CN111806246A - A kind of suspension system control method for maglev train - Google Patents

Abstract

The invention relates to a suspension system control method for a maglev train, which specifically includes the following steps: constructing a second-order sliding mode surface based on a suspension control dynamics model of the maglev train, and introducing an online real-time training neural network related to a positioning error signal The approximated nonlinear bounded function is used to obtain the final sliding mode variable structure control law model, which is used to construct the magnetic suspension controller of the suspension system; the physical parameters of the suspension system are input into the magnetic suspension controller; the magnetic suspension controller obtains the track and the car body in real time. After receiving the control signal, the peripheral hardware of the suspension system drives the suspension electromagnet to move to the target position within a limited time. Compared with the prior art, the present invention can realize arbitrary approximation of the uncertain working model parameters in practice, improve the adaptability of the controller to various environments, and finally improve the stability of the suspension system control.

Description

A kind of suspension system control method for maglev train

technical field

The invention relates to the field of maglev trains, in particular to a suspension system control method for a maglev train.

Background technique

Maglev train is a modern mode of transportation with non-contact electromagnetic levitation, guidance and drive systems. It relies on electromagnetic attraction or electric repulsion to suspend the train in the air so that there is no mechanical contact between the train and the track, and is driven by a linear motor. Maglev trains are ideal means of transportation due to their high speed, low energy consumption, comfortable ride and low noise. At present, according to the different suspension principles and methods used by maglev vehicles, maglev trains are generally divided into two categories, one is Electrodynamic Suspension, referred to as EDS type; the other is Electromagnetic Suspension, referred to as EMS type. . The EDS type maglev system uses electromagnetic repulsion to suspend the vehicle above the track, while the EMS type maglev system uses the attractive force generated by the electromagnet located under the track to lift the vehicle to ensure that it does not contact the track. The EDS type maglev system can suspend stably without applying control, while the EMS type maglev system needs to apply active control to ensure the stable suspension of the system. At present, the commercial operation is EMS type maglev train.

For the EMS maglev train, the suspension system is the key and core of the maglev train. However, the suspension system has strong nonlinearity and open-loop instability. In addition, the system parameters are uncertain, and the system is subject to external disturbances during operation. Therefore, the design of a high-performance suspension controller presents a high challenge.

The most pressing problem that the levitation system of the EMS type maglev train faces now is the uncertainty of the model and external disturbances, such as the uncertainty of passenger weight, the disturbance of wind, etc. Currently, most maglev vehicle suspension controllers are linear PID controllers. When the parameters are changed or the external disturbance is large, under the action of the current linear controller, the control performance of the system will be degraded, and the stability of the system will decrease or even lose its stability.

SUMMARY OF THE INVENTION

The purpose of the present invention is to provide a suspension system control method for a maglev train in order to overcome the above-mentioned defects of the prior art.

The object of the present invention can be realized through the following technical solutions:

A suspension system control method for a maglev train, characterized in that it specifically comprises the following steps:

S1. Construct a second-order sliding mode surface based on the suspension control dynamics model of the maglev train, and introduce a nonlinear bounded function that is related to the positioning error signal and approximated by an online real-time training neural network to obtain the final sliding mode variable structure control law model, Magnetic levitation controllers for building levitation systems;

S2. Input the set physical parameters of the suspension system in the magnetic suspension controller;

S3. The magnetic levitation controller obtains the gap data between the track and the car body in real time and outputs a control signal;

S4. After receiving the control signal, the peripheral hardware of the suspension system drives the suspension electromagnet to move to the target position within a limited time, and keeps it within the limit range of the position error.

Further, in the described step S1, the expression of the final sliding mode variable structure control law model is:

和

分别为未知的非线性有界函数f(·)和g(·)的神经网络逼近，x为网络输入，j为第j个隐含层节点，W^*,L^*是f(·)和g(·)的理想网络权重，h_f(x)和h_g(x)为神经网络的Koski方程，r为理想跟踪指令；where sgn(·) is the sign function, s is the dynamic sliding mode surface, e is the system error, c ₁ , c ₂ , η, μ are the control gain parameters,

and

Neural network approximation of unknown nonlinear bounded functions f( ) and g( ), respectively, x is the network input, j is the jth hidden layer node, W ^* , L ^* are f( ) and g ( ) ideal network weights, h _f (x) and h _g (x) are the Koski equations of the neural network, and r is the ideal tracking instruction;

和

According to the minimum parameter learning method, the adaptation rates of f( ) and g( ) are defined as a single parameter

and

Among them, the parameters γ ₁ , γ ₂ , Ω ₁ , Ω ₂ ∈R ⁺ .

