the agricultural machinery nominal system with a pseudo-Huberl function improved cost function, and solving the control input of the agricultural machinery nominal system; optimizing the solver, and using the reinforcement learning actor-critic algorithm to approximate the uncertainty compensation of the agricultural machinery model; using an adaptive parameter nonlinear disturbance observer to estimate and compensate for the external disturbances suffered by the agricultural machinery; through an event triggering mechanism, judging whether to trigger the optimization solver to optimize the solution according to the error between the real value and the predicted value of the agricultural machinery trajectory tracking, and obtaining the input of the agricultural machinery trajectory tracking controller. The present invention enhances the robustness of the control system and reduces the calculation frequency.

Description

Technical Field

[0001] The invention belongs to the field of control technology and relates to an event-triggered robust nonlinear model predictive control agricultural machinery trajectory tracking method. Background Art

[0002] In the context of agricultural production modernization, people are increasingly interested in agricultural machinery automatic driving technology. The emergence of automatic agricultural machinery has played an irreplaceable role in increasing crop yields and improving farmers' living standards. Accurate trajectory tracking control is a key technology for realizing agricultural machinery automatic driving, which is of great significance for saving labor costs and improving operation accuracy. The current mainstream trajectory tracking control algorithms include: fuzzy logic control, classical control theory, sliding mode control, enhanced PID control, and model predictive control (MPC). Nonlinear model predictive control (NMPC), as a multivariable controller, can better predict the future state of the vehicle and handle various constraints. It is widely used in agricultural machinery automatic driving. Many agricultural machinery nonlinear model predictive control trajectory tracking works are based on kinematic models. Although the model construction is simple, the accuracy of the model is lost, which has an adverse effect on the control accuracy. Some scholars use more accurate dynamic models for controller design, but often only consider the lateral control of agricultural machinery, and do not introduce longitudinal control into the control system. During the operation of agricultural machinery, the longitudinal tracking accuracy will be lost to compensate for the lateral accuracy. Model predictive controllers based on dynamic models have higher requirements for model accuracy, but often ignore the external disturbances to agricultural machinery and the uncertainty of model parameters. However, due to the relatively complex farmland environment, the land environment is relatively soft, slippery, and uneven, which inevitably brings unknown disturbances to the trajectory tracking control of agricultural machinery operation. In addition, due to the aging of the agricultural machinery itself, the parameters of the agricultural machinery itself will also be perturbed, such as the lateral stiffness of the tire, the moment of inertia, and its own mass will also change, which brings difficulties to the accurate establishment of the dynamic model of agricultural machinery. Although the classical nonlinear model predictive control performs well in improving control accuracy, it has a large computational complexity and amount of calculation. In practical applications, it is necessary to weigh the relationship between computational efficiency and control performance to ensure the real-time and reliability of the system. How to save computing resources is also a problem that needs to be considered. Summary of the invention

[0003] In order to solve many problems existing in the prior art, the present invention proposes an improved nonlinear model predictive control method. The control input is the longitudinal traction of the rear wheel and the front wheel steering angle, and the control variables are the vehicle position and yaw angle. The cost function is improved by using the pseudo-Huberl function to improve the turning control performance. The reinforcement learning actor-critic neural network (ACNN) is introduced to approximate the uncertain terms of the model. The adaptive nonlinear disturbance observer (NDOB) is used to estimate the unknown time-varying disturbance, which enhances the robustness of the controller. The event trigger mechanism is designed to reduce the calculation frequency of the NMPC to save computing resources.

[0004] The technical solution adopted by the present invention to solve the technical problem is: an event-triggered robust nonlinear model predictive control agricultural machinery trajectory tracking method, comprising:

[0005] Establish the transverse and longitudinal dynamics model of agricultural machinery and its state space equations to track the trajectory of agricultural machinery;

[0006] An improved nonlinear model predictive control system is constructed. The improved nonlinear model predictive control system includes a nonlinear controller of the nominal system of agricultural machinery with a pseudo-Huber function to improve the cost function, and solves the control input of the nominal system of agricultural machinery; an optimization solver is used to approximate the uncertainty compensation of the agricultural machinery model using a reinforcement learning actor-critic algorithm; an adaptive parameter nonlinear disturbance observer is used to estimate and compensate the external disturbances of the agricultural machinery; through an event trigger mechanism, the error between the true value and the reference value of the agricultural machinery trajectory tracking is used to determine whether to trigger the optimization solver for optimization solution; and finally the input of the agricultural machinery trajectory tracking controller is obtained:

[0007] ;

[0008] ;

[0009] ;

[0010] in, and They represent the upper and lower limits of the input of the agricultural machinery traction force and the front wheel angle respectively; It is the control input of the nominal system of agricultural machinery; is the model uncertainty; is the adaptive disturbance observer estimate; is the adaptation rate.

