Curve lane keeping method and system based on adaptive model predictive control

Abstract

The application discloses a kind of based on adaptive model predictive control curve lane keeping method and system, first according to the curve working condition information of vehicle travel determines the safe vehicle speed of vehicle over curve;Then the optimal time domain parameters of MPC controller under different road information are matched using genetic algorithm;Again, neural network MPC controller is constructed, so that the safe vehicle speed of vehicle over curve and the time domain parameters of MPC controller are adaptively adjusted according to the change of curve working condition;Finally, according to the trained neural network, the vehicle longitudinal controller and vehicle lateral controller are integrated, to realize the control of curve lane keeping.The application effectively solves the problem of vehicle sideslip or rollover caused by high speed over curve, and the poor lane keeping accuracy caused by fixed MPC time domain parameters, improves the accuracy of unmanned vehicle curve lane keeping, ensures the stability of vehicle driving, and has practical significance for promoting the rapid development of unmanned technology.

Description

Technical Field

[0001] This invention belongs to the technical field of autonomous driving, and mainly relates to a method and system for lane keeping in curves based on adaptive model predictive control. Background Technology

[0002] Lane keeping is a fundamental task of autonomous driving. Its purpose is to control the wheel angles so that the vehicle follows the centerline of the desired path, which falls under the category of lateral control of the vehicle.

[0003] A survey of lane-keeping control strategies both domestically and internationally revealed that commonly used controllers for lane keeping include PID (Proportional-Integral-Derivative) controllers, SMC (Slippery Module Control) controllers, LQR (Linear Quadratic Regulation) controllers, Stanley controllers, pure tracking controllers, and MPC (Model Predictive Control) controllers. Among these, the basic idea of ​​the MPC controller is to use an existing model, the current state of the system, and future control variables to predict the future output of the system. This is achieved by continuously solving a constrained optimization problem to minimize control error. Therefore, MPC is the most commonly used controller in vehicle control.

[0004] Existing lane keeping methods based on MPC controllers all employ constant time-domain parameters and assume that the vehicle maintains a constant speed, road adhesion conditions are good, and road curvature is small. However, in real-world environments, road curvature is time-varying, and road adhesion conditions change due to factors such as weather. When a vehicle corners at high speeds, dangerous accidents such as skidding or rollovers can occur. When the time-domain control parameters of the MPC controller are constant, the lane keeping performance may be good on a certain road segment, but when the vehicle's driving environment changes drastically, lane keeping will deviate, and may even affect the vehicle's safety and stability. The goal is to find a method that allows the vehicle speed and the time-domain control parameters of the MPC controller to adaptively adjust according to environmental changes, ensuring vehicle driving safety while improving the effectiveness of lane keeping in curves. Therefore, this patent proposes a curve lane keeping method and system based on adaptive model predictive control. Summary of the Invention

[0005] This invention addresses the problem of insufficient accuracy and stability in lane keeping for autonomous driving systems by providing an adaptive model predictive control (EMC) method and system for cornering lane keeping. First, the safe cornering speed is determined based on the vehicle's cornering conditions. Then, a genetic algorithm is used to match the optimal time-domain parameters of the MPC controller under different road conditions. Finally, a neural network MPC controller is constructed, enabling the safe cornering speed and the MPC controller's time-domain parameters to adaptively adjust according to changes in cornering conditions. The neural network model has the following inputs: road curvature radius and road adhesion coefficient; output: safe cornering speed and predicted time-domain parameters N.p and control time domain N c A two-input, three-output neural network is used; finally, based on the trained neural network, the vehicle's longitudinal controller and lateral controller are integrated to achieve lane keeping control during curves. This invention enables adaptive adjustment of vehicle speed and controller parameters during curve lane keeping, effectively solving the problems of vehicle sideslip or rollover caused by high cornering speeds and poor lane keeping accuracy caused by fixed MPC time-domain parameters. It improves the accuracy of lane keeping in curves for autonomous vehicles, ensures vehicle stability, and has practical significance for promoting the rapid development of autonomous driving technology.

[0006] To achieve the above objectives, the technical solution adopted by this invention is: an adaptive model predictive control method for lane keeping on curves, comprising the following steps:

[0007] S1: Determine the safe speed for the vehicle to corner based on the cornering conditions information of the vehicle; when the vehicle is cornering, only the front wheels are considered to participate in the vehicle steering, and the left front wheel and the right front wheel have the same rotation angle.

