Unmanned vehicle trajectory tracking control method based on adaptive time domain model predictive control

Abstract

The invention discloses an unmanned vehicle trajectory tracking control method based on adaptive time domain model predictive control, and relates to the technical field of automatic driving, and the method comprises the steps: obtaining a system predictive equation, converting a target function into a standard quadratic programming form function, and obtaining a system predictive equation based on a constraint condition and the standard quadratic programming form function; calculating to obtain a control increment sequence in a control time domain; constructing a model prediction trajectory tracking controller; a model prediction trajectory tracking controller is used for tracking the trajectories of the tested vehicle under different working conditions, and an optimal time domain data set is determined; training the optimal time domain data set by using an adaptive neural fuzzy inference system to obtain a time domain optimization module; and based on the time domain optimization module and the model prediction trajectory tracking controller, generating a self-adaptive time domain model prediction controller which is used for adjusting the prediction time domain and the control time domain of the tested vehicle under different working conditions in real time. The adaptability and trajectory tracking effect of the system are effectively improved, and the vehicle out-of-control risk is reduced.

Description

Technical Field

[0001] The present application generally relates to the field of autonomous driving technology, and specifically to a trajectory tracking control method for an unmanned vehicle based on adaptive time-domain model predictive control. Background Art

[0002] The existing intelligent system of unmanned vehicles is mainly composed of positioning perception, decision planning and tracking control. Among them, tracking control is an important research direction in intelligent unmanned systems and a key step in realizing the closed loop of autonomous driving systems. At present, the commonly used trajectory tracking control algorithms mainly include PID control, pure tracking control, sliding mode control (SMC), linear quadratic regulator control (LQR) and model predictive control (MPC). Compared with the above-mentioned other methods, model predictive control has the ability to predict future vehicle states and handle multi-objective constraints of the system. Therefore, it has been widely used in the research of trajectory tracking control.

[0003] However, traditional model predictive trajectory tracking controllers usually set the prediction time domain and control time domain parameters to fixed values, which makes it difficult to adapt to different working conditions. The commonly used improvement methods mainly improve the time domain parameters of the model predictive controller through various optimization algorithms. However, these improvement methods cannot cope with complex and changeable driving environments in real time. When the road conditions or vehicle speed change, the model predictive trajectory tracking controller is difficult to adjust the parameters in time to maintain tracking performance, which is prone to the risk of increased errors or even loss of control of the controlled vehicle. Therefore, we propose an unmanned vehicle trajectory tracking control method based on adaptive time domain model predictive control to solve the above problems. Summary of the invention

[0004] In view of the above-mentioned defects or deficiencies in the prior art, it is desired to provide an unmanned vehicle trajectory tracking control method based on adaptive time domain model predictive control, which improves the adaptability of the system and the effect of trajectory tracking, reduces the risk of vehicle loss of control and has strong applicability.

[0005] The present application provides an unmanned vehicle trajectory tracking control method based on adaptive time domain model predictive control, comprising the following steps: Obtaining a system prediction equation, taking a control increment of the system prediction equation as an optimization quantity, introducing a relaxation factor, and determining an objective function according to the control increment, the relaxation factor, and a deviation between an output of the system prediction equation and a reference trajectory; The objective function is converted into a standard quadratic programming form function, and constraint conditions are obtained, and based on the constraint conditions and the standard quadratic programming form function, a control increment sequence in a control time domain is calculated; Based on the rolling optimization strategy, the first increment value in the control increment sequence is used as the control input at the current moment to construct a model prediction trajectory tracking controller; The model prediction trajectory tracking controller is used to track the trajectory of the tested vehicle under different working conditions to determine an optimal time domain data set; the optimal time domain data set at least includes an optimal prediction time domain and an optimal control time domain corresponding to different working conditions; Using an adaptive neural fuzzy inference system to train the optimal time domain data set to obtain a time domain optimization module; Based on the time domain optimization module and the model prediction trajectory tracking controller, an adaptive time domain model prediction controller is generated; the adaptive time domain model prediction controller is used to adjust the prediction time domain and control time domain of the vehicle under test under different working conditions in real time.

[0006] According to the technical solution provided in the embodiment of the present application, obtaining the system prediction equation specifically includes the following steps: Based on the three-degree-of-freedom vehicle dynamics model, the nonlinear vehicle dynamics state space equation is established; Linearizing the nonlinear vehicle dynamics state space equation to obtain a linear time-varying state equation; Discretizing the linear time-varying state equation to obtain a discrete state space equation; An incremental model is introduced into the discrete state space equation to obtain a system prediction equation.

[0007] According to the technical solution provided in the embodiment of the present application, the three-degree-of-freedom vehicle dynamics model is obtained according to the following steps: Perform longitudinal, lateral and yaw dynamic force analysis on the vehicle under test to obtain the force balance and moment balance equations of the front and rear wheels of the vehicle under test in the longitudinal and lateral directions; Approximately linearizing the force balance and moment balance equations to obtain a linearized tire model; The linearized tire model is simplified, and a three-degree-of-freedom vehicle dynamics model is determined based on a geodesic inertial coordinate system and a vehicle body coordinate system.

[0008] According to the technical solution provided in the embodiment of the present application, the model prediction estimation tracking controller is used to track the trajectory of the vehicle under test under different working conditions, which specifically includes the following steps: Obtaining a speed collection interval of the vehicle under test, and selecting a plurality of speed collection points with equal intervals in the speed collection interval; Obtaining a road adhesion interval of the vehicle under test, and selecting a plurality of coefficient collection points with equal intervals in the road adhesion interval; Constructing a plurality of operating conditions according to the speed collection points and the road adhesion coefficient collection points; The model is used to predict the trajectory tracking controller to track the trajectory of the vehicle under test under different working conditions.