[0011] Preferably, the objective function and constraint conditions of the nonlinear control model of the agricultural machinery nominal system of the pseudo-Huber function improved cost function are specifically:

[0012] ;

[0013] ;

[0014] in, express The optimal solution of the time-nominal system is the control input sequence; Represents the set of state constraints of the system; is the terminal domain; For the prediction time domain, To control the time domain and ; is the terminal cost, is the weighting matrix;

[0015] is the stage cost function, which uses quadratic penalties on speed and control input, expressed as:

[0016] ;

[0017] is a weighted matrix that penalizes speed, is the penalty matrix for the control input; They are longitudinal velocity, lateral velocity and yaw rate respectively; is the yaw angle of the agricultural machinery; Indicates the time, represents the step length; and They represent the position coordinates in the geodetic coordinate system respectively;

[0018] Choose location cost As a pseudo-Huber function, Weighted, penalized differences between the actual horizontal and vertical postures of the agricultural machinery and the expected postures, expressed as:

[0019] ;

[0020] in, is a positive design parameter; the cost of agricultural machinery heading pass Weighted, penalized agricultural machinery heading surround, expressed as:

[0021] .

[0022] Preferably, the reinforcement learning actor-critic algorithm approximates the uncertainty compensation of the agricultural machinery model as follows:

[0023] ;

[0024] in, is an estimate of the optimal weight matrix, is the actor neural network basis function vector; is the input vector, is the approximate error.

[0025] Preferably, the adaptive parameter nonlinear disturbance observer is expressed as follows:

[0026] ;

[0027] in, is an estimate of the unknown disturbance, is the intermediate state variable, is the designed parameter matrix, ; and is the nominal coefficient matrix; is the state vector of the control system;

[0028] The adaptive rate is designed as:

[0029] ;

[0030] in, is the design positive parameter, is the initial value of the adaptive parameter, is the design parameter.

[0031] Preferably, the event triggering mechanism is:

[0032] ;

[0033] in, represents the nominal system control input; express The optimal solution of the time-nominal system is the control input sequence; For the prediction time domain, is the error threshold between the actual operation trajectory of the agricultural machinery and the reference trajectory; is the current triggering moment, For the next trigger moment; Indicated in After time The system state at the step time; is the set reference trajectory; is the vector's two-norm, norm , represents transpose, is the weighting matrix.

[0034] Compared with the prior art, the present invention has the following beneficial effects:

[0035] (1) Establish the lateral and longitudinal dynamics model of agricultural machinery, taking into account position disturbance and model uncertainty, and taking into account longitudinal motion control while controlling the lateral direction of the agricultural machinery. Use the pseudo-Huber function to improve the nonlinear model predictive control cost function and improve the tracking accuracy of the agricultural machinery in the turning trajectory;

[0036] (2) A disturbance observer with adaptive parameters is proposed to estimate the unknown time-varying disturbance and compensate it to the input to enhance the robustness of the nonlinear predictive controller. A reinforcement learning actor-critic neural network is used to deal with model uncertainty, and the uncertainty is approximated and compensated in the design process of the disturbance observer and the nonlinear predictive controller.