[0008] S2: Use a genetic algorithm to match the optimal time-domain parameters of the MPC controller under different road information. The optimal time-domain parameters are related to the safe cornering speed, road curvature radius and road adhesion coefficient in step S1.

[0009] S3: Construct a neural network MPC controller so that the vehicle's safe cornering speed obtained in step S1 and the time-domain parameters of the MPC controller determined in step S2 are adaptively adjusted according to changes in cornering conditions; the neural network model is as follows: the input is the road curvature radius and the road adhesion coefficient, and the output is the safe cornering speed and the predicted time-domain N. p and control time domain N c A two-input, three-output neural network;

[0010] S4: Based on the neural network trained in step S3, the vehicle longitudinal controller and the vehicle lateral controller are integrated to achieve lane keeping control on curves.

[0011] As an improvement of the present invention, in step S1, the safe speed for the vehicle to corner is the maximum critical speed at which the vehicle will not skid or roll over:

[0012]

[0013] Where, min{·} represents taking the minimum value; μ represents the adhesion coefficient between the tire and the road surface; g represents the gravitational acceleration; θ represents the road lateral slope angle; R represents the road curvature radius; B represents the wheelbase of the vehicle; H g This indicates the height of the vehicle's center of gravity.

[0014] As an improvement of the present invention, the safe speed for the vehicle to corner in step S1 is the maximum critical speed within the legal speed limit at which the vehicle will not skid or roll over, that is:

[0015] V safe =min{kV sr_maX V law}

[0016] Among them, V safe represents the safe speed for a vehicle to corner; k represents an environmental factor. On sunny days, the environmental factor k ranges from [0.32, 0.38]; on cloudy days, the environmental factor k ranges from [0.24, 0.32]; at night, the environmental factor k ranges from [0.20, 0.27]; and during rainy, snowy, foggy, or hazy weather, the environmental factor k ranges from [0.12, 0.18].

[0017] As another improvement of the present invention, step S2, which involves the selection of the optimal time-domain parameters for MPC based on a genetic algorithm, specifically includes:

[0018] S21: Initialize the MPC time-domain parameters and generate a population of N individuals;

[0019] S22: Use the ROUND function to round down the initialized population and assign the result to the prediction time domain N. p and control time domain N c ;

[0020] S23: Input the time-domain parameter values ​​from step S22 into the MPC controller to perform circular path tracking based on the corresponding safe cornering speed, and calculate the fitness function of each individual. The fitness function for the lane-keeping task is:

[0021]

[0022] Among them, J s The evaluation index representing lane-keeping stability; J e Evaluation metrics indicating lane-keeping accuracy; w s and w e These represent the weights used to measure lane-keeping stability and accuracy, respectively.

[0023] S24: Select and cross over individuals with higher scores; to prevent getting trapped in local optima, a small number of individuals also need to be mutated to generate new individuals;

[0024] S25: Repeat steps S22-S24 until the lane keeping performance requirements are met or the number of iterations is completed.

[0025] As another improvement of the present invention, in step S23, the evaluation index J of lane keeping stability sSpecifically:

[0026]

[0027] Among them, J s1 J is the evaluation index for yaw rate. s2 For lateral acceleration evaluation; w s1 and w s2 These represent the weights used to measure yaw rate and lateral acceleration, respectively.

[0028] in:

[0029]

[0030]

[0031] in, and These represent the vehicle yaw rate and yaw rate threshold, respectively; a y and t1 and t2 represent the longitudinal acceleration of the vehicle and the longitudinal acceleration threshold, respectively; t1 and t2 represent the start and end times of the experiment, respectively.

[0032] The lane keeping accuracy evaluation index J e Specifically:

[0033]

[0034] Among them, J e1 The evaluation index for lane keeping lateral deviation; J e2 The yaw rate deviation is used as an evaluation index; w e1 and w e2 These represent the weights for measuring lane keeping lateral deviation and yaw rate deviation, respectively.

[0035] in:

[0036]

[0037]

[0038] Among them, ye and These represent the lane-keeping lateral deviation and its lateral deviation threshold, respectively; β and t1 and t2 represent the vehicle yaw angle deviation and the yaw angle deviation threshold, respectively; t1 and t2 represent the start and end times of the experiment, respectively.