[0009] According to the technical solution provided in the embodiment of the present application, the optimal time domain data set is obtained according to the following steps: Calculating a lateral tracking error and a yaw angle error of the vehicle under test, and creating a fitness function of a particle swarm algorithm according to the lateral tracking error and the yaw angle error; Based on the fitness function of the particle swarm algorithm, the particle swarm algorithm is optimized to obtain an improved particle swarm optimization algorithm; The improved particle swarm optimization algorithm is used to optimize the time domain parameters to obtain the optimal time domain data set corresponding to different working conditions.

[0010] According to the technical solution provided in the embodiment of the present application, based on the fitness function of the particle swarm algorithm, the particle swarm algorithm is optimized to obtain an improved particle swarm optimization algorithm, which specifically includes the following steps: Get the number of particle populations, spatial dimensions, and maximum number of iterations; Initializing the speed and position of each particle, and calculating the initial fitness value of each particle based on the fitness function of the particle swarm algorithm; According to the nonlinear inertia weight and the asynchronous learning factor, the speed and position of each particle are updated, and the first fitness value of each particle is calculated; When it is determined that there is at least one of the first fitness values ​​that is less than the initial fitness value, and the current number of iterations is less than the maximum number of iterations, updating the position of the corresponding particle to a new position corresponding to the first fitness value, and continuing iteration; Perform random quasi-reflective learning or Gaussian mutation on the particles, calculate the corresponding second fitness value of each particle at the disturbed position, and introduce a greedy mechanism to update the position of each particle; When it is determined that there is at least one of the second fitness values ​​that is less than the first fitness value, and the current number of iterations is less than the maximum number of iterations, updating the position of the corresponding particle to a new position after the disturbance, and continuing the iteration; When it is determined that the current number of iterations is equal to the maximum number of iterations, the iteration is stopped to obtain an improved particle swarm optimization algorithm.

[0011] According to the technical solution provided in the embodiment of the present application, the optimal time domain data set is trained using an adaptive neural fuzzy inference system to obtain a time domain optimization module, which specifically includes the following steps: Using the real-time vehicle speed and road adhesion coefficient of the tested vehicle as input variables of the adaptive neural fuzzy inference system, using the prediction time domain and the control time domain as output variables of the adaptive neural fuzzy inference system, and determining the number of the input variables and the output variables; The type and quantity of membership functions are selected, and based on a preset training cycle and a preset error target, the optimal time domain data set is trained to obtain a time domain optimization module.

[0012] According to the technical solution provided in the embodiment of the present application, an adaptive time-domain model predictive controller is generated according to the following steps: Obtaining the current speed and road adhesion coefficient of the tested vehicle, and inputting them into the time domain optimization module to obtain the optimal prediction time domain and the optimal control time domain; Based on the optimal prediction time domain and the optimal control time domain, the model prediction trajectory tracking controller is updated to obtain an adaptive time domain model prediction controller.

[0013] It can be seen from the above technical solution that the present application has at least the following beneficial effects: The present application discloses a trajectory tracking control method for an unmanned vehicle based on an adaptive time-domain model predictive control, comprising: obtaining a system prediction equation, taking a control increment of the system prediction equation as an optimization quantity, introducing a relaxation factor, and determining an objective function according to the control increment, the relaxation factor, and the deviation between the output of the system prediction equation and a reference trajectory; converting the objective function into a standard quadratic programming form function, and obtaining constraint conditions, and calculating a control increment sequence in a control time domain based on the constraint conditions and the standard quadratic programming form function; based on a rolling optimization strategy, taking the first increment value in the control increment sequence as the control input at the current moment, and constructing a model predictive trajectory tracking controller; using the model predictive trajectory tracking controller to track the trajectory of a vehicle under test under different working conditions, and determining an optimal time-domain data set; the optimal time-domain data set at least includes an optimal prediction time domain and an optimal control time domain corresponding to different working conditions; then using an adaptive neural fuzzy inference system to train the optimal time-domain data set to obtain a time-domain optimization module; finally, generating an adaptive time-domain model predictive controller based on the time-domain optimization module and the model predictive trajectory tracking controller; the adaptive time-domain model predictive controller is used to adjust the prediction time domain and the control time domain of the vehicle under test under different working conditions in real time.

[0014] The present application uses an adaptive neural-fuzzy inference system to train the optimal time domain data set to obtain a time domain optimization module to adjust the model prediction trajectory tracking controller in real time, that is, the two form an adaptive time domain model prediction controller. Through this adaptive time domain model prediction controller, the prediction time domain and control time domain of the vehicle are adjusted in real time, which effectively improves the adaptability of the system and the trajectory tracking effect, reduces the risk of vehicle loss of control, eliminates the complexity and uncertainty of manual parameter adjustment, and enhances performance in changing environments. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] Other features, objects and advantages of the present application will become more apparent from the detailed description of non-limiting embodiments made with reference to the following drawings.

[0016] Figure 1 This is a flow chart of the unmanned vehicle trajectory tracking control method based on adaptive time domain model predictive control.

[0017] Figure 2 Flowchart for obtaining the system prediction equation.

[0018] Figure 3 Flowchart for obtaining the optimal time domain dataset.

[0019] Figure 4 To obtain the flowchart of the improved particle swarm optimization algorithm.

