A vehicle path tracking control method and control system adapted to complex roads

Abstract

The present invention provides a vehicle path tracking control method and control system that adapts to complex roads, the method comprising: solving the vehicle state quantity through discrete iteration of a preset probabilistic vehicle model and obtaining an optimal control sequence; inputting the first control quantity in the optimal control sequence into a real vehicle system to obtain an actual vehicle state quantity; solving the vehicle stability evaluation index based on the absolute value of the difference between the iterative vehicle state quantity and the actual vehicle state quantity to obtain a switching strategy at the current moment; selecting a set of controller weight coefficient combinations based on the switching strategy and calling a vehicle error model. The scheme of the present invention enables unmanned vehicles to effectively cope with road uncertainties, realizes the adaptive control of the path tracking algorithm of unmanned vehicles on roads with varying adhesion coefficients and roads with varying curvature, and effectively improves the stability of path tracking of smart vehicles under complex road conditions.

Description

Technical Field

[0001] The present invention relates to the field of unmanned driving moving vehicle control, and in particular to a vehicle path tracking control method and control system adaptable to complex road surfaces. Background Art

[0002] The real vehicle system is a highly coupled nonlinear system. During driving, it will encounter uncertain changes caused by the road. Therefore, when unmanned vehicles deal with road uncertainties, they are often prone to vehicle loss of control, such as deviation from the track, tire slippage and loss of steering ability.

[0003] In unmanned or intelligent driving vehicles, the trajectory tracking control system can control the vehicle's actuators (such as steering, driving, braking), etc. to make the vehicle travel along the desired trajectory. Generally, it can be divided into two situations: normal working conditions and extreme working conditions. For normal working conditions with small lateral acceleration, a simple geometric, kinematic model or linear dynamic model can usually meet the needs. In extreme conditions with large lateral acceleration, inaccurate models may lead to deterioration of tracking performance. Therefore, it is necessary to consider factors such as tire nonlinearity, slip, roll, pitch motion, and actuator dynamic characteristics to establish a highly accurate nonlinear dynamic model. However, the increase in the complexity of the model will inevitably lead to an increase in design difficulty and calculation. How to construct a mathematical model that is highly accurate and easy to control for extreme conditions is a major challenge in the field of intelligent driving.

[0004] There is a scheme that uses a neural network model as a vehicle dynamics model in two steps. First, a virtual vehicle dynamics data set is used to pre-train the neural network model, and then a real vehicle dynamics data set is used to optimize the neural network model, and finally a data-driven vehicle dynamics model is fitted. This method uses a completely black box vehicle model, which not only requires a large amount of data for model fitting, but also makes it difficult to verify the accuracy of the model and the safety of learning.

[0005] There is a scheme that uses the vehicle kinematic model to design the model predictive controller. The prediction model itself is not changed, but the error value of the real vehicle system feedback is input through the control quantity optimized by the controller to fuzzily control the weight coefficient of the model predictive controller. Although it can improve the path tracking stability of some vehicles during slip, the optimization of the optimized control quantity of the model predictive controller itself does not consider the accuracy of the model. Therefore, the solved control quantity may not be the optimal solution in the real scene.

[0006] The existing technology for regulating intelligent vehicles has the following defects:

[0007] 1. It is impossible to accurately model the uncertainties of the vehicle model, such as vehicle manufacturing errors and the influence of suspension roll. The inaccuracy of the vehicle model will lead to inaccurate prediction of the control quantity, which will fundamentally lead to inaccurate vehicle motion control.

[0008] 2. The existing technology is not perfect enough in modeling the forces of tire models under different adhesion coefficients, which makes the optimization control sequence of the model predictive controller inaccurate and leads to deviations in vehicle control.

[0009] 3. The existing technology mainly improves the adaptability of the controller by adjusting the controller weight. The mainstream vehicle model predictive control algorithm framework has fewer optimizable items, so there are fewer adjustable parameters. Summary of the invention

[0010] Based on the problems existing in the prior art, the present invention provides a vehicle path tracking control method and control system that is suitable for complex roads. The specific scheme is as follows:

[0011] A vehicle path tracking control method adapted to complex roads includes the following steps:

[0012] In the preset model prediction profile controller, the vehicle state quantity of the prediction step is solved by discrete iteration of the preset probability vehicle model, and the optimal control sequence at the current moment is obtained; the optimal control sequence includes predicting the control quantity at multiple moments after the current moment;

[0013] Recording the vehicle state quantity after the first iteration of the probabilistic vehicle model as the iterated vehicle state quantity; inputting the first control quantity in the optimal control sequence into the real vehicle system, and observing the actual vehicle state quantity of the real vehicle system;

[0014] Based on the absolute value of the difference between the iterated vehicle state quantity and the actual vehicle state quantity, the stability evaluation index of the vehicle is solved, and the switching strategy at the current moment is obtained by logically judging and classifying the stability evaluation index;

[0015] A set of controller weight coefficient combinations is selected based on the switching strategy and applied to the model predictive profile controller to complete parameter adjustment of the model predictive profile controller; wherein different controller weight coefficient combinations correspond to controller modes under different adhesion coefficient types of road surfaces;

[0016] Based on the switching strategy, a vehicle error model is called, and the vehicle error model is updated to the probabilistic vehicle model to complete the vehicle path tracking control at the current moment; wherein different vehicle error models are constructed based on vehicle data under different adhesion coefficient types of road surfaces.

[0017] In a specific embodiment, the probabilistic vehicle model includes a vehicle error model fitted by a Gaussian process model and a vehicle dynamics model based on tire side slip considerations.

[0018] In a specific embodiment, the vehicle error model E(X k ,U k ) is expressed as:

[0019]

[0020] Among them, GP(Z k ) represents a Gaussian process model with five input dimensions and three output dimensions, and its output data is and

[0021] Among them, the position of the vehicle's center of mass is (x, y), v x 、v y are the longitudinal and lateral velocities of the vehicle's center of mass, is the heading angular velocity;

[0022] The expression of the probabilistic vehicle model is:

[0023]

[0024] Among them, f(X k ,U k ) represents the discretized vehicle dynamics model, Indicates that at time k, according to the current vehicle state X k And the optimized control quantity U k The vehicle state quantity at the next moment is delivered.