[0039] As another improvement of the present invention, in the neural network of step S3, there is the following relationship between the road curvature radius, road adhesion coefficient, safe cornering speed, prediction time domain, and control time domain:

[0040]

[0041] Among them, V safe Indicates the safe speed for cornering; N p N represents the prediction time domain; c denoted by ; R represents the road curvature radius; μ represents the road adhesion coefficient; f1(·) and f2(·) represent different nonlinear functions with respect to R and μ, respectively.

[0042] As another improvement of the present invention, the neural network dataset in step S3 includes the road curvature radius and the road adhesion coefficient. The dataset labels include the safe speed for passing the curve, the prediction time domain, and the control time domain. A circular path tracking experiment is conducted at the safe speed for passing the curve to determine the optimal time domain parameters under different combinations of vehicle speed, road curvature radius, and road adhesion coefficient, and the neural network is trained.

[0043] To achieve the above objectives, the present invention also adopts the following technical solution: an adaptive model predictive control curve lane keeping system, comprising a computer program, wherein the computer program, when executed by a processor, implements the steps of any of the methods described above.

[0044] Compared with existing technologies, this invention introduces neural networks into the traditional control field, enabling adaptive adjustment of vehicle speed and model predictive control parameters under curve conditions. This improves the accuracy and safety stability of lane keeping, ensuring the safety of both vehicles and pedestrians. This invention provides a method and system for curve lane keeping based on adaptive model predictive control. Addressing the issue of skidding or rollover when vehicles curve at high speeds, this method designs a safe cornering vehicle based on road information. It also addresses the problem of lane keeping deviation caused by fixed-time-domain MPC controllers being unsuitable for complex road conditions. By leveraging the optimization capabilities of genetic algorithms to match optimal time-domain parameters under different road conditions, a neural network MPC controller is constructed. This allows for adaptive adjustment of vehicle speed and the time-domain parameters of the MPC controller according to the vehicle's driving conditions, thereby improving the accuracy of lane keeping in autonomous vehicles and ensuring vehicle stability. This has practical significance for promoting the rapid development of autonomous driving technology. Attached Figure Description

[0045] Figure 1 This is a flowchart of the steps of the method of the present invention;

[0046] Figure 2 This is a flowchart of step S2 of the method of the present invention, which is the optimization of the optimal time-domain parameters of MPC based on the genetic algorithm.

[0047] Figure 3 This invention relates to a time-domain parameter adaptive MPC control graph based on a neural network.

[0048] Figure 4This is a comparison diagram of the lane-keeping effects of the method of the present invention and existing methods. Detailed Implementation

[0049] The present invention will be further illustrated below with reference to the accompanying drawings and specific embodiments. It should be understood that the following specific embodiments are for illustrative purposes only and are not intended to limit the scope of the invention.

[0050] Example 1

[0051] Adaptive model predictive control-based lane keeping methods for curves, such as Figure 1 As shown, it includes the following steps:

[0052] S1: Determine the safe speed for the vehicle to corner based on the cornering conditions information of the vehicle; when the vehicle is cornering, only the front wheels are considered to participate in the vehicle steering, and the left front wheel and the right front wheel have the same rotation angle.

[0053] The vehicle and road environment settings for this method are as follows:

[0054] ① Only consider the front wheels participating in vehicle steering, and assume that the left and right front wheels have the same rotation angle.

[0055] The vehicle is a monorail model:

[0056] ②Assume the road surface is flat, and do not consider the vertical, pitch, and lateral motion of the vehicle;

[0057] ③ The coupling relationship of the vehicle suspension is not considered, and the vehicle is assumed to be rigid;

[0058] ④ The self-aligning torque generated by the tires when the wheels turn is not considered;

[0059] ⑤ The influence of the longitudinal and lateral coupling relationship of tire forces is not considered;

[0060] ⑥ Load transfer between the front and rear axles of the vehicle is not considered;

[0061] ⑦ The effects of aerodynamics on the vehicle's lateral and longitudinal movements are not considered.

[0062] S11: Calculate the vehicle's critical sideslip speed

[0063] When a vehicle is cornering, if the lateral adhesion force of the wheels exceeds the lateral friction force between the road surface and the tires, the wheels will sideslip. Therefore, the torque balance equation for the critical state of vehicle sideslip is:

[0064] F yl +F yr =(F zl +F zr μ (1.1)

[0065] In the formula, F yland F yr F represents the lateral force exerted on the inner and outer tires of the vehicle, respectively. zl and F zr These represent the vertical forces acting on the inner and outer tires of the vehicle, respectively, while μ represents the coefficient of adhesion between the tire and the road surface.