[0020] Figure 5 A schematic diagram of the application scenario.

[0021] Figure 6 Schematic diagram of the three-degree-of-freedom vehicle dynamics model.

[0022] Figure 7 Schematic diagram of training for the optimal time domain dataset. DETAILED DESCRIPTION

[0023] The present application is further described in detail below in conjunction with the accompanying drawings and embodiments. It is to be understood that the specific embodiments described herein are only used to explain the relevant invention, rather than to limit the invention. It is also necessary to explain that, for ease of description, only the parts related to the invention are shown in the accompanying drawings.

[0024] It should be noted that, in the absence of conflict, the embodiments and features in the embodiments of the present application can be combined with each other. The present application will be described in detail below with reference to the accompanying drawings and in combination with the embodiments.

[0025] In order to make the description of the following embodiments clear and concise, a brief introduction to the related technology is first given: In trajectory tracking control based on model predictive control, prediction time domain and control time domain are two key parameters, and their values ​​have a great influence on the tracking control effect. When other parameters of the controller of model predictive control remain unchanged, increasing the prediction time domain can enhance the prediction ability of vehicle state. However, too large a prediction time domain will reduce the control accuracy of closer positions, thereby increasing the tracking deviation; although too small a prediction time domain can improve real-time performance, the predicted vehicle state information is insufficient, which may easily lead to the failure of the front wheels of the vehicle to adjust the direction in time, aggravating the tracking error and even causing the vehicle to lose control. In addition, when other parameters of the controller of model predictive control remain unchanged, increasing the control time domain can improve tracking accuracy, but reduce real-time performance; reducing the control time domain can improve stability and real-time performance, but reduce tracking accuracy. Therefore, whether it is the prediction time domain or the control time domain, the value needs to comprehensively consider factors such as the accuracy, stability and real-time performance of tracking control.

[0026] At present, the commonly used improvement methods are mainly to improve the time domain parameters of the model predictive controller through various optimization algorithms. For example, the first improvement method is to apply the particle swarm optimization algorithm (PSO) to optimize the time domain parameters of the model predictive control, and find the optimal parameter combination under specific working conditions through offline optimization, but the traditional PSO algorithm is prone to fall into the local optimal solution, which makes it difficult to find the global optimal time domain parameter combination. In addition, the offline optimization process of the PSO algorithm takes a long time, and when the working conditions change, the parameters cannot be adjusted in real time, which limits its applicability in actual driving. The second improvement method is to combine fuzzy control with model predictive control and adjust the time domain parameters in real time through a fuzzy controller. However, the rule design of the fuzzy controller is highly dependent on expert experience, which leads to certain limitations when facing variable working conditions. Therefore, although the existing model predictive controller has made some progress in the application of unmanned vehicle trajectory tracking, it still has the defects of poor adaptability and tracking performance.

[0027] In view of this, the present application provides an unmanned vehicle trajectory tracking control method based on adaptive time domain model predictive control, which uses an adaptive neural-fuzzy inference system to train the optimal time domain data set to obtain a time domain optimization module to adjust the model prediction trajectory tracking controller in real time, that is, the two form an adaptive time domain model predictive controller. Through this adaptive time domain model predictive controller, the prediction time domain and control time domain of the vehicle are adjusted in real time, effectively improving the adaptability of the system and the trajectory tracking effect, reducing the risk of vehicle loss of control, and eliminating the complexity and uncertainty of manual parameter adjustment, thereby enhancing performance in changing environments.

[0028] In order to make the unmanned vehicle trajectory tracking control method based on adaptive time domain model predictive control provided in the embodiment of the present application clearer and easier to understand, the method is introduced below in conjunction with the accompanying drawings. Figure 1 As shown in the figure, this figure is a flow chart of an unmanned vehicle trajectory tracking control method based on adaptive time domain model predictive control provided by an embodiment of the present application, the method comprising: S100, obtaining a system prediction equation, taking a control increment of the system prediction equation as an optimization quantity, introducing a relaxation factor, and determining an objective function according to the control increment, the relaxation factor, and a deviation between an output of the system prediction equation and a reference trajectory.

[0029] Among them, Figure 2 As shown, the system prediction equation is obtained, which specifically includes the following steps: S101. Based on a three-degree-of-freedom vehicle dynamics model, a nonlinear vehicle dynamics state space equation is established.

[0030] It should be noted that the three-degree-of-freedom vehicle dynamics model is obtained according to the following steps: The longitudinal, lateral and yaw dynamic force analysis of the tested vehicle is carried out to obtain the force balance and moment balance equations of the front and rear wheels of the tested vehicle in the longitudinal and lateral directions.

[0031] Here, the force balance and moment balance equations are as follows: ; in, is the vehicle mass, is the longitudinal velocity of the center of mass in the vehicle coordinate system, is the lateral velocity of the center of mass in the vehicle coordinate system, is the front wheel turning angle of the vehicle, is the sideslip angle of the center of mass, is the yaw angle, is the yaw angular velocity, is the moment of inertia of the vehicle around the z-axis of the vehicle coordinate system, f is the front wheel of the vehicle, r is the rear wheel of the vehicle, is the distance from the center of mass to the front wheel, is the distance from the center of mass to the rear wheel, is the lateral force on the front wheels of the vehicle, is the lateral force on the rear wheels of the vehicle, is the longitudinal force on the front wheels of the vehicle, The longitudinal force acting on the rear wheels of the vehicle.

[0032] The force balance and moment balance equations are approximately linearized to obtain a linearized tire model.