[0025] In a specific embodiment, each control variable includes target items including hysteresis error, contour error, orientation error, boundary error, vehicle vector speed, path centerline speed, acceleration and deceleration aggressiveness, and steering aggressiveness;

[0026] The controller weight coefficient combination is composed of the weight coefficients of each target item in the control quantity, and the combination of weight coefficients of different target items obtains different controller weight coefficient combinations.

[0027] In a specific embodiment, in the model prediction contour controller, the road centerline is used as the reference path, and the path points are discretized and represented as a set {s0, s1, ..., s k ,...s max};

[0028] At each moment, the path point s k There are corresponding coordinates [x c (s k), y c (s k )], heading angle Track radius R c (s k ), road width W;

[0029] By introducing a centerline velocity v c (s k ) Discretize the path s k+1 =s k +v c (s k )t, t is the time of each prediction step to ensure the vehicle's driving progress in the direction of the road centerline.

[0030] In a specific embodiment, the hysteresis error e is defined according to the geometric position of the vehicle and the trajectory line. l , contour error e c , orientation error e o , boundary error e off The spatial pose error including:

[0031]

[0032]

[0033]

[0034]

[0035] Error Optimization Term of Model Predictive Contour Controller for:

[0036]

[0037] Among them, q c Indicates the contour error e c The weight coefficient, q l Represents the hysteresis error e l The weight coefficient, q o Represents the orientation error e o The weight coefficient, q off Represents the boundary error e off The weight coefficient of Represents the vehicle state at time k.

[0038] In one embodiment, the vehicle vector velocity and the path centerline velocity v c Speed ​​optimizations The expression is as follows:

[0039]

[0040] Among them, q v Represents the vehicle vector speed as the weight coefficient, q ce A weight factor representing the speed of the centerline of the path.

[0041] In a specific embodiment, the dynamic optimization items regarding the acceleration aggressiveness τ (0 < τ < 1)), the deceleration aggressiveness τ (-1 < τ < 0) and the steering aggressiveness δ are: The expression is as follows:

[0042]

[0043] Among them, q τ The weight coefficient indicating the degree of acceleration, q -τ The weight coefficient indicating the degree of deceleration positivity, q δ The weight coefficient indicating the degree of positivity of the turn.

[0044] In a specific embodiment, the process of obtaining the switching strategy includes:

[0045] The absolute value of the difference between the iterated vehicle state quantity and the actual vehicle state quantity is used as the original error expression; the vehicle driving state error quantity is selected;

[0046] Based on a preset state quantity screening matrix, extracting a true error expression about the vehicle driving state error quantity from the original error expression;

[0047] Constructing a plurality of stability evaluation indicators based on the true error expression;

[0048] Taking the stability evaluation index as input, the corresponding switching strategy involving the model predictive profile controller mode is selected.

[0049] In a specific embodiment, the control quantity U at time k is taken as k , input into the probabilistic vehicle model to obtain the iterative vehicle state at time k+1 Control quantity U k The actual vehicle state quantity input into the real vehicle system and observed at time k+1 Get the true error expression The expression is:

[0050]

[0051] Among them, the state quantity screening matrix B d =[I 3×3 , 0 3×3 ].

[0052] In a specific embodiment, three stability evaluation indicators are constructed based on the true error expression, namely Error1, Error2, and Error3;

[0053] Error1 is The second norm of

[0054] As the vehicle travels, the Error1 sequence can be obtained, K = [k, k-1, k-2, k-3, ..., kN], where N is the length of the historical Error1 sequence used at time k. Assuming that the linear regression function h w (k) = K T w k , in order to make the regression function more consistent with the true value, the least squares fitting loss function J is defined as

[0055]

[0056] w k It is the derivative of the linear regression function based on the past N+1 Error1 error points at each k moment, and Error2 is obtained k =w k ;

[0057] Error3 is the smoothing of Error1. Error3 is obtained by averaging the past N Error1 error points at each k moment.

[0058] In a specific embodiment, N types of adhesion coefficient type road surfaces are defined based on the adhesion coefficient;

[0059] A vehicle error model suitable for each type of adhesion coefficient road surface is obtained by training the data under each type of adhesion coefficient road surface;

[0060] N groups of controller weight coefficient combinations are set, and each group of controller weight coefficient combination corresponds to a type of adhesion coefficient road surface.

[0061] A vehicle path tracking control system adapted to complex roads, used to implement any of the above-mentioned vehicle path tracking control methods; the vehicle path tracking control system comprises:

[0062] Vehicle error model database, which is configured with vehicle error models under different adhesion coefficients;

[0063] A probabilistic vehicle model, connected to a vehicle error model database, configured with a vehicle dynamics model related to tire sideslip, and capable of calling a vehicle error model from the vehicle error model database;

[0064] A model prediction profile controller is connected to the probabilistic vehicle model and is used to perform multi-objective optimization solutions for each target item through discrete iteration within a prediction step using the probabilistic vehicle model as a prediction model to obtain an optimized control sequence;

[0065] a controller strategy switching module, connected to the error model database and the controller parameter adjustment module, for obtaining an iterative vehicle state quantity and an actual vehicle state quantity, solving a stability evaluation index of the vehicle based on an absolute value of a difference between the iterative vehicle state quantity and the actual vehicle state quantity, and obtaining a switching strategy at a current moment by performing logical judgment and classification on the stability evaluation index, and outputting the switching strategy to the error model database and the controller parameter adjustment module respectively;

[0066] The controller parameter adjustment module is connected to the model prediction profile controller and is configured with weight coefficient combinations under different adhesion coefficient road surfaces, and is used to output corresponding weight coefficient combinations to the model prediction profile controller according to a switching strategy.

[0067] In a specific embodiment, it also includes a real vehicle system;

[0068] The real vehicle system is connected to the model prediction profile controller and the controller strategy switching module respectively, and is used to receive the first control item of the optimal control sequence output by the model prediction profile controller, and detect the actual vehicle state quantity obtained under the control of the first control item.

[0069] The present invention has the following beneficial effects:

[0070] The present invention provides a vehicle path tracking control method and control system that are adaptable to complex road conditions, so that unmanned vehicles can effectively cope with road uncertainties, and realize the adaptive control of the path tracking algorithm of unmanned vehicles on roads with varying adhesion coefficients and roads with varying curvatures, effectively improving the stability of the path tracking of intelligent vehicles under complex road conditions. Based on different road conditions, corresponding vehicle error models are constructed, and the dynamic model of the vehicle can be adjusted according to the road conditions, so that the controller is reliable and stable and has the ability to adapt to road conditions. The execution difference of the real vehicle system and the vehicle model on the same control quantity is used as feedback to obtain different switching strategies, and then different weight coefficients and error models are obtained to achieve accurate switching of strategies.