[0066] At this point, the critical lateral acceleration at which the vehicle sideslips is:

[0067]

[0068] In the formula, a ys The value represents the critical lateral acceleration at which a vehicle skids, g represents the acceleration due to gravity, and θ represents the lateral slope angle of the road.

[0069] The relationship between lateral acceleration, vehicle critical sideslip velocity, and road curvature is shown in the following equation:

[0070] a yS =v s_max 2 R (1.3)

[0071] In the formula, v s_max R represents the critical speed at which a vehicle will roll over. R represents the road curvature radius. Considering that the road curvature radius of a straight road is ∞, and that the lane is approximately straight within the visual range when the road curvature radius exceeds 1000m, the range of the road curvature radius is set to (0m, 1000m).

[0072] Combining equations (1.2) and (1.3), the critical speed at which the vehicle begins to skid is obtained as follows:

[0073]

[0074] S12: Calculate the critical rollover speed of the vehicle

[0075] When a vehicle is cornering, excessive speed can cause it to lose lateral stability, resulting in excessive centrifugal force and the inner tires lifting off the ground. At this point, the vehicle is in a critical rollover state. Therefore, the moment balance equation for the critical rollover state is:

[0076]

[0077] In the formula, m represents the mass of the vehicle, and a yr B represents the critical acceleration at which a vehicle rolls over, and H represents the wheelbase of the vehicle. g θ represents the vehicle's center of gravity height, and θ represents the road's lateral slope angle.

[0078] The critical acceleration at which a vehicle overturns is:

[0079]

[0080] The relationship between lateral acceleration, vehicle critical rollover velocity, and road curvature is shown in the following formula:

[0081] a yr =v r_max 2 R (1.7)

[0082] In the formula, v r_max R represents the critical speed at which a vehicle will roll over, and R represents the radius of curvature of the road.

[0083] Combining equations (1.6) and (1.7), the critical speed at which the vehicle rolls over is obtained as follows:

[0084]

[0085] S13: Safe speed selection for cornering:

[0086] Therefore, regarding the safe speed for a vehicle to corner, combining equations (1.4) and (1.8), the maximum critical speed at which the vehicle will not skid or roll over can be obtained as follows:

[0087]

[0088] In the formula, min{·} represents taking the minimum value.

[0089] Furthermore, according to the "Road Traffic Safety Law of the People's Republic of China," motor vehicles must comply with the speed limits of the corresponding roads when driving. The legally limited speed for the relevant roads is V. law Meanwhile, environmental factors have a significant impact on vehicle safety. To better reflect vehicle driving characteristics under different environments, an environmental factor k is introduced. Therefore, the relationship between the safe cornering speed, the maximum critical cornering stability speed, and the legally limited speed is as follows:

[0090] V safe =min{kV sr_max V law (1.10)

[0091] In the formula, V safe denoted by , where represents the safe speed for a vehicle to corner, and k represents the environmental factor. On sunny days, the environmental factor ranges from [0.32, 0.38]; on cloudy days, it ranges from [0.24, 0.32]; at night, it ranges from [0.20, 0.27]; and during rain, snow, fog, or haze, it ranges from [0.12, 0.18].

[0092] In summary, the factors affecting vehicle safety when cornering include wheelbase, road curvature radius, road surface adhesion conditions, lateral slope angle, and road speed limit. Since the vehicle wheelbase and road speed limit are known, and this method assumes that the vehicle is traveling on a flat road surface and does not consider the vehicle's vertical, pitch, and roll motions, it only considers the influence of the road curvature radius and road surface adhesion coefficient on the vehicle's safe cornering speed.

[0093] S2: Use a genetic algorithm to match the optimal time-domain parameters of the MPC controller under different road information. The optimal time-domain parameters are related to the safe cornering speed, road curvature radius and road adhesion coefficient in step S1.

[0094] When using an MPC controller for lane keeping, the effectiveness of lane keeping is highly dependent on the time-domain parameters of the MPC controller. Prediction time domain N p The degree to which the system predicts the future, N p When the value is large, the lane keeping deviation is significant, resulting in poor timeliness; N p When the pressure is too low, the wheels may not be able to steer in time, leading to loss of vehicle control. c When the value is large, the system stability and real-time performance are poor. c When N is small, the system control accuracy is poor. Therefore, N p and N c The value of not only affects the lane keeping performance of the controller, but also the real-time performance of the controller in solving the lane keeping quadratic programming problem.