[0033] Here, based on the above force balance and moment balance equations, the magic formula is used to derive the vehicle tire force and obtain a linearized tire model, which is approximately expressed as a linear function: ; in, is the longitudinal stiffness of the vehicle's front tire, is the longitudinal stiffness of the vehicle's rear tire, is the vehicle front tire cornering stiffness, is the cornering stiffness of the rear tire of the vehicle, is the vehicle's front tire slip rate, is the rear tire slip rate of the vehicle, is the vehicle's front tire slip angle, is the rear tire slip angle of the vehicle.

[0034] The magic formula above is a tire longitudinal force, lateral force and self-aligning moment model established using a special sine function.

[0035] The linearized tire model is simplified, and the three-degree-of-freedom vehicle dynamics model is determined based on the earth inertial coordinate system and the vehicle body coordinate system.

[0036] Here, since the linearized tire model has a higher fitting accuracy when the tire slip angle is small, a small angle assumption is adopted, and when the front wheel steering angle and the tire slip angle are both within a small angle range, the trigonometric function is simplified using approximate numbers.

[0037] The formula for the small angle assumption is: ; Here, the approximate number is expressed using the following equation: ; Based on the above formula, the fixed three-degree-of-freedom vehicle dynamics model is obtained: .

[0038] S102. Linearize the nonlinear vehicle dynamics state space equation to obtain a linear time-varying state equation.

[0039] Here, the nonlinear vehicle dynamics state space equation is: ; in, is the current time; , which is the state quantity; , which is the control quantity; , which is the output; is the output matrix, ; T represents the transpose of the matrix.

[0040] Assuming that the control quantity of the system does not change, by The above nonlinear vehicle dynamics state space equation is linearized by expanding the Taylor formula, ignoring the high-order terms and retaining the first-order terms to obtain the linear time-varying state space equation: ; in, express The state matrix at time is Relative to The Jacobian matrix of ; express The control matrix at time is Relative to The Jacobian matrix of ; ; for The amount of control at any moment; for The control quantity of the moment reference point; for The state quantity at the moment; for The state quantity of the reference point at a certain moment.

[0041] S103. Discretize the linear time-varying state equation to obtain a discrete state space equation.

[0042] Here, the forward Euler method is used for discretization, and after sorting, the discrete state equation at any time is obtained as follows: ; in, ; ; represents the identity matrix; is the sampling period of the discrete system; for any time; is the discrete state error matrix from time t to time k, ; represents the output matrix; .

[0043] S104. Introduce an incremental model into the discrete state space equation to obtain a system prediction equation.

[0044] Here, the incremental model is introduced to change the control quantity in the above discrete state space equation into the control increment form: ; set up , and the new state equation is: ; in, ; ; ; ; I is the unit matrix, d is the current state error, which is also a matrix.

[0045] According to the basic principle of model predictive control, the prediction time domain and control time domain are defined as and , the system prediction equation can be expressed in matrix form as: ; in, is the output of the system in the prediction time domain; is the state quantity of the system in the prediction time domain; is the control increment of the system in the control time domain; is the deviation of the system in the prediction time domain; , , are the corresponding coefficient matrices.

[0046] Here, the objective function is: ; in, represents the objective function; and Represent the prediction time domain and control time domain respectively; , represents the weight matrix; is the weight coefficient of the relaxation factor; is the output reference value; Is an index variable used to traverse from 1 to and An integer.

[0047] S200, converting the objective function into a standard quadratic programming form function, obtaining constraint conditions, and calculating a control increment sequence in a control time domain based on the constraint conditions and the standard quadratic programming form function.

[0048] It should be noted that, in order to facilitate computer solution, the objective function is assumed and derived, and the constant term that has no effect on the solution of the objective function is ignored, and it is converted into a standard quadratic programming formal function: ; in, is a positive definite Hessian matrix, ; , Θ is the coefficient matrix that controls the increment.

[0049] After comprehensively considering the standard quadratic programming formal function and constraints, the controller needs to solve the following problems in each control cycle: ; in, , are the maximum and minimum values ​​of the control quantity respectively; , are the maximum and minimum values ​​of the control increment respectively; , They are hard constraint output and soft constraint output respectively.

[0050] In each control cycle, the control increment sequence in the control time domain can be obtained based on the above formula; ; in, represents the control increment sequence, Represents the control increment.

[0051] S300, based on the rolling optimization strategy, taking the first increment value in the control increment sequence as the control input at the current moment, and constructing a model prediction trajectory tracking controller.

[0052] It should be noted that the first increment value of the control increment sequence is used as the control input at the current moment, that is, applied to the system as the actual control increment: ; in, Indicates the current control amount of the vehicle. The control amount of the vehicle at the last moment.

[0053] The system continuously generates new control sequences through rolling optimization until trajectory tracking control is achieved and a model-predicted trajectory tracking controller is obtained.

[0054] S400, using a model-predicted trajectory tracking controller to track the trajectory of the vehicle under test under different operating conditions, and determine an optimal time domain data set; the optimal time domain data set at least includes an optimal prediction time domain and an optimal control time domain corresponding to different operating conditions.

[0055] The model prediction estimation tracking controller is used to track the trajectory of the vehicle under test under different working conditions, which specifically includes the following steps: The speed collection interval of the vehicle under test is obtained, and a plurality of speed collection points with the same interval are selected in the speed collection interval.

[0056] The road adhesion interval of the vehicle under test is obtained, and a plurality of coefficient collection points with the same interval are selected in the road adhesion interval.