Further, in the step S1, the final sliding mode variable structure control law model construction method includes:

S11. Establish an affine nonlinear mathematical model;

S12. Design the sliding film control law of the sliding mode controller according to the affine nonlinear mathematical model, and obtain a sliding mode variable structure control law model and two nonlinear bounded functions f(·) and g(·);

S13. Approximate the nonlinear bounded functions f(·) and g(·) in the sliding mode variable structure control law model through online learning of the RBF neural network, and obtain the final sliding mode variable structure control law model.

Further, in the described step S11, the expression of the affine nonlinear mathematical model is:

in,

In the formula, z ₁ represents the air gap spacing, z ₂ represents the change speed of the air gap spacing, z ₃ represents the air gap spacing acceleration, m is the body mass, μ ₀ is the vacuum permeability, N _m is the number of coil windings, A _m is the cross-sectional area of the magnet, and R _m represents the electromagnet winding resistance.

Further, in the described step S12, the expression of the sliding mode variable structure control law model is:

In the formula, u _SMC (x, t) is the control output, η, μ ∈ R ⁺ represent the constant arrival coefficient and exponential arrival coefficient, respectively, r is the ideal tracking command, c ₁ , c ₂ ∈ R ⁺ is the positive position control gain, e is the systematic error and s is the dynamic sliding surface.

Further, the suspension system includes a gap sensor, a chopper, a magnetic levitation controller and an electromagnet, the gap sensor is connected to the magnetic levitation controller through the chopper, and the electromagnet passes through the peripheral hardware magnetic levitation controller, so the The gap sensor is installed on the electromagnet.

Further, the gap sensor is connected to the chopper through the gap processing board, the control board and the interface conversion board.

Further, the magnetic levitation controller includes computer hardware and algorithm software, which is used to execute the operation of the neural network approximation algorithm, obtain the set physical parameters of the suspension system of the input, and obtain and output the gap data between the track and the vehicle body in real time. control signal.

Compared with the prior art, the present invention has the following advantages:

The invention re-establishes the magnetic suspension controller of the suspension system, constructs the second-order sliding mode surface based on the suspension control dynamics model of the magnetic suspension train, and introduces the nonlinear bounded function related to the positioning error signal and approximated by the online real-time training neural network. , the final sliding mode variable structure control law model can be obtained, which can achieve arbitrary approximation of the uncertain working model parameters in practice, improve the adaptability of the controller to various environments, and finally improve the stability of the suspension system control.

Description of drawings

FIG. 1 is a schematic flow chart of the present invention.

FIG. 2 is a schematic diagram of the suspension control dynamics model of the maglev train.

FIG. 3 is a schematic diagram of a closed-loop control system of a magnetic levitation controller.

FIG. 4 is a schematic diagram of the structure of the suspension system.

Detailed ways

The present invention will be described in detail below with reference to the accompanying drawings and specific embodiments. This embodiment is implemented on the premise of the technical solution of the present invention, and provides a detailed implementation manner and a specific operation process, but the protection scope of the present invention is not limited to the following embodiments.

As shown in FIG. 1 , this embodiment provides a suspension system control method for a maglev train, which specifically includes the following steps:

Step S1, construct a second-order sliding mode surface based on the suspension control dynamics model of the maglev train, and introduce a nonlinear bounded function related to the positioning error signal and approximated by an online real-time training neural network to obtain a final sliding mode variable structure control law model , a magnetic levitation controller for building a levitation system;

Step S2, input the physical parameters of the suspension system set in the magnetic suspension controller;

Step S3, the magnetic levitation controller obtains the gap data between the track and the vehicle body in real time and outputs a control signal;

In step S4, the peripheral hardware of the suspension system drives the suspension electromagnet to move to the target position within a limited time after receiving the control signal, and keeps it within the position error limit range to achieve a stable and reliable suspension control effect.

In step S1, the final sliding mode variable structure control law model construction method includes:

Step S11, establishing an affine nonlinear mathematical model;

Step S12, design the synovial film control law of the sliding mode controller according to the affine nonlinear mathematical model, and obtain a sliding mode variable structure control law model and two nonlinear bounded functions f(·) and g(·);

In step S13, the nonlinear bounded functions f(·) and g(·) in the sliding mode variable structure control law model are approximated by online learning through the RBF (radial basis) neural network, and the final sliding mode variable structure control law model is obtained.

The specific expansion of the above steps is as follows:

As shown in Figure 2, the suspension control dynamics model of the maglev train is a system of electromagnetic and mechanical equations established using Newton's law and Maxwell's equations:

为气隙间距的变化速度，状态量x₃(t)＝i_m(t)为控制输入电流，N_m为线圈绕组个数，A_m为磁体的截面积，m为车身质量，g为重力常数，u_m(t)为控制电压，f_d为干扰度，μ₀为真空磁导率，R_m表示电磁铁绕组电阻。In the above formula, the state quantity x ₁ (t)=x _m (t) is the air gap distance, and the state quantity

is the change speed of the air gap spacing, the state quantity x ₃ (t)=im (t) is the control input current, N _m is the number of coil windings, A _m is the cross-sectional area of the magnet, _m is the body mass, and g is the gravity Constant, _um (t) is the control voltage, f _d is the interference degree, μ ₀ is the vacuum permeability, and R _m is the electromagnet winding resistance.