[0037] (3) Based on the state prediction deviation caused by the remaining unknown time-varying disturbances and uncertainties, an event-triggered nonlinear model predictive control optimization solution triggering strategy is designed. Based on the error between the system state and the reference trajectory, an event trigger mechanism is designed. The optimization problem needs to be solved only when the error reaches the trigger level, which reduces the frequency of optimization solutions and saves computing resources;

[0038] (4) An S-type reference trajectory was designed to simulate the real working scenario of agricultural machinery. The robustness and effectiveness of the proposed algorithm were verified. The optimization problem can be solved in milliseconds, meeting the real-time requirements. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] Figure 1 for vehicle lateral dynamics;

[0040] Figure 2 for the longitudinal dynamics of the vehicle;

[0041] Figure 3 It is a principle flow chart of the event-triggered robust nonlinear model predictive control agricultural machinery trajectory tracking method of the present invention;

[0042] Figure 4 It is the typical operation path (operation trajectory) of agricultural machinery;

[0043] Figure 5 The event triggering mechanism proposed by the present invention is subjected to Markov time-varying disturbance;

[0044] Figure 6 is an S-shaped trajectory tracking curve of the method proposed in an embodiment of the present invention in a disturbance environment;

[0045] Figure 7 is the trajectory tracking performance in a disturbance environment in an embodiment of the present invention ( );

[0046] Figure 8 The trigger signal for optimizing the solution based on the event trigger mechanism proposed by the present invention;

[0047] Fig. 9 It is the calculation time of each step of the optimization solution proposed by the present invention. DETAILED DESCRIPTION

[0048] The present invention is further described below in conjunction with specific embodiments and the accompanying drawings. More details are set forth in the following description to facilitate a full understanding of the present invention. However, the present invention is obviously capable of being implemented in a variety of other ways different from the description herein. For those skilled in the art, any substitution, improvement or change made to the embodiments of the present invention is within the protection scope of the present invention, and the protection scope of the present invention should not be limited by the content of this specific embodiment.

[0049] The present invention provides an event-triggered robust nonlinear model predictive control method for agricultural machinery trajectory tracking, the method comprising the following detailed steps:

[0050] 1. Constructing the prediction model of the lateral and longitudinal trajectory tracking controller

[0051] A more accurate transverse and longitudinal dynamics model of agricultural machinery is used as the prediction model of the transverse and longitudinal trajectory tracking controller. When studying the tracking problem on the horizontal plane, the operation motion of agricultural machinery is simplified and the following assumptions are made:

[0052] (1) The agricultural machinery working scene is a two-dimensional plane. The vertical movement of the agricultural machinery is not considered, and the vertical movement caused by the external environment can be ignored.

[0053] (2) Without considering the weight distribution between the front and rear axles and the width of the agricultural machine, the center of gravity coincides with the origin of the frame.

[0054] (3) The two front wheels and rear wheels of the agricultural machinery maintain the same posture. During the operation of the agricultural machinery, the rear wheels provide propulsion for driving, and the front wheel angle controls the steering of the agricultural machinery.

[0055] First, establish the lateral dynamics model of agricultural machinery. The force analysis of agricultural machinery during turning is as follows: Figure 1 As shown. represents the geodetic coordinate system, Represents the body coordinate system of the agricultural machinery. The motion relationship of the agricultural machinery in the vehicle earth coordinate system is:

[0056] .

[0057] In the establishment of the lateral dynamics model, the two degrees of freedom of the vehicle's lateral displacement and yaw are mainly considered. According to the force relationship, it can be obtained:

[0058] ;

[0059] in, and They represent the position coordinates in the geodetic coordinate system, They are longitudinal velocity, lateral velocity and yaw rate respectively; is the yaw angle of the agricultural machinery; and They represent the position coordinates in the geodetic coordinate system respectively; and Respectively represent the distance from the front axle and rear axle of the agricultural machinery to the center of mass, It is the quality of agricultural machinery. It's agricultural machinery around The yaw moment of the axis, represents the front wheel turning angle, is the front wheel cornering stiffness, is the rear wheel cornering stiffness, and Respectively represent the lateral forces on the front and rear tires.

[0060] Considering that the front wheel turning angle and its rate of change are generally small, we can get , The present invention adopts the pacejka tire model, and the current tire side deflection angle and rear tire lateral slip angle When the slip ratio is small, the tire lateral force can be linearly approximated as:

[0061] .

[0062] When establishing the longitudinal dynamics model, the traction and resistance of the agricultural machinery are mainly considered. The traction is provided by the engine to the rear wheels, and the resistance mainly includes air resistance and rolling resistance. Figure 2 shown.

