A rear wheel steering vehicle stability control method based on BP neural network optimization

Abstract

The application is suitable for the field of automobile control technology, and provides a rear wheel steering vehicle stability control method based on BP neural network optimization, including the following steps: a front wheel steering angle reference model is used to solve a desired mass center side slip angle and a desired yaw rate; an MPC control algorithm with rear wheel steering angle as a control variable is designed; an adjustable parameter for the MPC algorithm is designed, including a BP neural network controller for adaptive real-time optimization of system output weight parameters and a prediction time domain; and finally, the calculated rear wheel steering angle is output to the vehicle. Through BP neural network optimization, the control parameters of the traditional model prediction algorithm are optimized and improved, the accuracy of the control system model is improved, and the stability of the four-wheel steering system is ensured.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of automobile control, and particularly relates to a rear wheel steering vehicle stability control method based on BP neural network optimization. BACKGROUND

[0002] With the increasing demand for automobile handling stability and active safety, the active rear wheel steering technology is gradually valued by domestic and foreign automobile manufacturers, is widely researched and applied in real vehicle field, and is a hot direction for the development of future intelligent vehicles.

[0003] At present, the control methods for the handling stability of the rear wheel steering vehicle mainly include the following: (1) feedforward proportional control: using K value to represent the proportional coefficient between front and rear wheel steering angles, the rear wheel steering angle of the vehicle is simply controlled, the control method is simple, and the control effect needs to be improved; (2) feedforward proportional + yaw rate feedback control: adding yaw rate feedback on the basis of the feedforward proportional control strategy, the traditional classical algorithm, which improves the stability of the vehicle; (3) LQR control: the feedforward and feedback matrix K are solved by establishing the Riccati equation, so that the rear wheel steering angle acts on the system. This method greatly improves the stability and dynamic response characteristics of the vehicle, but has the problems of being unable to handle state quantity constraints and actuator constraints, and requiring high accuracy of model parameters; (4) other methods, such as PID, sliding mode variable structure control, etc., have the problems of being greatly dependent on the adjustment parameters, chattering, etc. SUMMARY

[0004] The purpose of the present application is to provide a rear wheel steering vehicle stability control method based on BP neural network optimization, which aims to solve the problems proposed in the above background.

[0005] To achieve the above purpose, the present application provides the following technical scheme:

[0006] A rear wheel steering vehicle stability control method based on BP neural network optimization, comprising the following steps:

[0007] Step S1, establishing a rear wheel steering vehicle reference model, comprising: according to the obtained vehicle speed v and front wheel steering angle δ f , taking the rear wheel steering vehicle as a reference model, considering the upper limit of the yaw rate, solving the control target of the active rear wheel steering system;

[0008] Step S2, designing a BP neural network, comprising: inputting the vehicle speed v and the front wheel steering angle δ f into the BP neural network, determining the connection weight and the threshold value through the training process of the BP neural network; determining the system output weight q β , q γ and the prediction time domain N pReal-time feedback to the MPC controller;

[0009] Step S3, designing a model predictive controller, including: building a linear two-degree-of-freedom four-wheel steering model of the vehicle, designing a model predictive controller considering constraints, inputting real-time feedback of the vehicle system state quantities: vehicle speed v, yaw rate γ and mass center side slip angle β into the nonlinear controller module; combining a reference model, solving the vehicle rear wheel steering angle through quadratic programming, and outputting the optimized vehicle rear wheel steering angle to the vehicle system.

[0010] Further, the expected mass center side slip angle β ref is considered as zero, and the expected yaw rate γ ref is considered as the upper limit, and the establishment of the front-wheel steering vehicle reference model in step S1 is as follows:

[0011] β ref = 0

[0012]

[0013] In the formula, u is the vehicle speed, L is the vehicle wheelbase, K is the stability factor, μ is the friction coefficient, g is the gravity acceleration, δ f is the front wheel steering angle.

[0014] Further, in step S2, the specific steps of designing the BP neural network are as follows:

[0015] Step S21, designing an experimental group: setting input x = {x1, x2} and output y = {y1, y2, y3}, wherein x1 is the vehicle speed v, x2 is the front wheel steering angle δ f ; y1 is the system output weight q β of the mass center side slip angle, y2 is the system output weight q γ of the yaw rate, and y3 is the prediction time domain N p ;

[0016] Sampling records the input data and corresponding output data of each group to obtain training samples

[0017] Step S22, selecting a training set and a test set, and improving the network training of the BP neural network, finally determining the connection weight and the threshold, and recording the training error and test error under the optimal network structure of each group;

[0018] Step S23, after the network training is completed, the double-hidden-layer BP neural network is determined by comparing the minimum point of the test error.