[0057] Multiple working conditions are constructed based on the speed collection points and the road adhesion coefficient collection points.

[0058] The model predictive trajectory tracking controller is used to track the trajectory of the vehicle under test under different working conditions.

[0059] For example, since vehicle speed has a great impact on vehicle control, especially in the tracking control of unmanned vehicles, from low-speed automatic parking to highway cruising, there are corresponding technical requirements and control strategies. Therefore, 10-80 km / h is selected as the speed collection interval for unmanned vehicles, and 8 data collection points are selected for equal intervals, covering a variety of driving conditions from low speed to high speed, while adapting to different road types and mission requirements, and providing sufficient test data for the optimization of unmanned vehicles. Further refinement of speed collection points will introduce data redundancy, increase processing complexity, and have limited effect on improving system performance.

[0060] The above 8 vehicle speed collection points are divided from low speed to high speed as follows: 10 km / h (very low speed driving, suitable for high-precision low-speed navigation, such as automatic parking in parking lots and unmanned delivery vehicles); 20 km / h (lower speed driving, suitable for autonomous cruising in closed environments such as campuses, factories, and ports); 30 km / h (medium and low speed driving, suitable for low-speed urban transportation and smart logistics vehicles driving in industrial parks); 40 km / h (low-medium speed driving, suitable for driverless taxis on urban secondary roads); 50 km / h (medium speed, suitable for unmanned logistics vehicles driving on urban main roads and suburban roads); 60 km / h (high-speed driving, suitable for dynamic lane changes of unmanned vehicles on expressways and suburban connection lines); 70 km / h (medium-high speed driving, suitable for dense traffic flow of unmanned vehicles on suburban expressways or at the entrance of expressways); 80 km / h (ultra-high-speed driving, suitable for unmanned vehicles on highways or open roads in suburbs).

[0061] Different road adhesion coefficients will affect the stability and braking performance of the vehicle. Generally speaking, the adhesion coefficient ranges from 0 to 1. The closer to 1, the drier the road surface and the greater the friction. The closer to 0, the road surface is very slippery or frozen, and the friction is very small. In the vehicle driving environment, an adhesion coefficient below 0.4 indicates extreme road conditions and is difficult to adapt to vehicle operation. An adhesion coefficient above 0.9 indicates a near-ideal state and has low test value. Therefore, the road adhesion coefficient interval (0.4-0.9) is selected to divide 6 data collection points at equal intervals to cover most actual road scenes. The 6 data points cover a balanced range from low adhesion coefficients (such as slippery, muddy roads) to high adhesion coefficients (such as dry, good roads), ensuring that there is data for control and adjustment under various road conditions. Further refinement of the collection points will lead to data redundancy, increase processing complexity, and will not significantly improve system performance.

[0062] The above six road adhesion coefficients are specifically divided into the following forms: 0.4 (low road adhesion coefficient, suitable for slippery roads, the system focuses on anti-skid and speed control); 0.5 (medium-low road adhesion coefficient, suitable for roads after rain or with a small amount of mud and sand); 0.6 (medium road adhesion coefficient, suitable for dry ordinary roads); 0.7 (medium to high surface adhesion coefficient, suitable for high-quality dry roads); 0.8 (higher road adhesion coefficient, suitable for ideal dry roads); 0.9 (high road adhesion coefficient, suitable for extreme driving performance testing).

[0063] Here, in order to obtain comprehensive test data, 8 vehicle speed collection points and 6 road adhesion coefficient collection points are combined to form 48 typical driving conditions. Each condition represents a unique combination of vehicle speed and road conditions. Collecting the optimal time domain data under these conditions can optimize the system's control strategy and ensure that the vehicle can operate stably and safely in various complex environments.

[0064] Furthermore, if Figure 3 As shown, the optimal time domain data set is obtained according to the following steps: S401, calculating the lateral tracking error and the yaw angle error of the vehicle under test, and creating a fitness function of the particle swarm algorithm according to the lateral tracking error and the yaw angle error.

[0065] It should be noted that the lateral tracking error of the vehicle is the reference lateral displacement The actual displacement of the vehicle The yaw angle error is the reference yaw angle The actual yaw angle of the vehicle The former reflects the accuracy of tracking, while the latter reflects the stability of tracking.

[0066] The lateral tracking error and the yaw angle error The index is used as the fitness function of the particle swarm algorithm. This index comprehensively considers the influence of vehicle tracking accuracy and stability. The fitness function created is: ; in, Represents the fitness calculation formula; Indicates the current time; for The lateral tracking error at time, for The yaw angle error at time , , is the weight coefficient of each error.

[0067] S402, based on the fitness function of the particle swarm algorithm, optimizing the particle swarm algorithm to obtain an improved particle swarm optimization algorithm.

[0068] It should be noted that in order to improve the optimization ability of the particle swarm algorithm, the improvement process of the particle swarm algorithm is as follows: (i) Use Logistic chaos mapping to initialize the particle swarm and enhance the randomness of the population; (ii) Perform nonlinear updates on the inertia weight and learning factor to improve the global optimization ability and accuracy; (iii) Randomly perturb the particle position by fusing quasi-reflective learning and Gaussian mutation, and combine the greedy mechanism to avoid it from falling into the local optimum.

[0069] like Figure 4 As shown in the figure, in order to improve the optimization ability of the particle swarm algorithm, the particle swarm algorithm is optimized to obtain an improved particle swarm optimization algorithm, which specifically includes the following steps: S4020, obtain the number of particle populations, spatial dimensions, and maximum number of iterations.