[0095] In real-world scenarios, assuming we ignore subjective human factors and vehicle performance limitations, the steering wheel angle of a vehicle changes with vehicle speed, road curvature radius, and road adhesion conditions. Therefore, vehicle speed, road curvature radius, and road adhesion conditions are all crucial factors affecting lane keeping performance during curves. By calculating the safe cornering speed using road condition information, and considering the road curvature radius and road adhesion coefficient, the optimal MPC time-domain control parameters can be selected for different curve conditions.

[0096] The optimal time-domain parameters of the MPC controller under different combinations of vehicle speed, road curvature radius, and road adhesion coefficient are selected using the optimization capability of a genetic algorithm. Genetic algorithms have advantages such as strong global search capability, fast convergence speed, and high search efficiency, and can obtain the globally optimal solution. Therefore, a genetic algorithm is used to optimize the time-domain parameters of the MPC controller. After multiple iterations, the MPC time-domain parameters with the best lane-keeping performance are selected, such as... Figure 2 As shown.

[0097] The process of selecting the optimal time-domain parameters for MPC based on genetic algorithm is as follows:

[0098] S21: Initialize the MPC time-domain parameters and generate a population of N individuals;

[0099] S22: Since the time-domain parameters of the MPC controller must be integers, the ROUND function is used to round down the initialized population and assign the value to the prediction time domain N. p and control time domain N c ;

[0100] S23: Input the time-domain parameter values ​​from step S22 into the MPC controller to perform circular path tracking based on the corresponding safe cornering speed, and calculate the fitness function of each individual. The fitness function for the lane-keeping task is:

[0101]

[0102] In the formula, J s J is an evaluation index representing lane-keeping stability. e The evaluation index representing lane keeping accuracy, w s and w e These represent the weights used to measure lane-keeping stability and accuracy, respectively.

[0103] in:

[0104]

[0105] In the formula, J s1 J is the evaluation index for yaw rate. s2 w is a lateral acceleration evaluation index. s1 and w s2 These represent the weights used to measure yaw rate and lateral acceleration, respectively.

[0106] in:

[0107]

[0108]

[0109] In the formula, and These represent the vehicle yaw rate and yaw rate threshold, respectively. y and t1 and t2 represent the longitudinal acceleration of the vehicle and the longitudinal acceleration threshold, respectively, and t1 and t2 represent the start and end times of the experiment, respectively.

[0110]

[0111] In the formula, J e1 J is the evaluation index for lane keeping lateral deviation. e2 w is the evaluation index for yaw angle deviation. e1and w e2 These represent the weights for measuring lane keeping lateral deviation and yaw rate deviation, respectively.

[0112] in:

[0113]

[0114]

[0115] In the formula, ye and These represent the lane-keeping lateral deviation and its lateral deviation threshold, β and β, respectively. t1 and t2 represent the vehicle yaw angle deviation and the yaw angle deviation threshold, respectively, and t1 and t2 represent the start and end times of the experiment, respectively.

[0116] S24: Select and crossover individuals with higher scores. Furthermore, to prevent getting trapped in local optima, a small number of individuals need to be mutated to generate new individuals.

[0117] S25: Repeat steps S22-S24 until the lane keeping performance requirements are met or the number of iterations is completed.

[0118] S3: Construct a neural network MPC controller so that the vehicle's safe cornering speed obtained in step S1 and the time-domain parameters of the MPC controller determined in step S2 are adaptively adjusted according to changes in cornering conditions; the neural network model is as follows: the input is the road curvature radius and the road adhesion coefficient, and the output is the safe cornering speed and the predicted time-domain N. p and control time domain N c A two-input, three-output neural network;

[0119] In step S2, the optimal MPC time-domain parameters for different curve conditions have been selected. However, in real-world environments, the road curvature radius and adhesion conditions are time-varying. Finding the optimal time-domain parameters for different curve conditions requires a certain amount of iterative process, compromising real-time performance and jeopardizing vehicle safety. Since the optimal time-domain parameters of the MPC controller are related to the safe cornering speed, road curvature radius, and road adhesion coefficient, the following relationship exists between the road curvature radius, road adhesion coefficient, safe cornering speed, prediction time domain, and control time domain:

[0120]

[0121] In the formula, V safe N represents the safe speed for cornering. p N represents the prediction time domain. c Let f1(·) represent the control time domain, R represent the road curvature radius, μ represent the road adhesion coefficient, and f1(·) and f2(·) represent different nonlinear functions with respect to R and μ, respectively.