[0019] Further, in step S22, the Levenberg-Marquardt algorithm is used to improve the network training of the BP neural network, and the training process includes forward calculation, error calculation and back propagation.

[0020] Further, in step S3, the specific steps of designing the model predictive controller are as follows:

[0021] Step S31, a two-degree-of-freedom four-wheel steering vehicle model is built, and the expression is as follows:

[0022]

[0023]

[0024] In the formula, k1 and k2 are the tire cornering stiffness of the front and rear axles of the vehicle, m is the mass of the vehicle, a and b are the wheelbase of the vehicle mass center to the front and rear axles, I z is the moment of inertia of the vehicle, δ r is the rear wheel steering angle;

[0025] Step S32, the linearized model is discretized, and the expression is as follows:

[0026] x(k+1)=A k x(k)+B k u(k)+D k w(k)

[0027] Wherein, A k =TA, B k =TB, D k =TD;

[0028] In the formula, k is the current sampling time, and T is the sampling period;

[0029] Step S33, the prediction model is derived, and the output form in the prediction time domain is expressed as:

[0030] Y=ψξ(k)+ΘΔU(k)+Γφ(k);

[0031] Step S34, design the objective function:

[0032]

[0033] Step S35, the BP neural network feeds back the system output weight q β , q γ and the prediction time domain N p to the MPC controller in real time.

[0034] Further, in step S35, set the control weight R=I, and the control time domain N c= 0.3N p If the quadratic programming problem has a solution, output the control increment ΔU(k), and apply the control quantity U(k) = U(k-1) + ΔU(k) to the vehicle system.

[0035] Compared with the prior art, the present application has the beneficial effects that:

[0036] The rear wheel steering vehicle stability control method based on BP neural network optimization optimizes and improves the control parameters of the traditional model prediction algorithm, improves the precision of the control system model, and ensures the stability of the four-wheel steering system. BRIEF DESCRIPTION OF DRAWINGS

[0037] Figure 1 The figure is a whole vehicle system control block diagram of the present application.

[0038] Figure 2 The figure is a BP neural network training flowchart of the present application.

[0039] Figure 3 The figure is a model prediction control implementation flowchart of the present application. DETAILED DESCRIPTION

[0040] In order to make the purpose, technical scheme and advantages of the present application clearer, the present application is further described in detail below in combination with the drawings and examples. It should be understood that the specific examples described herein are only used to explain the present application, and are not used to limit the present application.

[0041] The specific implementation of the present application is described in detail below in combination with specific examples.

[0042] As shown in the figure, the rear wheel steering vehicle stability control method based on BP neural network optimization provided by one embodiment of the present application comprises the following steps: Figure 1

[0043] Step S1, establishing a wheel steering vehicle reference model, comprising: according to the obtained vehicle speed v and front wheel steering angle δ f , taking the front wheel steering vehicle as the reference model, considering the upper limit of the yaw rate, solving the control target of the active rear wheel steering system;

[0044] Step S2, designing a BP neural network, comprising: inputting the vehicle speed v and front wheel steering angle δ f into the BP neural network, determining the connection weight and threshold value through the training process of the BP neural network; feeding back the system output weight q β , q γ and the prediction time domain N p to the MPC controller in real time;

[0045] ​Step S3, designing a model predictive controller, comprising: building a linear two-degree-of-freedom four-wheel steering model of the vehicle, designing a model predictive controller considering constraints, inputting real-time feedback vehicle system state quantities: vehicle speed v, yaw rate γ and mass center side slip angle β into the nonlinear controller module; combining reference model output quantities, solving the vehicle rear wheel steering angle through quadratic programming, and outputting the optimized vehicle rear wheel steering angle to the vehicle system.

[0046] As a preferred embodiment of the present application, in steps S1 and S3, the control target and the model output quantity of the active rear wheel steering system are both the desired mass center side slip angle β ref and the desired yaw rate γ ref .

[0047] As a preferred embodiment of the present application, the desired mass center side slip angle β ref is considered to be zero, and the desired yaw rate γ ref is considered to have an upper limit, and the establishment of the front wheel steering vehicle reference model in step S1 is as follows:

[0048] β ref = 0

[0049]

[0050] In the formula, u is the vehicle speed, L is the wheelbase of the vehicle, K is the stability factor, μ is the friction coefficient, g is the gravity acceleration, δ f is the front wheel steering angle.

[0051] In the embodiment of the present application, for step S1, the actual mass center side slip angle is estimated based on the unscented Kalman filter. This method uses probability distribution to approximate nonlinearity, and can effectively overcome the problems of low estimation accuracy, poor stability and large calculation amount of EKF.