[0063] The longitudinal force of the vehicle is expressed as:

[0064] ;

[0065] ;

[0066] ;

[0067] in, The traction force of the rear wheels of agricultural machinery. is the longitudinal wind resistance, is the rolling resistance, is the drag coefficient, is the windshield area, is the air density, is the rolling resistance coefficient.

[0068] According to the force analysis of the lateral and longitudinal dynamics systems of agricultural machinery, the lateral and longitudinal coordinated nominal dynamics controlled model is:

[0069] ;

[0070] in, , , , They are the nominal mass of the agricultural machinery, the stiffness of the front wheels, the stiffness of the rear wheels, and the moment of inertia. Due to external disturbances and the mechanical aging of the agricultural machinery itself, the traction and front wheel steering angle control inputs will be affected by disturbances. In addition, the parameters of the agricultural machinery itself will also be perturbed, such as the lateral stiffness of the tires, the mass of the agricultural machinery itself, and the moment of inertia. Considering the matching disturbance and model uncertainty, the specific expression can be obtained according to the nominal dynamic model:

[0071] ;

[0072] in, , , , ; , , , It is the uncertain part of the model; , It is the unknown disturbance part of the agricultural machinery control input.

[0073] The dynamic model can be written in the form of state-space equations:

[0074] ;

[0075] in represents the state vector, is the control input vector, is the unknown perturbation vector; the coefficient matrix and It is expressed as follows:

[0076] ;

[0077] ;

[0078] Consider the existence of agricultural machinery brake saturation:

[0079] ;

[0080] in, and It represents the upper and lower limits of the input of the agricultural machinery traction force and the front wheel angle. Therefore, the control input can be defined as:

[0081] .

[0082] 2. Radial Basis Function Neural Network Approximation

[0083] The radial basis function neural network (RBFNN) has good universal approximation ability and can approximate any continuous function on the emergency set. It is widely used to process the uncertain model of unmanned systems. Given the following function:

[0084] ;

[0085] In the formula, is the input vector, is the input dimension of the neural network, is the weight vector, is the radial basis function vector. The Gaussian function is chosen as follows:

[0086] ;

[0087] In the formula, is the basis function center, is the width of the Gaussian function. It is defined in the emergency set A continuous function, for any given constant , there exists a constant vector , so that the following equation holds:

[0088] ;

[0089] Defining the optimal weights :

[0090] .

[0091] The present invention adopts radial basis function neural network as the basis function of reinforcement learning actor-critic neural network to approximate the uncertainty of agricultural machinery model.

[0092] In this invention, some useful definitions are proposed: Indicated in After time The system state at the step time, is the vector's two-norm, norm , represents transpose, represents the minimum eigenvalue of the Euclidean vector. For the agricultural machinery tracking process, a All perturbations and model uncertainties are assumed to be bounded.

[0093] 2. Building a nonlinear predictive control system

[0094] The nonlinear predictive control system is composed of a nonlinear model predictive controller based on a pseudo-Huber function to improve the cost function, an actor-critic neural network approximation, an adaptive parameter nonlinear disturbance observer, and an event trigger mechanism, taking into account the input saturation of the actuator. The control scheme principle of this control system is as follows Figure 3 shown.

[0095] 1. Nonlinear prediction model

[0096] For the nominal agricultural machinery system, there is a state space equation:

[0097] ;

[0098] in, represents the nominal system state space expression, represents the control input under the nominal system. For the above state equation, the fourth-order Runge-Kutta method is used for discretization. Relative to time The iterative equation can be obtained:

[0099] ;

[0100] in, , , , .here, and Agricultural machinery The state vector and control input at time, is the sampling time interval.

[0101] In order to facilitate the establishment of nonlinear prediction models, the discretized state space model can be obtained:

[0102] ;

[0103] in, is the fourth-order Runge-Kutta transformation Discretization function of; define the prediction time domain , control time domain , and exists ; The predicted state can be obtained in the following order:

[0104] .

[0105] 2. Optimization solver

[0106] According to the nonlinear model predictive optimization control problem (OCP) of agricultural machinery trajectory tracking, this paper proposes an improved objective function based on the pseudo-Huber function:

[0107] .