[0071] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, preferred embodiments are given below and described in detail with reference to the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0072] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings required for use in the embodiments are briefly introduced below. It should be understood that the following drawings only show certain embodiments of the present invention and therefore should not be regarded as limiting the scope. For ordinary technicians in this field, other related drawings can be obtained based on these drawings without creative work.

[0073] Figure 1 is a flow chart of a vehicle path tracking control method according to an embodiment of the present invention;

[0074] Figure 2 is a schematic diagram of a vehicle path tracking control system module according to an embodiment of the present invention;

[0075] Figure 3 is a schematic diagram of a vehicle dynamics model considering tire cornering characteristics according to an embodiment of the present invention;

[0076] Figure 4 Schematic diagram of the logic determination principle of the stability evaluation index of the embodiment of the present invention;

[0077] Figure 5 2 is a diagram showing simulation results of an embodiment of the present invention.

[0078] Figure numerals: 1 - vehicle error model database; 2 - probabilistic vehicle model; 3 - model prediction profile controller; 4 - controller strategy switching module; 5 - controller parameter adjustment module; 6 - real vehicle system. DETAILED DESCRIPTION

[0079] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0080] It should be noted that the controller mode in the present invention refers to different control modes of the model prediction profile controller.

[0081] Example 1

[0082] This embodiment proposes a vehicle path tracking control method that is adapted to complex roads, which can solve the adaptive control problem of the path tracking algorithm of unmanned vehicles on roads with varying adhesion coefficients and curvatures, and improve the stability of the path tracking of smart vehicles under complex road conditions. The flow chart of the vehicle path tracking control method is shown in the attached manual. Figure 1 The specific scheme is as follows:

[0083] A vehicle path tracking control method that adapts to complex roads can predict the vehicle state quantity at each moment in the future N moments, solve the optimal control sequence at the future M moments, and complete a moment of cyclic control based on the vehicle state quantity and the optimal control sequence. The cycle at each moment includes the following steps:

[0084] 101. In the preset model prediction profile controller, the vehicle state quantity of the prediction step is solved by discrete iteration of the preset probability vehicle model, and the optimal control sequence at the current moment is obtained; the optimal control sequence includes the control quantity predicted at multiple moments after the current moment;

[0085] 102. Record the vehicle state quantity after the first iteration of the probabilistic vehicle model as the iterated vehicle state quantity; input the first control quantity in the optimal control sequence into the real vehicle system, and observe the actual vehicle state quantity of the real vehicle system;

[0086] 103. Based on the absolute value of the difference between the iterative vehicle state quantity and the actual vehicle state quantity, the stability evaluation index of the vehicle is solved, and the switching strategy at the current moment is obtained by logically determining and classifying the stability evaluation index;

[0087] 104. Based on the switching strategy, a set of controller weight coefficient combinations are selected and applied to the model prediction profile controller to complete parameter adjustment of the model prediction profile controller; wherein different controller weight coefficient combinations correspond to controller modes under different adhesion coefficient types of road surfaces;

[0088] 105. Based on the switching strategy, a vehicle error model is called, and the vehicle error model is updated to the probabilistic vehicle model to complete the vehicle path tracking control at the current moment; wherein different vehicle error models are constructed based on vehicle data under different adhesion coefficient types of road surfaces.

[0089] The vehicle path tracking control method of this embodiment needs to be completed with the help of a specific vehicle path tracking control system. Figure 2 A vehicle path tracking control system is provided. The system includes a vehicle error model database, a probabilistic vehicle model, a model prediction profile controller, a controller parameter adjustment module, and a controller strategy switching module.

[0090] It should be noted that the vehicle path tracking control method of this embodiment is applicable to complex road conditions. In view of the road conditions that the vehicle is most likely to encounter during driving and for safety considerations, a vehicle error model that can adapt to complex road conditions is constructed. Preferably, this embodiment obtains three vehicle error models by training and optimizing the Gaussian process model for vehicle driving data on three different adhesion coefficient roads. In practical applications, other classification methods or classification numbers can be adopted for classification according to actual conditions.

[0091] Based on the adhesion coefficient, N types of adhesion coefficient type roads are defined; a vehicle error model suitable for the adhesion coefficient type road surface is obtained by data training under each adhesion coefficient type road surface, and at least N vehicle error models are preset in the error model database. N groups of controller weight coefficient combinations are set, and each group of controller weight coefficient combinations corresponds to an adhesion coefficient type road surface.

[0092] Specifically, the adhesion coefficient is used to quantify the road surface. According to the adhesion coefficient statistical data of different road surfaces, they are classified according to the numerical values ​​and divided into high adhesion coefficient road surface, medium adhesion coefficient road surface and low adhesion coefficient road surface, as shown in Table 1.

[0093] Table 1 Examples of high, medium and low adhesion coefficient roads

[0094] Adhesion coefficient Road surface type High adhesion coefficient road surface <![CDATA[μ1=0.9]]> Dry asphalt or concrete, wet concrete Medium adhesion coefficient road surface <![CDATA[μ2=0.6]]> Wet asphalt, gravel, dry or wet dirt roads Low adhesion road surface <![CDATA[μ3=0.25]]> Snow, rainy days

[0095] The collection of driving data involves the extreme state of the vehicle. The collection of real vehicle driving data on low adhesion coefficient roads involves the extreme working conditions of the vehicle, which requires an extremely high threshold and is accompanied by the side slip phenomenon of the vehicle. The risk is relatively high and the driving safety of the real vehicle cannot be guaranteed. Therefore, this embodiment uses high-accuracy vehicle dynamics simulation software to replace the real vehicle for driving data collection. The road adhesion coefficient is classified by value, and the vehicle driving data is collected in a simulation environment to train the Gaussian process vehicle error model, and a vehicle error model database that can be called arbitrarily is generated.

[0096] According to the classification method in Table 1, the road pavement adhesion coefficients in the vehicle dynamics simulation software are set to μ1, μ2, and μ3 respectively. In the route setting of the vehicle dynamics software, a route with a large curvature change is selected. The route refers to the existing international F1 formula racing track. The use of the racing track can maximize the vehicle's extreme working conditions and is conducive to collecting vehicle driving data in the critical stability area of ​​the vehicle.