[0052] As shown in Figure 2 , as a preferred embodiment of the present application, in step S2, the specific steps of designing the BP neural network are as follows:

[0053] Step S21, designing an experimental group: setting input x = {x1, x2} and output y = {y1, y2, y3}, wherein x1 is the vehicle speed v, x2 is the front wheel steering angle δ f ; y1 is the system output weight q β of the mass center side slip angle, y2 is the system output weight q γ of the yaw rate, and y3 is the prediction time domain N p ;

[0054] Sample and record the input data and corresponding output data of each group to obtain training samples

[0055] Step S22, select the training set and test set, and improve the network training of BP neural network, finally determine the connection weight and threshold, record the training error and test error under the optimal network structure of each group;

[0056] Step S23, after network training, determine to use double hidden layer BP neural network by comparing the minimum point of test error.

[0057] In the embodiment of the application, in step S21, first set the number of groups 3600 groups, each group corresponds to different front wheel turning angle 1-30° and vehicle speed 1-120km / h. Then for any group of vehicles, record the input x={x1, x2}. Initialize the system output weight q β , q γ and prediction horizon N p , run the control program, simulate to obtain the steady-state vehicle centroid side slip angle output and steady-state yaw rate output, if the absolute value of the centroid side slip angle error exceeds the allowable range, increase q β , decrease q γ ; if the absolute value of the yaw rate error exceeds the allowable range, decrease q β , increase q γ . If the speed increases or the turning angle increases, appropriately increase the prediction horizon N p . Repeat until the absolute values of the centroid side slip angle and the yaw rate of the vehicle are within the allowable range. At the end of each data collection period, record the input data and the corresponding output data of each group, thereby obtaining the training sample

[0058] In step S22, first train the data pre-processing: randomly select 3300 groups from 3600 groups as the neural network training set, and the other 300 groups as the test set; because the linear function purelin is used as the output layer of the network, normalization processing is not needed. Then select the BP neural network algorithm: set the learning rate of the network to 0.05, the maximum number of training to 2000, use tan-sigmoid function between the input layer and the hidden layer, the expression is use purelin as the transfer function between the output layer and the hidden layer; finally determine the connection weight and threshold, record the training error and test error under the optimal network structure of each group;

[0059] In step S23, after network training, determine to use double hidden layer BP neural network by comparing the minimum point of test error, the best network topology is 12-6-3. Thus, the training of the BP neural network is completed, which can be used for online adjustment of the system output weight q β , q γ and prediction horizon N p.

[0060] As a preferred embodiment of the present application, in the step S22, the Levenberg-Marquardt algorithm is used to improve the network training of the BP neural network, and the training process includes forward calculation, error calculation and back propagation.

[0061] As shown in Figure 3 , as a preferred embodiment of the present application, in the step S3, the specific steps of designing the model predictive controller are as follows:

[0062] Step S31, a two-degree-of-freedom four-wheel steering vehicle model is built, and the expression is as follows:

[0063]

[0064]

[0065] In the formula, k1 and k2 are respectively the tire cornering stiffness of the front and rear axles of the vehicle, m is the mass of the vehicle, a and b are respectively the wheelbase of the vehicle from the mass center to the front and rear axles, I z is the moment of inertia of the vehicle, δ r is the rear wheel steering angle;

[0066] Step S32, discretize and linearize the model, and the expression is as follows:

[0067] x(k+1)=A k x(k)+B k u(k)+D k w(k)

[0068] Wherein, A k =TA, B k =TB, D k =TD;

[0069] In the formula, k is the current sampling time, and T is the sampling period;

[0070] Step S33, derive the prediction model, and the output quantity in the prediction time domain is expressed as:

[0071] Y=ψξ(k)+ΘΔU(k)+Γφ(k);

[0072] Step S34, design the objective function:

[0073]

[0074] Step S35, the BP neural network feeds back the system output weight q β , q γ and the prediction time domain N p to the MPC controller in real time.

[0075] In the embodiment of the present application, the continuous model is converted into a discrete time-varying model by the forward Euler method, which facilitates the design of the MPC controller. In step S33, in order to facilitate the constraint on the original state variable x and the control variable u later, a new state variable ξ(k) is constructed:

[0076]

[0077] η(k) = Cξ(k) + φ(k) k ξ(k)

[0078] wherein,

[0079]

[0080] The output η(k) at the current time can be calculated from the state variable ξ(k) at the current time, the disturbance variable w(k) at the current time and the unknown control increment Δu(k) at the current time. Similarly, the output Y in the prediction time domain N can be calculated, i.e. the size of the yaw rate and the mass center side slip angle in the future time, which can be expressed in the following form: Y = ψξ(k) + ΘΔU(k) + Γφ(k). p

[0081] wherein,

[0082]

[0083]

[0084]

[0085]

[0086] In step S34, the optimization objective function with constraints is converted into a quadratic programming problem by the MPC, and the objective function is defined as:

[0087]

[0088] s.t.ΔU min ≤ΔU(t+i|t)≤ΔU max

[0089] U min ≤U(t+i|t)≤U max

[0090] y hc,min ≤y hc ≤y hc,max

[0091] y sc,min -ε≤y sc ≤y​sc,max + ε

[0092] where ΔU is a series of control increments in the control horizon, ε is a relaxation factor, y hc is a hard constraint output, y sc is a soft constraint output which can be adjusted by the relaxation factor.