[0122] Furthermore, according to S13, the safe speed for cornering is also related to the road curvature radius and the road adhesion coefficient as follows:

[0123] V safe =f3(R, μ) (3.2)

[0124] In the formula, f3(·) represents a nonlinear function with respect to R and μ.

[0125] Combining equations (3.1) and (3.2), we obtain:

[0126]

[0127] Since f1(·) and f3(·), f2(·) and f3(·) are all nonlinear functions of R and μ, equation (3.3) can be rewritten as:

[0128]

[0129] In the formula, f4(·) and f5(·) represent different nonlinear functions with respect to R and μ, respectively.

[0130] As shown in the above equation, there are nonlinear relationships between the safe cornering speed, the prediction time domain, and the control time domain, and the road curvature radius and road adhesion coefficient. Considering that neural networks have strong nonlinear fitting ability and self-learning ability, and have a simple structure, this method constructs a neural network with the road curvature radius and road adhesion coefficient as inputs, and the safe cornering speed and prediction time domain N as outputs. p and control time domain N c A two-input, three-output neural network is used. Before the controller runs, the neural network is trained to establish the relationship between different combinations of road curvature radii, road adhesion coefficients, safe cornering speeds, prediction time domain, and control time domain. The time-domain parameter adaptive MPC control diagram based on the neural network is shown below. Figure 3 As shown, after the neural network is trained, the road curvature radius and road adhesion coefficient are obtained based on the reference path. The road curvature radius and road adhesion coefficient are then input into the neural network to obtain the time-domain control parameters N of the MPC controller. p and N c And the safe cornering speed V for lane keeping safe This enables adaptive adjustment of vehicle speed and MPC time-domain control parameters under curve conditions.

[0131] Neural Network Dataset Collection: Vehicle driving conditions with different combinations of road curvature radii and adhesion coefficients were set up. The autonomous vehicle used the genetic algorithm described in step S2 to conduct circular path tracking experiments at a safe cornering speed to determine the optimal time-domain parameters under different combinations of vehicle speed, road curvature radius, and road adhesion coefficient. A dataset was then constructed to train the neural network. The dataset includes the following information:

[0132]

[0133] S4: Based on the neural network trained in step S3, the vehicle longitudinal controller and the vehicle lateral controller are integrated to achieve lane keeping control on curves.

[0134] This method targets autonomous vehicles operating on curves. Step S1 designs a safe cornering speed to ensure vehicle stability. However, existing lane-keeping systems only control wheel angles to keep the vehicle following the centerline of the desired path, neglecting the influence of vehicle speed and focusing solely on lateral control. Therefore, this method necessitates research into longitudinal speed control. Since both lateral and longitudinal control performance are affected by road curvature radius and adhesion conditions, the longitudinal controller also employs the time-domain parameter adaptive MPC controller designed in step S3, with the time-domain parameters set identically to those of the lane-keeping lateral controller. The inputs to the MPC longitudinal controller are the desired vehicle speed and current acceleration, while the output is the desired acceleration.

[0135] Both the lateral and longitudinal control performance of a vehicle are affected by road environmental factors such as road curvature radius and road adhesion conditions, and there is a certain degree of coupling between them. Therefore, the lateral MPC controller designed in step S3 is used for lane-keeping lateral control. The inputs of the lateral MPC controller are reference position error, reference yaw angle error, longitudinal speed, lateral speed, and yaw rate, and the output is wheel steering angle. The longitudinal MPC controller is used to independently control the speed. The time-domain parameters of the longitudinal controller are set the same as those of the lateral MPC controller, realizing coordinated longitudinal and lateral control of the vehicle to ensure lane keeping on curves.

[0136] Example 2

[0137] The adaptive model predictive control-based lane keeping method for curves includes the following steps:

[0138] S1: The weather is clear. A vehicle is traveling on a curve with a speed limit of 30 km / h. The radius of curvature of the road is R = 100 m, the coefficient of adhesion is μ = 0.8, the lateral slope angle is θ = 0°, the wheelbase is B = 1460 mm, and the vehicle's center of gravity height is H. g=540mm, select the safe cornering speed for the vehicle:

[0139] S11: Calculate the vehicle's critical sideslip speed:

[0140]

[0141] S12: Calculate the critical rollover speed of the vehicle:

[0142]

[0143] S13: Safe speed for vehicles cornering:

[0144] Taking environmental factor k = 0.35, the safe speed for the vehicle when cornering is:

[0145]

[0146] S2: The selection of optimal time-domain parameters for MPC is influenced by vehicle speed, road curvature radius, and road adhesion coefficient. Based on the optimization capability of the genetic algorithm, optimal time-domain parameters are selected for different combinations of road curvature radius and road adhesion coefficient.