[0097] The vehicle driving data collected on μ1, μ2, and μ3 road surfaces can be used to establish three Gaussian process vehicle error models respectively according to the vehicle error model modeling method, and stored in the industrial computer to form a vehicle error model database. The three vehicle error models are stored in parallel and do not interfere with each other. The vehicle controller can only call one of the vehicle error models at a time. For different adhesion coefficient road surfaces, the appropriate probabilistic vehicle model can be dynamically adjusted according to the custom switching strategy. The use of a multi-Gaussian process error model database method to fit the vehicle model mismatch error under different adhesion coefficient road surfaces reduces the model complexity and avoids the computational complexity of the traditional vehicle dynamics model's precise modeling. At the same time, the vehicle dynamics simulation environment can be used to collect training data, which is economical and safe.

[0098] The vehicle error model database module has multiple preset vehicle error models. Under typical road surfaces with different adhesion coefficients, the actual vehicle driving data and controller prediction data are collected, and Gaussian process vehicle error models of the vehicle under the above typical adhesion coefficient types of high, medium and low road surfaces are established to generate a vehicle error model database.

[0099] Specifically, as the prediction model of the model prediction profile controller, the probabilistic vehicle model includes a vehicle error model fitted by a Gaussian process model and a vehicle dynamics model based on tire side slip considerations. Figure 3 A schematic diagram of the vehicle dynamics model considering tire cornering characteristics is given.

[0100] It is stipulated that in the global coordinate system XOY, the position of the vehicle's center of mass is (x, y), is the heading angle of the vehicle, which is specified in the vehicle's own coordinate system, v x 、v y are the longitudinal and lateral velocities of the vehicle's center of mass, is the angular velocity of the heading, m is the mass of the vehicle, I is the moment of inertia of the vehicle, and L r , L f is the distance from the front and rear axles of the wheels to the center of mass, F f，y 、F r，y is the lateral tire force on the front and rear axles, F f，x 、F r，x is the longitudinal tire force of the front and rear axles, and δ is the front wheel turning angle. The differential equation expression of the vehicle dynamics model is obtained as follows:

[0101]

[0102] Input vector U = [δ, τ] T , τ∈[-1, 1] represents the acceleration or deceleration command of the vehicle pedal, and the longitudinal driving force F x Linearly related to τ, denoted as F x =C a τ, C a is the gain factor, the value changes when accelerating or decelerating. f，x 、F r，x The longitudinal tire force on the front and rear axles is defined as F r，x =ξF x , F f，x =(1-ξ)F x , ξ∈[0,1].

[0103] The vehicle dynamics model is discretized using the fourth-order Runge-Kutta method to obtain:

[0104] X k+1 =f(X k , Uk )

[0105] The vehicle error model is implemented using non-parametric modeling Gaussian process regression. First, a training data set for the Gaussian process model needs to be established. Based on each time sampling point K, the training data set is defined as

[0106] D={(x i ,y i )|i=1,2,...,n}

[0107] Multidimensional input sample points Multidimensional output sample points Output matrix C Y =[0 3×3 , I 3×3 ].

[0108] Gaussian process vehicle error model E(X k , U k ) is expressed as:

[0109]

[0110] Among them, GP(Z k ) represents a Gaussian process model with five input dimensions and three output dimensions. and∑ k It can be understood as an intermediate variable of output, and its output data is and

[0111] Based on the above derivation, the probabilistic vehicle model is expressed as

[0112]

[0113] Among them, f(X k , U k ) represents the discretized vehicle dynamics model, Indicates that at time k, according to the current vehicle state X k And the optimized control quantity U k The vehicle state quantity at the next moment is passed. In the model predictive controller, at any k moment, the current vehicle state quantity X k And the optimized control quantity U k Recursively calculate the vehicle state at the next moment

[0114] Step 101, applying the probabilistic vehicle model to the model prediction profile controller. Through the discrete iteration of the probabilistic vehicle model, the vehicle state quantity of each prediction step can be solved, and the model prediction profile controller is used to perform multi-objective optimization tasks according to optimization items and constraint items within the prediction step to solve the optimal control sequence.

[0115] Preferably, each control quantity includes target items including hysteresis error, contour error, orientation error, boundary error, vehicle vector speed, path centerline speed, acceleration and deceleration aggressiveness, and steering aggressiveness. The controller weight coefficient combination is composed of the weight coefficients of each target item in the control quantity, and the combination of weight coefficients of different target items obtains different controller weight coefficient combinations.

[0116] The larger number of weight coefficients in the model prediction contour controller can make more detailed requirements definition for the optimization task of the controller. At the same time, the three groups of controller weight coefficient combinations that are pre-set can be switched to cope with the weight coefficient combinations of the adhesion coefficient road surface, making the controller reliable, stable and adaptive. Compared with the fuzzy controller's real-time adjustment of the control weight coefficient method which requires setting expert experience rules, the solution of this embodiment is simpler and more efficient.

[0117] In the model predictive contour controller, for target items such as lag error, contour error, orientation error, boundary error, path centerline speed, acceleration and deceleration aggressiveness, and steering aggressiveness, multi-objective optimization is performed within the prediction step to obtain the optimal control sequence.

[0118] The controller reference model predictive contour control algorithm framework of this embodiment takes the road centerline as the reference path and discretizes the path points into a set {s0, s1, ..., s k ,...s max}; Every moment path point s k There are corresponding xy axis coordinates [x c (s k ), y c (s k )], heading angle Track radius R c (s k ), road width W; by introducing a centerline speed v c (s k ), the speed ensures that the vehicle projection ensures the vehicle projection's driving progress in the centerline direction. Discretize the path as follows:

[0119] s k+1 =s k +v c (s k )t

[0120] Where t is the sampling time of each prediction step, and the trajectory sequence is: s0, s1, ..., s k ,...

[0121] In this embodiment, each target item includes an error optimization item, a speed optimization item, and a dynamic optimization item, and a cost function is defined by integrating each target item.

[0122] The hysteresis error e is defined based on the geometric position of the vehicle and the trajectory. l , contour error e c , orientation error e o , boundary error e off The spatial pose error.