[0070] Among them, the number of particle populations, spatial dimensions and maximum number of iterations can be set according to actual requirements.

[0071] S4021, initialize the speed and position of each particle, and calculate the initial fitness value of each particle based on the fitness function of the particle swarm algorithm.

[0072] Here, the speed of the particle swarm algorithm is and location The change follows the following formula: ; in, is the number of iterations; is the particle number; is the inertia weight; , They are self-learning factor and group learning factor respectively; , for Random numbers within is the individual's best historical position; is the global optimal historical position.

[0073] Furthermore, the position of each particle is initialized using Logistic chaotic mapping. Logistic chaotic mapping has a simple form and rich dynamic behavior. Applying it to the particle swarm algorithm can enhance the randomness and diversity of the initial solution of the population. Its chaotic sequence formula is: ; in, For the Chaotic state variables of time steps; is the control parameter, which is Take the value in The larger it is, the more chaotic the system is.

[0074] The chaotic sequence is used to initialize the particle position, so that it has good randomness and ergodicity in the solution space. The formula for generating the initial solution of the PSO algorithm by the chaotic sequence is: ; in, Indicates the sequence number of the particle; and Respectively represent the maximum and minimum positions of the particles; It is the corresponding sequence value in the chaotic sequence generated by the above chaotic sequence formula.

[0075] S4022. Update the speed and position of each particle according to the nonlinear inertia weight and the asynchronous learning factor, and calculate the first fitness value of each particle.

[0076] Here, in order to balance the speed and accuracy of the search, a nonlinear decreasing function is used to improve the inertia weight. In the early stage, a larger inertia weight is maintained to enhance the global search capability; in the later stage, the inertia weight is gradually reduced to enhance the ability to converge to the optimal solution. The improved update formula is: ; in, , are the maximum and minimum values ​​of the inertia weight respectively; , They are the current iteration number and the maximum iteration number of the particle respectively.

[0077] Learning Factor , The step lengths of the particles moving towards their own optimal and global optimal directions are adjusted respectively. In the initial stage of iterative optimization, The value is large, The value is small, which strengthens the global search capability and prevents the "premature maturity" phenomenon. In the later stage of iterative optimization, The value is small, A large value can enhance local search capabilities and improve algorithm performance.

[0078] ; in, , are the maximum and minimum values ​​of the individual learning factor, respectively; , are the maximum and minimum values ​​of the group learning factor respectively.

[0079] S4023: When it is determined that there is at least one first fitness value that is less than the initial fitness value, and the current number of iterations is less than the maximum number of iterations, the position of the corresponding particle is updated to a new position corresponding to the first fitness value, and iteration continues.

[0080] Among them, if the fitness value of the new position is better than the original position, that is, the first fitness value is less than the initial fitness value, the particle position is updated; otherwise, the particle position remains unchanged.

[0081] S4024. Perform random quasi-reflection learning or Gaussian mutation on the particles, calculate the corresponding second fitness value of each particle at the disturbed position, and introduce a greedy mechanism to update the position of each particle.

[0082] Here, in order to solve the problem that the PSO algorithm is prone to fall into the local optimum, the quasi-reflective learning strategy and Gaussian mutation are integrated to improve it. Quasi-reflective learning determines the location of the quasi-reflective solution by the intermediate value between the current solution and the upper and lower bounds of the algorithm. The formula is as follows: ; in, Learn solutions for the current quasi-reflection; is the current solution; ; and Respectively represent the maximum and minimum positions of the particles; for Random numbers within.

[0083] Gaussian mutation uses random numbers that obey Gaussian distribution to act on the original position to generate a new position, expanding the search breadth of the particle. The formula is: ; in, is the current Gaussian variation solution; is the current solution; is a random variable that follows a Gaussian distribution.

[0084] Using the perturbation update function Quasi-reflective learning and Gaussian mutation are dynamically and randomly executed using Random numbers and Compare, if If the value of is small, quasi-reflective learning perturbation is performed, otherwise Gaussian variation perturbation is performed. Perturbation update function Defined as: ; in, is the current iteration number, is the total number of iterations, and the 0.3 in the formula is used to control the type of disturbance variation.

[0085] S4025. When it is determined that there is at least one second fitness value that is less than the first fitness value, and the current number of iterations is less than the maximum number of iterations, the position of the corresponding particle is updated to a new position after the disturbance, and the iteration continues.

[0086] Here, a greedy mechanism is introduced after the disturbance update to compare the fitness of the new position generated by the disturbance with the original position. If the fitness value of the new position is better than the original position, the particle position is updated; otherwise, the particle position remains unchanged. The formula is: ; in, is the current iteration number of the algorithm; is the particle number; Indicates the new position; Indicates the original position; is the fitness calculation function .

[0087] S4026. When it is determined that the current number of iterations is equal to the maximum number of iterations, the iteration is stopped to obtain an improved particle swarm optimization algorithm.

[0088] S403, using an improved particle swarm optimization algorithm to optimize time domain parameters and obtain optimal time domain data sets corresponding to different working conditions.

[0089] Among them, under each typical working condition, by comparing the fitness value of the particle swarm algorithm, the current optimal prediction time domain and control time domain are obtained, and all working conditions are integrated and extracted from the optimization results to obtain the offline optimal time domain data set, as shown in Table 1. Among them, the first data in the brackets in Table 1 is the prediction time domain , the second data is the control time domain .

[0090] Table 1 Optimal time domain dataset

[0091] S500, using an adaptive neural fuzzy inference system to train the optimal time domain data set to obtain a time domain optimization module.