[0147] When the radius of curvature of the circular path is R = 700m and the road adhesion coefficient μ = 0.8, the calculated safe vehicle speed for cornering is 90.68km / h. Experiments were conducted under these conditions to select the optimal MPC time-domain parameters. The initial population size of the genetic algorithm was set to 100, the number of iterations to 50, the crossover probability to 0.5, the mutation probability to 0.1, and the prediction time domain N. p The value range is [1, 40], and the control time domain N c The value range is [1, 20]. Choose w. s =w e =w s1 =w s2 =w e1 =w e2 =0.5, the algorithm stabilizes after 12 iterations, and the optimal time-domain parameter combination is [26, 11].

[0148] S3: Design a time-domain parameter adaptive MPC controller based on neural networks:

[0149] The input is the road curvature radius and the road adhesion coefficient, and the output is the safe cornering speed and the prediction time domain N. p and control time domain N c A two-input, three-output neural network was constructed. The neural network was set to have one hidden layer, 10 nodes, 50 iterations (Epoch = 50), and a learning rate (lr = 0.01). 80% of the dataset was used as the training set, and 20% of the dataset was used as the test set.

[0150] Neural network dataset collection:

[0151] Using 0.1 as the interval for road adhesion coefficient, the range of road adhesion coefficient is set to 0.1–0.9; using 10m as the interval for road curvature radius, the range of road curvature radius is set to 10–1000m; and the prediction time domain N is set. p The value range is [1, 40], and the control time domain N is set. c The value range of is [1, 20], and satisfies N p >N c A circular path tracking experiment was conducted using the genetic algorithm described in S22 to determine the optimal time-domain parameters under different combinations of road curvature radii and road adhesion coefficients. A total of 900 experiments were performed. After the experiments, a dataset was built to train the neural network. Partial datasets are shown in the table below:

[0152]

[0153] S4: Based on the neural network trained in step S3, the vehicle longitudinal controller and the vehicle lateral controller are integrated to achieve lane keeping control on curves.

[0154] After designing the vehicle's longitudinal and lateral cooperative control system, performance verification was conducted. The vehicle was tested for lane keeping in curves with time-varying road curvature radius and road adhesion coefficient. The lane keeping performance of this method was compared with existing methods. The comparison results are as follows: Figure 4 As shown, Figure 4 As can be seen, the adaptive model predictive control-based lane keeping method for cornering proposed in this invention has a significantly improved control effect compared with the traditional model predictive control method, and the lane keeping accuracy is better. In addition, the safe cornering speed proposed in this invention can ensure the stability of lane keeping.

[0155] In summary, the adaptive model predictive control-based lane keeping method for curves proposed in this invention enables adaptive adjustment of vehicle speed and controller parameters during curve lane keeping. This effectively solves the problems of vehicle sideslip or rollover caused by high cornering speeds and poor lane keeping accuracy caused by fixed MPC time-domain parameters, thereby improving the accuracy of lane keeping for autonomous vehicles and ensuring vehicle stability.

[0156] It should be noted that the above content merely illustrates the technical concept of the present invention and should not be construed as limiting the scope of protection of the present invention. For those skilled in the art, various improvements and modifications can be made without departing from the principle of the present invention, and all such improvements and modifications fall within the scope of protection of the claims of the present invention.