[0123] Hysteresis error l for:

[0124]

[0125] Contour error e c for:

[0126]

[0127] Orientation error e o for:

[0128]

[0129] Boundary error e off for:

[0130]

[0131] The above errors are integrated to obtain the geometric error optimization term of the model prediction contour controller. The error optimization term for:

[0132]

[0133] Among them, q c Indicates the contour error e c The weight coefficient, q l Represents the hysteresis error e l The weight coefficient, q o Represents the orientation error e o The weight coefficient, q off Represents the boundary error e off The weight coefficient of Represents the vehicle state at time k.

[0134] Set vehicle speed optimization items in the vehicle speed optimization task, specifically involving vehicle vector speed and the path centerline velocity v c , speed optimization The expression is as follows:

[0135]

[0136] Among them, q v Represents the vehicle vector speed as the weight coefficient, q ce represents the weight coefficient of the path centerline velocity, is the 2-norm of the vehicle vector velocity; v c is the centerline speed, and the vehicle speed optimization strategy can be adjusted through the weight coefficient.

[0137] The dynamic optimization item can change the degree of activeness of the control amount change by adjusting the weight coefficient of the control item, including the degree of activeness of acceleration, the degree of activeness of deceleration and the degree of activeness of steering. Dynamic optimization items about the degree of activeness of acceleration τ (0<τ<1)), the degree of activeness of deceleration τ (-1<τ<0) and the degree of activeness of steering δ The expression is as follows:

[0138]

[0139] Among them, q τ The weight coefficient indicating the degree of acceleration, q -τ The weight coefficient indicating the degree of deceleration positivity, q δ The weight coefficient indicating the degree of positivity of the turn.

[0140] According to the optimization items described above, this embodiment provides a comprehensive optimization problem of iterating the model predictive controller within the prediction step length when the vehicle state quantity is estimated as X0(t) at time t. Considering the above multiple objectives, the cost function is defined as:

[0141]

[0142] The constraints are:

[0143]

[0144] s k+1 =s k +v c (s k )t

[0145] X0=X(t),s0=s(k)

[0146] δ min <δ k <δ max ,

[0147] -1<τ<1,

[0148] Among them, the vehicle turning angle δ k , acceleration and deceleration instructions τ have constraints, and the weight coefficients are as follows q l ,q c ,q o ,q off ,q v ,q ce ,q δ ,q -τ ,q τ , different optimization effects can be achieved through different combinations. There are many parameters that can be optimized and regulated, and the control precision and accuracy are improved. Through the optimization solver (such as IPOPT, FORCES PRO) to perform the above optimization tasks, the controller can plan an optimal driving trajectory and the optimal control sequence based on the current path.

[0149] After obtaining the optimal control sequence, take the first control quantity in the optimal control sequence. In this embodiment, the algorithm is cycled once at each real moment, and the first control quantity of the prediction step is taken each time. The model will predict the control quantity from the current moment to the future M moments, and the control quantity at each moment constitutes the optimal control sequence. Taking the first control quantity as the actual control quantity is to take the control quantity at the next moment of the current moment. Similarly, the vehicle state quantity after the first iteration of the probabilistic vehicle model is recorded because the vehicle state quantity after the first iteration represents the vehicle state quantity at the next moment of the current moment. The reason for predicting the state of the next multiple moments instead of just predicting the state of the next moment is to make a prediction of the vehicle state at multiple moments in the future at the current moment and optimize a predictive control sequence. By optimizing the state for a period of time in the future, the optimal control within a period of time in the future is solved to achieve the purpose of predictive control. Taking the first control quantity takes into account that the situation encountered by the system in the future may change, so it is done every time, which is called rolling optimization in model predictive control. This not only guarantees the prediction ability, but also guarantees the rolling optimization ability of the system.

[0150] In the path tracking control system of this embodiment, a real vehicle system is preset. The vehicle state quantity after the first iteration of the probabilistic vehicle model is recorded as the iterative vehicle state quantity; the first control quantity in the optimal control sequence is input into the real vehicle system, and the actual vehicle state quantity of the real vehicle system is observed. The control quantity contains various instructions for controlling the vehicle's actions. The real vehicle system generates corresponding vehicle state quantities according to the control quantities, and obtains the real vehicle state quantities through vehicle sensors. The real vehicle state quantity and the iterative vehicle state quantity reflect the reactions of the real vehicle and the vehicle model to the same instruction. Only when the reactions are closer, the vehicle model can be closer to the real vehicle. Therefore, it is necessary to calculate the absolute value of the difference between the real vehicle state quantity and the iterative vehicle state quantity, so as to define the vehicle stability evaluation index.

[0151] The process of obtaining the switching strategy includes: taking the absolute value of the difference between the iterated vehicle state quantity and the actual vehicle state quantity as the original error expression; selecting the vehicle driving state error quantity; extracting the real error expression about the vehicle driving state error quantity from the original error expression based on the preset state quantity screening matrix; constructing multiple stability evaluation indicators based on the real error expression; taking the stability evaluation indicator as input, selecting the corresponding switching strategy involving the model prediction profile controller mode. The vehicle controller mode switching strategy based on the real-time vehicle state error data, the vehicle error model designed by the vehicle stability evaluation indicator and the controller weight coefficient combination switching rule.

[0152] The switching strategy of the vehicle controller mode is to make a decision on the call of the vehicle error model in the probabilistic vehicle model and to select the combination of controller weight coefficients in the controller weight coefficient adjustment module. Regarding the stability evaluation index, preferably, the scheme of this embodiment sets three indicators. Regarding the vehicle driving state error amount, for example, the vehicle position (x, y) and heading angle can be selected. State quantity error.

[0153] The optimal control sequence U at each k-th moment is known k = {U k ,U k+1 ,…,U k+m-1}, where m is the length of the control sequence. Take the control quantity U at time k k , input into the probabilistic vehicle model to obtain the iterative vehicle state at time k+1 Control quantity U k The actual vehicle state quantity input into the real vehicle system and observed at time k+1 Get the true error expression The expression is:

[0154]

[0155] Among them, the state quantity screening matrix B d =[I 3×3 ,0 3×3 ]. Filter matrix B by state quantity d =[I 3×3 ,0 3×3 ] can obtain the true error expression of the real-time vehicle state regarding the vehicle driving state error.

[0156] Based on the true error expression, three stability evaluation indicators are constructed as the input of the switching strategy, namely Error1, Error2, and Error3.