[0092] Among them, the optimal time domain data set is trained by using an adaptive neural fuzzy inference system to obtain a time domain optimization module, which specifically includes the following steps: The real-time vehicle speed and road adhesion coefficient of the tested vehicle are used as input variables of the adaptive neural fuzzy inference system, the prediction time domain and the control time domain are used as output variables of the adaptive neural fuzzy inference system, and the number of input variables and output variables is determined; The type and quantity of membership functions are selected, and based on a preset training cycle and a preset error target, the optimal time domain data set is trained to obtain a time domain optimization module.

[0093] Here, the membership function type is, for example, a Gaussian membership function, which has good smoothness, simple calculation, and strong adaptability. It can handle continuous input variables in unmanned vehicles well, ensuring that the system can respond smoothly under different vehicle speeds and road adhesion conditions. The number of membership functions is 7, which is enough to subdivide complex input variables (vehicle speed, road adhesion coefficient) into multiple intervals, ensuring that the system can provide reasonable control strategies in each interval while avoiding overfitting and unnecessary computational burden.

[0094] The preset training cycle is, for example, 50, so as to provide the system with sufficient training time so that the model can fully adjust parameters to adapt to different vehicle speeds and road conditions; the preset error target is, for example, 0, so as to ensure that the model minimizes errors during the training process and improves accurate response to different vehicle speeds and adhesion conditions.

[0095] S600, generating an adaptive time-domain model prediction controller based on a time-domain optimization module and a model prediction trajectory tracking controller; the adaptive time-domain model prediction controller is used to adjust the prediction time domain and the control time domain of the vehicle under test in real time under different working conditions.

[0096] It should be noted that the adaptive time-domain model predictive controller is generated according to the following steps: Obtain the current speed and road adhesion coefficient of the vehicle under test, and input them into the time domain optimization module to obtain the optimal prediction time domain and the optimal control time domain; Based on the optimal prediction time domain and the optimal control time domain, the model predictive trajectory tracking controller is updated to obtain an adaptive time domain model predictive controller.

[0097] If the data collected in step S100 changes, the time domain parameters are updated again until the tracking process ends.

[0098] For example: Figure 5 A schematic diagram of the application scenario provided in this application. First, a reference lateral displacement is obtained according to the path planning module. , reference yaw angle and road adhesion coefficient ; The navigation and positioning module collects the status information of the controlled vehicle, including the current location of the vehicle , yaw angle , yaw rate and lateral speed Longitudinal speed ; Use the above information to design an adaptive time-domain model predictive controller. Figure 6 The three-degree-of-freedom vehicle dynamics model is used to design a model-predicted trajectory tracking controller, and then the vehicle lateral tracking error (reference lateral displacement The actual displacement of the vehicle The difference between the yaw angle error and the yaw angle error (reference yaw angle The actual yaw angle of the vehicle The difference of The index is used as the fitness function of the particle swarm algorithm. The improved particle swarm algorithm is used to find the optimal prediction time domain and control time domain under different working conditions to obtain the optimal time domain data set, such as Figure 7 As shown in the figure, the optimal time domain data set is trained through the adaptive neural fuzzy inference system to obtain an optimization module that can adjust the time domain online. The time domain optimization module transmits the optimal prediction time domain and control time domain to the model predictive controller for trajectory tracking according to the current working conditions (vehicle speed and road adhesion coefficient). The controller calculates the control amount and controls the vehicle through the throttle, brake, and steering mechanism. If the working conditions change, the time domain is recalculated, otherwise until the trajectory tracking process ends.

[0099] This application uses Logistic chaos mapping to enhance population randomness, nonlinearly adjust inertia weights and learning factors, and uses a combination of Gaussian mutation and quasi-reflective learning to improve global optimization capabilities and accelerate convergence, thereby quickly and accurately finding the optimal prediction time domain and control time domain. Furthermore, an adaptive neural fuzzy inference system (ANFIS) is used to adjust the prediction time domain and control time domain, which can adjust the time domain parameters online according to real-time driving conditions (vehicle speed and road adhesion coefficient), effectively improving the system's adaptability and trajectory tracking effects, and reducing the risk of vehicle loss of control. In addition, the flexibility of the system is improved, the complexity and uncertainty of manual parameter adjustment are eliminated, and the performance in a changing environment is enhanced.

[0100] The above description is only a preferred embodiment of the present application and an explanation of the technical principles used. Those skilled in the art should understand that the scope of the invention involved in the present application is not limited to the technical solution formed by a specific combination of the above technical features, but should also cover other technical solutions formed by any combination of the above technical features or their equivalent features without departing from the inventive concept. For example, the above features are replaced with (but not limited to) technical features with similar functions disclosed in the present application.

Claims

1. An unmanned vehicle trajectory tracking control method based on adaptive time-domain model predictive control, characterized in that: The following steps are involved: Obtaining a system prediction equation, taking a control increment of the system prediction equation as an optimization quantity, introducing a relaxation factor, and determining an objective function according to the control increment, the relaxation factor, and a deviation between an output of the system prediction equation and a reference trajectory; The objective function is converted into a standard quadratic programming form function, and constraint conditions are obtained, and based on the constraint conditions and the standard quadratic programming form function, a control increment sequence in a control time domain is calculated; Based on the rolling optimization strategy, the first increment value in the control increment sequence is used as the control input at the current moment to construct a model prediction trajectory tracking controller; The model prediction trajectory tracking controller is used to track the trajectory of the tested vehicle under different working conditions to determine an optimal time domain data set; the optimal time domain data set at least includes an optimal prediction time domain and an optimal control time domain corresponding to different working conditions; Using an adaptive neural fuzzy inference system to train the optimal time domain data set to obtain a time domain optimization module; Based on the time domain optimization module and the model prediction trajectory tracking controller, an adaptive time domain model prediction controller is generated; the adaptive time domain model prediction controller is used to adjust the prediction time domain and control time domain of the vehicle under test under different working conditions in real time.