[0093] where the system output weight is set as:

[0094]

[0095] As a preferred embodiment of the present application, in step S35, the control weight R = I, the control horizon N c = 0.3N p , if the quadratic programming problem has a solution, output the control increment ΔU(k), and apply the control quantity U(k) = U(k-1) + ΔU(k) to the vehicle system.

[0096] The above is only a preferred embodiment of the present application, and it should be noted that, for those skilled in the art, without departing from the concept of the present application, a number of modifications and improvements can be made, which should also be considered as the protection scope of the present application, and these will not affect the effect and practicality of the present application.

Claims

1. A rear wheel steering vehicle stability control method based on BP neural network optimization, characterized in that, The method comprises the following steps: Step S1, establishing a turning vehicle reference model, including: according to the obtained vehicle speed and front wheel steering angle , taking the front wheel steering vehicle as a reference model, considering the upper limit of the yaw rate, solving the control target of the active rear wheel steering system; Step S2, designing BP neural network, including: inputting vehicle speed and front wheel steering angle to BP neural network, determining connection weight and threshold value through training process of BP neural network; feeding back system output weight , and prediction time domain to MPC controller in real time; Step S3, designing a model predictive controller, including: building a linear two-degree-of-freedom four-wheel steering model of the vehicle, designing a model predictive controller considering constraints, inputting real-time feedback vehicle system state quantities: vehicle speed , yaw rate and center of mass side slip angle into a nonlinear controller module; combining a reference model, solving the vehicle rear wheel steering angle through quadratic programming, and outputting the optimized vehicle rear wheel steering angle to the vehicle system.

2. The BP neural network optimization-based rear wheel steering vehicle stability control method according to claim 1, characterized by, the desired yaw angle considering the desired yaw rate the upper limit, the establishment of the front wheel steering vehicle reference model in step S1 is as follows: ； wherein is the vehicle speed, is the wheelbase of the vehicle, is the stability factor, is the friction coefficient, is the center of gravity acceleration, is the front wheel steering angle.

3. The BP neural network optimization based rear wheel steering vehicle stability control method according to claim 1, characterized in that, In the step S2, the specific steps of designing the BP neural network are as follows: Step S21, design experiment group: set input quantity , output quantity , wherein is vehicle speed , is front wheel steering angle ; is the system output quantity weight of the mass center side slip angle , is the system output quantity weight of the yaw rate , is the prediction time domain ; The sampling records the input data and the corresponding output data of each group to obtain training samples ; Step S22, select the training set and test set, and improve the network training of the BP neural network, finally determine the connection weight and threshold, and record the training error and test error under the optimal network structure of each group; Step S23, after the network training is completed, the minimum point of the test error is compared to determine that the double-hidden-layer BP neural network is adopted.

4. The BP neural network optimization based rear wheel steering vehicle stability control method according to claim 3, characterized in that, In the step S22, the Levenberg-Marquardt algorithm is adopted to improve the network training of the BP neural network, and the training process comprises forward calculation, error calculation and backward propagation.

5. The BP neural network optimization based rear wheel steering vehicle stability control method according to claim 1, characterized in that, In the step S3, the specific steps of designing the model predictive controller are as follows: Step S31, build a two-degree-of-freedom four-wheel steering vehicle model, and the expression is as follows: ； ； wherein, , are the tire cornering stiffness of the front and rear axles, respectively, is the vehicle mass, , are the wheel base of the vehicle from the center of mass to the front and rear axles, respectively, is the vehicle moment of inertia, is the rear wheel steering angle; Step S32, discretize the linearized model, and the expression is as follows: ； wherein , , ; In the formula, k is the current sampling time, and T is the sampling period; Step S33, derive the prediction model, and the output quantity in the prediction time domain is expressed as: ； Step S34, design the objective function: ； Step S35, the BP neural network weights the system output , and the prediction horizon feeds back to the MPC controller in real time.

6. The BP neural network optimization-based rear wheel steering vehicle stability control method according to claim 5, characterized in that, The step S35, set control weight , control time domain =0.3 , by solving quadratic programming output control amount , and its effect on the vehicle system.

Images