[0157] The first indicator Error1 is The second norm of

[0158] The second indicator Error2 is defined as follows: As the vehicle travels, the Error1 sequence can be obtained, K = [k, k-1, k-2, k-3, ..., kN], N is based on the length of the historical Error1 sequence used at time k, assuming that the linear regression function h w (k) = K T w k , in order to make the regression function more consistent with the true value, the least squares fitting loss function J is defined as:

[0159]

[0160] w k It is the derivative of the linear regression function based on the past N+1 Error1 error points at each k moment, and we get:

[0161] Error2 k =w k

[0162] The third indicator Error3 is the smoothing of Error1. At each k moment, the average of the past N Error1 error points is used to obtain Error3.

[0163] Based on the defined vehicle stability evaluation index, a switching rule based on the vehicle stability evaluation index is designed, such as Figure 4 As shown, the input data are Error1, Error2, and Error3 at each sampling moment.

[0164] In each pre-set controller mode, there is a corresponding model and a corresponding weight coefficient in the controller. The logic judgment classification is specifically as follows:

[0165] When Error1 > N1 or Error2 > N2, the controller determines that the vehicle is about to enter an out-of-control state. Considering the handling safety of the driverless vehicle, it directly switches to Model 3 with the highest safety factor and the corresponding controller weight parameters, which is called GP3 mode; GP3 can be understood as a conservative driving mode.

[0166] When Error1 < N3, Error2 < N4, and Error3 < N5, the controller determines that the vehicle enters a relatively controllable state and switches to Model 2 and the corresponding controller weight parameters, which is called GP2 mode; GP2 can be understood as a normal driving mode.

[0167] When Error1 < N6 and Error3 < N7, the controller believes that the vehicle enters a controllable state and switches to Model 1 and the corresponding controller parameters, which is called GP1 mode. GP1 can be understood as an aggressive driving mode.

[0168] There are size relationships among N1, N2, N3, N4, N5, N6, and N7, all of which are obtained by empirical parameter tuning. Table 2 gives examples of N1, N2, N3, N4, N5, N6, and N7.

[0169] Table 2 Examples of N1, N2, N3, N4, N5, N6, and N7

[0170] N1 N2 N3 N4 N5 N6 N7 0.2 0.02 0.15 0.02 0.09 0.005 0.009

[0171] The parameters in Table 2 are only examples, and in actual applications, parameter tuning needs to be carried out comprehensively considering the vehicle, road surface conditions, etc. The system can select the appropriate controller mode according to the size of the real-time vehicle stability evaluation index.

[0172] After obtaining the switching strategy, the switching strategy will be input into the controller parameter adjustment module. From the multiple groups of controller weight coefficient combinations preset in this module, the corresponding weight coefficient combination will be retrieved and applied to the model predictive contour controller to complete the parameter adjustment of the controller and take effect in the controller optimization task of the next cycle. The specific implementation method is as follows: According to the high, medium, and low adhesion coefficient road surfaces defined above, configure the corresponding Gaussian process vehicle error model for the controller under each type of road surface. On the racing track where the vehicle extreme conditions are triggered, use the high-fidelity vehicle model in the vehicle dynamics simulation software to simulate the real vehicle, and carry out the parameter tuning of the controller weight coefficient combination. Debug the controller weight coefficient combination for the high, medium, and low adhesion coefficient road surfaces respectively. The parameter tuning process can use the method of reinforcement learning to make the vehicle control under each type of road surface basically maintain fast and stable driving. Define the three groups of controller weight coefficient combinations obtained from the debugging as aggressive, normal, and conservative controller driving modes to cope with the high, medium, and low adhesion coefficient road surfaces.

[0173] Based on the switchable design of multiple groups of controller weight coefficient combinations under the model prediction contour control framework, the controller weight coefficients debugged in advance ensure the reliable stability of the controller. At the same time, the switchable design increases the controller's adaptability to roads with different adhesion coefficients, which is simple and efficient.

[0174] After the vehicle error model database receives the call instruction, it inputs the corresponding vehicle error model into the probabilistic vehicle dynamics model, that is, it completes a cycle of the vehicle path tracking control algorithm and enters the cycle of the next moment. The probabilistic vehicle dynamics model generated by combining the vehicle dynamics model and the Gaussian process vehicle error model has advantages in reliability, accuracy and interpretability compared to the neural network model as the vehicle dynamics model, and the amount of prior training data required is small.

[0175] Instructions attached Figure 5 A simulation diagram is provided. The feasibility of the vehicle path tracking control method of this embodiment is proved through simulation. Figure 5 In the figure, point E is the starting point of the vehicle, and the road adhesion coefficient of path EA is set to 0.9, the road adhesion coefficient of path AB is set to 0.6, the road adhesion coefficient of path BC is set to 0.25, the road adhesion coefficient of path CD is set to 0.9, and the road adhesion coefficient of path DE is set to 0.6. The trajectory line in the figure is the trajectory of the vehicle driving three laps on the track. The colors from blue to yellow correspond to the vehicle vector speed range of 5m / s to 25m / s. It can be seen that the vehicle did not run out of control and run off the track under the above three road combination conditions with different adhesion surfaces, which proves that the vehicle path tracking control method of the present application can adapt the vehicle tracking control to complex roads.

[0176] This embodiment provides a vehicle path tracking control method that adapts to complex road conditions, so that unmanned vehicles can effectively cope with road uncertainties, and realize the adaptive control of the path tracking algorithm of unmanned vehicles on roads with varying adhesion coefficients and roads with varying curvatures, effectively improving the stability of the path tracking of smart vehicles under complex road conditions. Based on different road conditions, corresponding vehicle error models are constructed, and the vehicle's dynamic model can be adjusted according to road conditions, so that the controller is reliable and stable and has the ability to adapt to road conditions. The execution difference of the real vehicle system and the vehicle model on the same control quantity is used as feedback to obtain different switching strategies, and then different weight coefficients and error models are obtained to achieve accurate switching of strategies.

[0177] Example 2

[0178] This embodiment proposes a vehicle path tracking control system that is suitable for complex roads, which can implement a vehicle path tracking control method that is suitable for complex roads proposed in Example 1. The module diagram of the vehicle path tracking control system is shown in the attached manual. Figure 2 The specific scheme is as follows:

[0179] The vehicle path tracking control system includes:

[0180] Vehicle error model database 1 is configured with vehicle error models for different adhesion coefficients. According to the customized high, medium and low adhesion coefficient types of roads, the actual vehicle driving data and controller prediction data in the dynamic simulation environment are collected to model the Gaussian process vehicle error model for high, medium and low typical adhesion coefficient types of roads.