[0108] The constraints are:

[0109] ;

[0110] In the formula, Indicates the nominal system The optimal control input sequence set is solved by moment prediction. represents the set of state constraints of the system, is the terminal domain; is the terminal cost, and the weight matrix is ; is the stage cost function.

[0111] In order to improve the tracking accuracy of the turning path of agricultural machinery, the present invention adopts the pseudo-Huber function to improve the stage cost function of the optimization problem, and the following can be obtained:

[0112] ;

[0113] The stage cost in the above formula uses quadratic penalties for speed and control input, and the weight matrix is and . Choose location cost As a pseudo-Huber function, Weighted, penalized the difference between the actual horizontal and vertical posture of the agricultural machinery and the expected posture, and given as follows:

[0114] ;

[0115] is a positive design parameter. It provides a quadratic penalty when the orthogonal position error is small and a linear penalty when the position error is large, which helps improve the tracking performance of the agricultural machinery when a large position error is observed in the turning trajectory. To avoid the heading wrapping problem, the cost function of the agricultural machinery heading is ,by For weights, choose as follows:

[0116] .

[0117] The optimization solver solves the above nonlinear predictive control optimization problem and obtains the following control input sequence results:

[0118] .

[0119] 3. Input Saturation Model

[0120] A smooth model is defined to replace the saturation model of the actuator. The smooth model represents the saturation nonlinearity in a smooth form and is designed as:

[0121] ;

[0122] in, , is the Gaussian error function, is the current running time, is the standard symbolic function, and Respectively represent the maximum and minimum values ​​of the actuator input, is the input of the sign function. The smooth model can describe the asymmetric saturation in a smooth form, and the input saturation can be defined as:

[0123] ;

[0124] It is the ideal control input that the control system finally obtains The actual input of the agricultural machinery actuator after input saturation limitation, is the difference. Substituting it into the state space equation rewritten by the dynamic model, we can get:

[0125] .

[0126] Through this smooth input saturation model, the control input given by the control system can be Smoothly limited to the actuator allowable range.

[0127] 4. Compensation for model uncertainty

[0128] The actor-critic neural network strategy based on reinforcement learning is usually used to model nonlinear functions. In order to compensate for the model uncertainty of agricultural machinery, the actor neural network Used to approximate unknown dynamic model parts and , described as:

[0129] ;

[0130] in, is the optimal weight matrix, is the actor neural network basis function vector, using radial basis function neural network; is the input vector, is the approximate error.

[0131] The critic function evaluates the system performance to adjust the weights of the actor neural network. Define the critic neural network as:

[0132] ;

[0133] in, is the design parameter, is the critic neural network basis function vector; vector Defined as:

[0134] .

[0135] In this form, the Restricted to ,in is the design parameter, It is the error between the actual value of the agricultural machinery status and the reference value.

[0136] In the actual design process, the adaptive law can be used to approximate the optimal weight vector. is designed to approximate the actual critic function and is expressed as follows:

[0137] ;

[0138] Then, the adaptive control rate and As follows:

[0139] ;

[0140] ;

[0141] in, is a positive design parameter.

[0142] Define the state vector of the control system . Nominal coefficient matrix and It is expressed as follows:

[0143] ;

[0144] .

[0145] The state space expression of the agricultural machinery control system considering disturbance and model uncertainty is:

[0146] .

[0147] 5. Adaptive Parameter Nonlinear Disturbance Observer

[0148] In order to estimate the unknown disturbance of the agricultural machinery system, the nonlinear disturbance observer is designed as follows:

[0149] ;

[0150] in, is an estimate of the unknown disturbance, is the intermediate state variable, is the designed parameter matrix, , It is the reinforcement learning neural network's estimate of the model uncertainty.

[0151] The adaptive rate is designed as:

[0152] ;

[0153] in, is the adaptation rate, is the design positive parameter, is the initial value of the adaptive parameter, is the design parameter.

[0154] By compensating for the unknown disturbance observer term and the model uncertainty term, the final control input of the agricultural machinery trajectory tracking controller is designed as follows:

[0155] .