2. The unmanned vehicle trajectory tracking control method based on adaptive time-domain model predictive control according to claim 1 is characterized in that: Obtaining the system prediction equation includes the following steps: Based on the three-degree-of-freedom vehicle dynamics model, the nonlinear vehicle dynamics state space equation is established; Linearizing the nonlinear vehicle dynamics state space equation to obtain a linear time-varying state equation; Discretizing the linear time-varying state equation to obtain a discrete state space equation; An incremental model is introduced into the discrete state space equation to obtain a system prediction equation.

3. The unmanned vehicle trajectory tracking control method based on adaptive time domain model predictive control according to claim 2 is characterized in that: The three-degree-of-freedom vehicle dynamics model is obtained according to the following steps: Perform longitudinal, lateral and yaw dynamic force analysis on the vehicle under test to obtain the force balance and moment balance equations of the front and rear wheels of the vehicle under test in the longitudinal and lateral directions; Approximately linearizing the force balance and moment balance equations to obtain a linearized tire model; The linearized tire model is simplified, and a three-degree-of-freedom vehicle dynamics model is determined based on a geodesic inertial coordinate system and a vehicle body coordinate system.

4. The unmanned vehicle trajectory tracking control method based on adaptive time-domain model predictive control according to claim 1 is characterized in that: The model prediction estimation tracking controller is used to track the trajectory of the vehicle under test under different working conditions, which specifically includes the following steps: Obtaining a speed collection interval of the vehicle under test, and selecting a plurality of speed collection points with equal intervals in the speed collection interval; Obtaining a road adhesion interval of the vehicle under test, and selecting a plurality of coefficient collection points with equal intervals in the road adhesion interval; Constructing a plurality of operating conditions according to the speed collection points and the road adhesion coefficient collection points; The model is used to predict the trajectory tracking controller to track the trajectory of the vehicle under test under different working conditions.

5. The unmanned vehicle trajectory tracking control method based on adaptive time-domain model predictive control according to claim 1 is characterized in that: The optimal time domain data set is obtained by following the steps below: Calculating a lateral tracking error and a yaw angle error of the vehicle under test, and creating a fitness function of a particle swarm algorithm according to the lateral tracking error and the yaw angle error; Based on the fitness function of the particle swarm algorithm, the particle swarm algorithm is optimized to obtain an improved particle swarm optimization algorithm; The improved particle swarm optimization algorithm is used to optimize the time domain parameters to obtain the optimal time domain data set corresponding to different working conditions.

6. The unmanned vehicle trajectory tracking control method based on adaptive time-domain model predictive control according to claim 5 is characterized in that: Based on the fitness function of the particle swarm algorithm, the particle swarm algorithm is optimized to obtain an improved particle swarm optimization algorithm, which specifically includes the following steps: Get the number of particle populations, spatial dimensions, and maximum number of iterations; Initializing the speed and position of each particle, and calculating the initial fitness value of each particle based on the fitness function of the particle swarm algorithm; According to the nonlinear inertia weight and the asynchronous learning factor, the speed and position of each particle are updated, and the first fitness value of each particle is calculated; When it is determined that there is at least one of the first fitness values ​​that is less than the initial fitness value, and the current number of iterations is less than the maximum number of iterations, updating the position of the corresponding particle to a new position corresponding to the first fitness value, and continuing iteration; Perform random quasi-reflective learning or Gaussian mutation on the particles, calculate the corresponding second fitness value of each particle at the disturbed position, and introduce a greedy mechanism to update the position of each particle; When it is determined that there is at least one of the second fitness values ​​that is less than the first fitness value, and the current number of iterations is less than the maximum number of iterations, updating the position of the corresponding particle to a new position after the disturbance, and continuing the iteration; When it is determined that the current number of iterations is equal to the maximum number of iterations, the iteration is stopped to obtain an improved particle swarm optimization algorithm.

7. The unmanned vehicle trajectory tracking control method based on adaptive time-domain model predictive control according to claim 1 is characterized in that: The optimal time domain data set is trained by using an adaptive neural fuzzy inference system to obtain a time domain optimization module, which specifically includes the following steps: Using the real-time vehicle speed and road adhesion coefficient of the tested vehicle as input variables of the adaptive neural fuzzy inference system, using the prediction time domain and the control time domain as output variables of the adaptive neural fuzzy inference system, and determining the number of the input variables and the output variables; The type and quantity of membership functions are selected, and based on a preset training cycle and a preset error target, the optimal time domain data set is trained to obtain a time domain optimization module.

8. The unmanned vehicle trajectory tracking control method based on adaptive time-domain model predictive control according to claim 1, characterized in that: Follow these steps to generate an adaptive time-domain model predictive controller: Obtaining the current speed and road adhesion coefficient of the tested vehicle, and inputting them into the time domain optimization module to obtain the optimal prediction time domain and the optimal control time domain; Based on the optimal prediction time domain and the optimal control time domain, the model prediction trajectory tracking controller is updated to obtain an adaptive time domain model prediction controller.

Images