[0181] The probabilistic vehicle model 2 is connected to the vehicle error model database 1 and is configured with a vehicle dynamics model regarding tire sideslip, and the vehicle error model can be called from the vehicle error model database 1. The probabilistic vehicle model 2 is composed of a vehicle dynamics model considering tire sideslip and a Gaussian process vehicle error model called from the vehicle error database, and serves as a prediction model for the model prediction profile controller 3.

[0182] The model prediction profile controller 3 is connected to the probabilistic vehicle model 2 and is used to optimize the target items by discrete iteration within the prediction step length using the probabilistic vehicle model 2 as the prediction model to obtain the optimal control sequence. The optimal control sequence is generated by optimizing the hysteresis error, profile error, orientation error, boundary error, path centerline speed, acceleration and deceleration aggressiveness, and steering aggressiveness within the prediction step length taking into account the above-mentioned multi-objective optimization solutions.

[0183] The controller strategy switching module 4 is connected to the error model database and the controller parameter adjustment module 5, and is used to obtain the iterative vehicle state quantity and the actual vehicle state quantity, and solve the vehicle stability evaluation index based on the absolute value of the difference between the iterative vehicle state quantity and the actual vehicle state quantity, and obtain the switching strategy at the current moment by logically judging and classifying the stability evaluation index, and output the switching strategy to the error model database and the controller parameter adjustment module 5 respectively. The absolute value of the difference between the actual vehicle state quantity and the vehicle state quantity obtained by the prediction model is used as the original error data, and a suitable vehicle driving state error quantity is selected, such as using the vehicle position (x, y) and heading angle State quantity error. The vehicle controller mode switching strategy is the call decision of the vehicle error model in the probabilistic vehicle model 2 and the selection of the controller weight coefficient combination in the controller weight coefficient adjustment module. The original error data used is the absolute value of the difference between the actual vehicle state quantity and the iterative vehicle state quantity, which is used to define the vehicle stability evaluation index.

[0184] The controller parameter adjustment module 5 is connected to the model prediction profile controller 3 and is configured with weight coefficient combinations under different adhesion coefficient roads, and is used to output the corresponding weight coefficient combination to the model prediction profile controller 3 according to the switching strategy. The weight coefficient can be reasonably adjusted through rules according to the real-time error data returned during the vehicle driving process to achieve the controller effect of different weight coefficient combinations.

[0185] The real vehicle system 6 is connected to the model prediction profile controller 3 and the controller strategy switching module 4 respectively, and is used to receive the first control item of the optimal control sequence output by the model prediction profile controller 3, and detect the actual vehicle state quantity obtained under the control of the first control item.

[0186] The present invention provides a vehicle path tracking control method and control system that are adaptable to complex road conditions, so that unmanned vehicles can effectively cope with road uncertainties, and realize the adaptive control of the path tracking algorithm of unmanned vehicles on roads with varying adhesion coefficients and roads with varying curvatures, effectively improving the stability of the path tracking of intelligent vehicles under complex road conditions. Based on different road conditions, corresponding vehicle error models are constructed, and the dynamic model of the vehicle can be adjusted according to the road conditions, so that the controller is reliable and stable and has the ability to adapt to road conditions. The execution difference of the real vehicle system and the vehicle model on the same control quantity is used as feedback to obtain different switching strategies, and then different weight coefficients and error models are obtained to achieve accurate switching of strategies.

[0187] It should be understood by those skilled in the art that the modules or steps of the present invention described above can be implemented by a general-purpose computing device, they can be concentrated on a single computing device, or distributed on a network composed of multiple computing devices, optionally, they can be implemented by a program code executable by a computer device, so that they can be stored in a storage device and executed by the computing device, or they can be made into individual integrated circuit modules, or multiple modules or steps therein can be made into a single integrated circuit module for implementation. Thus, the present invention is not limited to any specific combination of hardware and software.

[0188] Note that the above are only preferred embodiments of the present invention and the technical principles used. Those skilled in the art will understand that the present invention is not limited to the specific embodiments herein, and that various obvious changes, readjustments and substitutions can be made by those skilled in the art without departing from the scope of protection of the present invention. Therefore, although the present invention has been described in more detail through the above embodiments, the present invention is not limited to the above embodiments, and may include more other equivalent embodiments without departing from the concept of the present invention, and the scope of the present invention is determined by the scope of the appended claims.

[0189] The above disclosure is only a few specific implementation scenarios of the present invention, but the present invention is not limited thereto, and any changes that can be conceived by those skilled in the art should fall within the protection scope of the present invention.

Claims

1. A vehicle path tracking control method adapted to complex roads, characterized in that: include: In the preset model prediction profile controller, the vehicle state quantity of the prediction step is solved by discrete iteration of the preset probability vehicle model, and the optimal control sequence at the current moment is obtained; the optimal control sequence includes predicting the control quantity at multiple moments after the current moment; Recording the vehicle state quantity after the first iteration of the probabilistic vehicle model as the iterated vehicle state quantity; inputting the first control quantity in the optimal control sequence into the real vehicle system, and observing the actual vehicle state quantity of the real vehicle system; Based on the absolute value of the difference between the iterated vehicle state quantity and the actual vehicle state quantity, the stability evaluation index of the vehicle is solved, and the switching strategy at the current moment is obtained by logically judging and classifying the stability evaluation index; A set of controller weight coefficient combinations is selected based on the switching strategy and applied to the model predictive profile controller to complete parameter adjustment of the model predictive profile controller; wherein different controller weight coefficient combinations correspond to controller modes under different adhesion coefficient types of road surfaces; Based on the switching strategy, a vehicle error model is called, and the vehicle error model is updated to the probabilistic vehicle model to complete the vehicle path tracking control at the current moment; wherein different vehicle error models are constructed based on vehicle data on roads with different adhesion coefficient types; The probabilistic vehicle model includes a vehicle error model fitted by a Gaussian process model and a vehicle dynamics model based on tire side slip considerations; Wherein, the vehicle error model E(X k ,U k ) is expressed as: Among them, GP(Z k ) represents a Gaussian process model with five input dimensions and three output dimensions, and its output data is and Among them, the position of the vehicle's center of mass is (x, y), v x 、v y are the longitudinal and lateral velocities of the vehicle’s center of mass, is the heading angular velocity; The expression of the probabilistic vehicle model is: Among them, f(X k ,U k ) represents the discretized vehicle dynamics model, Indicates that at time k, according to the current vehicle state X k And the optimized control quantity U k The vehicle state quantity at the next moment is delivered.