[0156] 6. Event-triggered trajectory tracking optimization control strategy

[0157] Solving the optimal problem of nonlinear model predictive control requires a lot of computational cost, which may lead to a decrease in real-time performance. Considering the practical adaptability of agricultural machinery trajectory tracking, an event-triggered mechanism is used to reduce the calculation frequency of the optimal problem. Compared with the periodically executed nonlinear model predictive control, event-triggered nonlinear model predictive control is an attractive method to reduce the number of solutions of the optimization solver and calculate the control input only when the definition occurs. The non-periodic implementation of event triggering saves a lot of computing resources. Assumptions To solve the current triggering time of the optimization problem, To solve the next triggering time of the optimization problem. The event triggering mechanism of trajectory tracking optimization control can be described as:

[0158] .

[0159] The solution of the optimization problem only occurs in the prediction time domain For a fixed moment in the cycle or when the error between the actual state and the reference state exceeds a certain threshold, is the trigger threshold set. A larger threshold will affect the tracking performance of the agricultural machinery, while a smaller threshold cannot improve the computing performance. Therefore, an appropriate trigger threshold can be selected according to actual needs to balance the tracking performance and computing performance. According to the design of the event trigger mechanism, the interval between two triggers is bounded. Obviously, there is a trigger upper bound For the event triggering strategy proposed in this paper, there is a minimum positive sampling interval as the triggering lower bound, that is, there is no Zeno phenomenon, avoiding infinite triggering in a finite time.

[0160] Based on the above analysis, the pseudo code of the proposed algorithm is shown in Table 1.

[0161] Table 1 Pseudo code of the robust model predictive control agricultural machinery trajectory tracking algorithm based on event triggering

[0162] .

[0163] The event-triggered robust nonlinear model predictive control agricultural machinery trajectory tracking method proposed in the present invention is simulated and analyzed:

[0164] In order to simulate the actual operation path and verify that the agricultural machinery has good tracking performance in both straight and curved trajectories, the present invention sets an S-type reference trajectory. This trajectory consists of a straight trajectory and a thumb-shaped turning trajectory, simulating straight-line operation and turning scenes in the field, such as Figure 4 shown. and Respectively indicate that the agricultural machinery is fixed on the ground The longitudinal and lateral displacements in the inertial coordinate system.

[0165] The present invention designs a Markov time-varying disturbance to verify the robustness of the method of the present invention. The following first-order Markov process is used to describe the external environmental disturbance to the agricultural machinery:

[0166] ;

[0167] in, Indicates the disturbance to which the agricultural machinery is subjected. is the designed constant matrix, is zero-mean Gaussian white noise, express The amplitude matrix of .

[0168] Figure 5 is the Markov time-varying disturbance of agricultural machinery, Figure 6 , Figure 7 The method proposed in the present invention is an agricultural machinery tracking performance under a preset trajectory. The method proposed in the present invention has high tracking accuracy and robustness, and the tracking accuracy meets the actual operation requirements of the agricultural machinery. Figure 8 In order to optimize the solution trigger signal, under the action of the event trigger mechanism, the control system reduces the number of optimization solutions as much as possible. Fig. 9 This is the operation time of the method proposed in the present invention at each sampling moment. It can be seen that when the optimization solver is not performing optimization solution, the operation time is significantly less than the operation time of re-optimization solution, thus saving the control system computing resources.

[0169] The present invention studies the problem of agricultural machinery trajectory tracking under the influence of model uncertainty and unknown disturbances. The lateral and longitudinal dynamic models of agricultural machinery are established, and an event-triggered robust nonlinear model predictive control method for agricultural machinery trajectory tracking is proposed, and the feasibility and stability of the method are verified. The Runge-Kutta discretization and cost function are used to obtain the optimal solution for the sampling time step at each triggering moment. In view of the problem of reduced accuracy of turning trajectory tracking, a pseudo-Huber improved cost function is proposed. On this basis, an adaptive nonlinear disturbance observer combined with a reinforcement learning actor-critic neural network is used to deal with model uncertainty and interference, and an event triggering mechanism is proposed to reduce the calculation frequency of nonlinear model predictive control and reduce the calculation time. The simulation results prove that the method of the present invention has higher stability and robustness, has advantages in saving computing resources, reduces the dependence on high computing performance processors, and is of great significance for agricultural machinery to achieve high-precision trajectory tracking.