Claims

1. A curve lane keeping method based on adaptive model predictive control, characterized in that... The steps include the following: S1: Determine the safe speed for the vehicle to corner based on the cornering conditions information of the vehicle; when the vehicle is cornering, only the front wheels are considered to participate in the vehicle steering, and the left front wheel and the right front wheel have the same rotation angle. S2: Use a genetic algorithm to match the optimal time-domain parameters of the MPC controller under different road information. The optimal time-domain parameters are related to the safe cornering speed, road curvature radius and road adhesion coefficient in step S1. S21: Initialize the MPC time-domain parameters and generate a population of N individuals; S22: Use the ROUND function to round down the initialized population and assign the result to the prediction time domain. and control time domain ; S23: Input the time-domain parameter values ​​from step S22 into the MPC controller to perform circular path tracking based on the corresponding safe cornering speed, and calculate the fitness function of each individual. The fitness function for the lane-keeping task is: ； in, Evaluation metrics indicating lane-keeping stability; Evaluation indicators representing lane keeping accuracy; and These represent the weights used to measure lane-keeping stability and accuracy, respectively. S24: Select and cross over individuals with higher scores; to prevent getting trapped in local optima, a small number of individuals also need to be mutated to generate new individuals; S25: Repeat steps S22-S24 until the lane keeping performance requirements are met or the number of iterations is completed; S3: Construct a neural network MPC controller so that the vehicle's safe cornering speed obtained in step S1 and the time-domain parameters of the MPC controller determined in step S2 are adaptively adjusted according to changes in cornering conditions; the neural network model has the following inputs: road curvature radius and road adhesion coefficient, and outputs: safe cornering speed and predicted time-domain parameters. and control time domain A two-input, three-output neural network; S4: Based on the neural network trained in step S3, the vehicle longitudinal controller and the vehicle lateral controller are integrated to achieve lane keeping control on curves.

2. The adaptive model predictive control-based lane keeping method for curves as described in claim 1, characterized in that: In step S1, the safe speed for the vehicle to corner is the maximum critical speed at which the vehicle will not skid or roll over. ； in, This indicates taking the minimum value; This indicates the coefficient of adhesion between the tire and the road surface; Represents gravitational acceleration; Indicates the lateral slope angle of the road; Indicates the radius of curvature of the road; Indicates the wheelbase of the car; This indicates the height of the vehicle's center of gravity.

3. The adaptive model predictive control-based lane keeping method for curves as described in claim 2, characterized in that: In step S1, the safe speed for the vehicle to corner is the maximum critical speed within the legal speed limit at which the vehicle will not skid or roll over, i.e.: ； in, Indicates the safe speed for a vehicle when cornering; This represents environmental factors; on a sunny day, the environmental factors... The range of values ​​is On cloudy days, environmental factors The range of values ​​is At night, environmental factors The range of values ​​is During rainy, snowy, foggy, or hazy weather, environmental factors The range of values ​​is .

4. The adaptive model predictive control-based lane keeping method for curves as described in claim 3, characterized in that: In step S23, the evaluation index of lane keeping stability Specifically: ； in, The yaw rate is used as an evaluation index. It is used as an evaluation index for lateral acceleration; and These represent the weights used to measure yaw rate and lateral acceleration, respectively. in: ； ； in, and These represent the vehicle's yaw rate and yaw rate threshold, respectively. and These represent the vehicle's longitudinal acceleration and longitudinal acceleration threshold, respectively. and These represent the start and end times of the experiment, respectively. The evaluation index of lane keeping accuracy Specifically: ； in, The evaluation index for lane keeping lateral deviation; As an evaluation index for yaw angle deviation; and These represent the weights for measuring lane keeping lateral deviation and yaw rate deviation, respectively. in: ； ； in, and These represent the lane keeping lateral deviation and its lateral deviation threshold, respectively. and These represent the vehicle yaw angle deviation and the yaw angle deviation threshold, respectively. and These represent the start and end times of the experiment, respectively.

5. The adaptive model predictive control-based lane keeping method for curves as described in claim 4, characterized in that: In the neural network of step S3, the following relationships exist between the road curvature radius, road adhesion coefficient, safe cornering speed, prediction time domain, and control time domain: ； in, Indicates the safe speed for cornering; Indicates the prediction time domain; Indicates control over the time domain; Indicates the radius of curvature of the road; Indicates the road adhesion coefficient; and They represent about and Different nonlinear functions.

6. The adaptive model predictive control-based lane keeping method for curves as described in claim 4, characterized in that: The neural network dataset in step S3 includes road curvature radius and road adhesion coefficient. The dataset labels include safe speed, prediction time domain and control time domain. A circular path tracking experiment is conducted at the safe speed for cornering to determine the optimal time domain parameters under different combinations of vehicle speed, road curvature radius and road adhesion coefficient, and the neural network is trained.

7. An adaptive model predictive control curve lane keeping system, including a computer program, characterized in that: When the computer program is executed by a processor, it implements the steps of the method as described in any one of claims 1-6 above.

Images