2. The vehicle path tracking control method according to claim 1, characterized in that: Each control variable includes a plurality of target items, wherein the target items include a hysteresis error, a contour error, a heading error, a boundary error, a vehicle vector speed, a path centerline speed, an acceleration / deceleration aggressiveness, or a steering aggressiveness; The controller weight coefficient combination is composed of the weight coefficients of each target item in the control quantity, and the combination of weight coefficients of different target items obtains different controller weight coefficient combinations.

3. The vehicle path tracking control method according to claim 2, characterized in that: In the model prediction contour controller, the road centerline is used as the reference path, and the path points are discretized into a set {s0,s1,…,s k ,…s max }; At each moment, the path point s k There are corresponding coordinates [x c (s k ),y c (s k )], heading angle Track radius R c (s k ), road width W; By introducing a centerline velocity v c (s k ) Discretize the path s k+1 =s k +v c (s k )t, t is the time of each prediction step to ensure the vehicle's driving progress in the direction of the road centerline.

4. The vehicle path tracking control method according to claim 2, characterized in that: The hysteresis error e is defined based on the geometric position of the vehicle and the trajectory. l , contour error e c , orientation error e o , boundary error e off The spatial pose error including: Error Optimization Term of Model Predictive Contour Controller for: Among them, q c Indicates the contour error e c The weight coefficient, q l Represents the hysteresis error e l The weight coefficient, q o Represents the orientation error e o The weight coefficient, q off Represents the boundary error e off The weight coefficient of Represents the vehicle state at time k.

5. The vehicle path tracking control method according to claim 4, characterized in that: Involving vehicle vector speed and the path centerline velocity v c Speed ​​optimizations The expression is as follows: Among them, q v Represents the vehicle vector speed as the weight coefficient, q ce A weight factor representing the speed of the centerline of the path.

6. The vehicle path tracking control method according to claim 5, characterized in that: Dynamic optimization terms for acceleration aggressiveness τ, 0<τ<1, deceleration aggressiveness -τ, -1<-τ<0, and steering aggressiveness δ The expression is as follows: Among them, q τ The weight coefficient indicating the degree of acceleration, q -τ The weight coefficient indicating the degree of deceleration positivity, q δ The weight coefficient indicating the degree of positivity of the turn.

7. The vehicle path tracking control method according to claim 1, characterized in that: The process of obtaining the switching strategy includes: The absolute value of the difference between the iterated vehicle state quantity and the actual vehicle state quantity is used as the original error expression; the vehicle driving state error quantity is selected; Based on a preset state quantity screening matrix, extracting a true error expression about the vehicle driving state error quantity from the original error expression; Constructing a plurality of stability evaluation indicators based on the true error expression; Taking the stability evaluation index as input, the corresponding switching strategy involving the model predictive profile controller mode is selected.

8. The vehicle path tracking control method according to claim 7, characterized in that: Take the control quantity U at time k k , input into the probabilistic vehicle model to obtain the iterative vehicle state at time k+1 Control quantity U k The actual vehicle state quantity input into the real vehicle system and observed at time k+1 Get the true error expression at time k+1 The expression is: Among them, the state quantity screening matrix B d =[I k×3 ,0 3×3 ].

9. The vehicle path tracking control method according to claim 8, characterized in that: Based on the true error expression, three stability evaluation indicators are constructed, namely Error1, Error2, and Error3; Error1 is The second norm of As the vehicle moves, the Error1 sequence can be obtained, K = [k, k-1, k-2, k-3, ..., kN], N is based on the length of the historical Error1 sequence used at time k, assuming that the linear regression function h w (k) = K T w k , in order to make the regression function more consistent with the true value, the least squares fitting loss function J is defined as w k It is the derivative of the linear regression function based on the past N+1 Error1 error points at each k moment, and Error2 is obtained k =w k ; Error3 is the smoothing of Error1. Error3 is obtained by averaging the past N Error1 error points at each k moment.

10. The vehicle path tracking control method according to claim 1, characterized in that: Define N types of adhesion coefficient type road surfaces based on the adhesion coefficient; A vehicle error model suitable for each type of adhesion coefficient road surface is obtained by training the data under each type of adhesion coefficient road surface; N groups of controller weight coefficient combinations are set, and each group of controller weight coefficient combination corresponds to a type of adhesion coefficient road surface.

11. A vehicle path tracking control system adapted to complex roads, characterized in that: Used to implement the vehicle path tracking control method according to any one of claims 1 to 10; The vehicle path tracking control system includes: Vehicle error model database, which is configured with vehicle error models under different adhesion coefficients; A probabilistic vehicle model, connected to a vehicle error model database, configured with a vehicle dynamics model related to tire sideslip, and capable of calling a vehicle error model from the vehicle error model database; A model prediction profile controller is connected to the probabilistic vehicle model and is used to perform multi-objective optimization solutions for each target item through discrete iteration within a prediction step using the probabilistic vehicle model as a prediction model to obtain an optimized control sequence; a controller strategy switching module, connected to the error model database and the controller parameter adjustment module, for obtaining an iterative vehicle state quantity and an actual vehicle state quantity, solving a stability evaluation index of the vehicle based on an absolute value of a difference between the iterative vehicle state quantity and the actual vehicle state quantity, and obtaining a switching strategy at a current moment by performing logical judgment and classification on the stability evaluation index, and outputting the switching strategy to the error model database and the controller parameter adjustment module respectively; The controller parameter adjustment module is connected to the model prediction profile controller and is configured with weight coefficient combinations under different adhesion coefficient road surfaces, and is used to output corresponding weight coefficient combinations to the model prediction profile controller according to a switching strategy.

12. The vehicle path tracking control system according to claim 11, characterized in that: Also includes real vehicle systems; The real vehicle system is connected to the model prediction profile controller and the controller strategy switching module respectively, and is used to receive the first control item of the optimal control sequence output by the model prediction profile controller, and detect the actual vehicle state quantity obtained under the control of the first control item.

Images