Claims

1. An event-triggered robust nonlinear model predictive control method for agricultural machinery trajectory tracking, characterized in that: include: Establish the transverse and longitudinal dynamics model of agricultural machinery and its state space equations to track the trajectory of agricultural machinery; An improved nonlinear model predictive control system is constructed. The improved nonlinear model predictive control system includes a nonlinear controller of the agricultural machinery nominal system with a pseudo-Huber function improving the cost function, and solves the control input of the agricultural machinery nominal system; The optimization solver uses the reinforcement learning actor-critic algorithm to approximate the uncertainty compensation of the agricultural machinery model; the adaptive parameter nonlinear disturbance observer is used to estimate and compensate the external disturbances of the agricultural machinery; through the event trigger mechanism, the error between the true value and the reference value of the agricultural machinery trajectory tracking is used to determine whether to trigger the optimization solver to optimize the solution; finally, the input of the agricultural machinery trajectory tracking controller is obtained: ； ； ； in, and They represent the upper and lower limits of the input of the agricultural machinery traction force and the front wheel angle respectively; Control inputs of nominal systems of agricultural machinery; is the model uncertainty; Adaptive disturbance observer estimate; is the adaptation rate; The objective function and constraint conditions of the nonlinear control model of the nominal system of agricultural machinery using the pseudo-Huber function to improve the cost function are specifically as follows: ； ； in, express The optimal solution of the nominal system at time instant is the control input sequence, represents the set of state constraints of the system, is the terminal domain; For the prediction time domain, To control the time domain and ; is the terminal cost, is the weighting matrix; represents the state space expression of the system after fourth-order Runge-Kutta discretization; It is the preset reference trajectory of agricultural machinery; is the stage cost function, which uses quadratic penalties on speed and control input, expressed as: ； is a weighted matrix that penalizes speed, is the penalty matrix for the control input; They are longitudinal velocity, lateral velocity and yaw rate respectively; is the yaw angle of the agricultural machinery; Indicates the time, represents the step length; and They represent the position coordinates in the geodetic coordinate system respectively; Indicated in After time The state of the agricultural machinery system at the step time; express The nominal system control input of the agricultural machinery at the time; Choose location cost As a pseudo-Huber function, Weighted, penalized differences between the actual horizontal and vertical postures of the agricultural machinery and the expected postures, expressed as: ； in, is a positive design parameter; , Respectively represent the set values ​​of the horizontal position coordinate and the vertical position coordinate of the preset reference trajectory; Agricultural machinery direction cost pass Weighted, penalized agricultural machinery heading surround, expressed as: ； Indicates the set value of the yaw angle in the preset reference trajectory of the agricultural machinery; The uncertainty compensation of the reinforcement learning actor-critic algorithm approximating the agricultural machinery model is expressed as follows: ； in, is an estimate of the optimal weight matrix, is the actor neural network basis function vector; is the input vector, is the approximation error; is the input dimension of the neural network; Represent the reference setting values ​​of the longitudinal speed, lateral speed and yaw rate of the agricultural machinery respectively, They respectively represent the reference set values ​​of the agricultural machinery's lateral position coordinates, longitudinal position coordinates and yaw angle.

2. The event-triggered robust nonlinear model predictive control agricultural machinery trajectory tracking method according to claim 1, characterized in that: The adaptive parameter nonlinear disturbance observer is expressed as follows: ； in, is an estimate of the unknown disturbance, is the intermediate state variable, is the designed parameter matrix, ; and is the nominal coefficient matrix; is the state quantity of the control system; The adaptive rate is designed as: ； in, is the design positive parameter, is the initial value of the adaptive parameter, is the design parameter; A reference set value state quantity representing the lateral position coordinate, longitudinal position coordinate and yaw angle of the agricultural machine; The state quantity that represents the lateral position coordinates, longitudinal position coordinates and yaw angle of the agricultural machinery.

3. The event-triggered robust nonlinear model predictive control agricultural machinery trajectory tracking method according to claim 1, characterized in that: The event triggering mechanism is: ； in, represents the nominal system control input; express The optimal solution of the time-nominal system is the control input sequence; For the prediction time domain, is the error threshold between the actual operation trajectory of the agricultural machinery and the reference trajectory; is the current triggering moment, For the next trigger moment; is the set reference trajectory; is the vector's two-norm, norm , represents transpose, is the weighting matrix.

Images