Step S11, establishing an affine nonlinear mathematical model.

In order to obtain the affine nonlinear mathematical model of the suspension system, according to the principle of nonlinear coordinate transformation, the nonlinear transformation coordinates are selected as:

η=[z ₁ z ₂ z ₃ ] ^T ∈Ω _η , Ω _η ∈R ³

in,

In the formula, z ₁ represents the air gap spacing, z ₂ represents the change speed of the air gap spacing, and z ₃ represents the air gap spacing acceleration; x ₁ , x ₂ , and x ₃ are the state quantities x ₁ (t), x ₂ (t) , x ₃ (t).

Derivating the above formula, we get:

To simplify the formulation, define k and L as follows:

还可以表示为：According to Maxwell's equations and Biot-Savar theory,

It can also be expressed as:

Therefore, the affine nonlinear mathematical model is obtained:

Among them, two nonlinear bounded functions are defined:

Step S12 , design the control law of the synovial membrane according to the affine nonlinear mathematical model.

定义系统误差为e，误差变化率为

其中e＝z₁-r，r为期望位置。The state variables of the magnetic levitation system are selected as z ₁ =x ₁ =x _m (t),

Define the systematic error as e, and the error rate of change

where e=z ₁ -r, where r is the desired position.

The sliding surface is designed as:

Among them, c ₁ , c ₂ ∈ R ⁺ is a positive position control gain, e=z ₁ -r=x ₁ -r.

Derivation of the above formula yields:

Among them, f(x) and g(x) are two nonlinear bounded functions f(x) and g(x) in the affine nonlinear mathematical model.

Thus, the sliding mode variable structure control law model is obtained as follows:

Among them, u _SMC (x, t) is the controller output, η, μ ∈ R ⁺ represent the constant arrival coefficient and exponential arrival coefficient, respectively, and r is the ideal tracking command.

Verify the sliding mode variable structure control law model:

The Lyapunov equation is defined as:

This equation is positive semi-definite, and the derivative of both sides can be obtained at the same time:

Therefore, based on the Lyapunov theory, the control system is globally stable.

The above sliding mode variable structure control law model, but f(·), g(·) are uncertain or even unknown, so it is necessary to approximate f(·), g(·) through RBF neural network online learning.

In step S13, the sliding mode variable structure control law model is revised to obtain the final sliding mode variable structure control law model by approximating f(·) and g(·) through online learning of the RBF neural network.

A schematic diagram of the closed-loop control system is shown in Figure 3.

The input and output of the RBF neural network are:

f(x)=W ^*T h _f (x)+ε _f ,g(x)=L ^*T h _g (x)+ε _g

Among them, x is the network input, j is the jth hidden layer node, h=[h _j ] ^T is the output of the Gaussian equation, W, L are the approximate network approximation weights, W ^* , L ^* are f( ),g (·) ideal network weight, ε _f , ε _g are the network approximation errors, |ε _f |≤ε _Mf , |ε _g |≤ε _Mg , c is the center vector of the neural network node, and b is the width of the Gaussian basis function.

When x=[x1x2x3] ^T is determined, the output of the RBF neural network is:

和

是f(x)和g(x)的估计网络权重。where h _f (x) and h _g (x) are the Koski equations of the RBF neural network,

and

are the estimated network weights for f(x) and g(x).

是φ的在线估计值，学习误差

使

是正实数，

是

的在线估计值，学习误差为

The neural network minimum parameter learning method is used for f(·) and g(·), respectively. Let φ=W* ² , φ is a positive real number,

is the online estimate of φ, the learning error

Make

is a positive real number,

Yes

The online estimate of , and the learning error is

Therefore, the final sliding mode variable structure control law model is expressed as:

where sgn( ) is the symbolic equation, s is the dynamic sliding mode surface, and _um (x, t) is the control output.

和

According to the minimum parameter learning method, the adaptive laws of f( ) and g( ) are defined as a single parameter

and

Among them, γ ₁ , γ ₂ , Ω ₁ , Ω ₂ ∈R ⁺ .

Verify the final sliding mode variable structure control law model:

According to step S13, the equation can be obtained:

The Lyapunov equation is defined as:

It can be known: V≥0

in,

Have,

Similarly,

Because the RBF neural network approximation error _εf is a very small real number, it can be rewritten as follows:

Comprehensive adaptation rate:

in,

Available as follows:

in,

可得：solve inequalities

Available:

It can be seen that the system is ultimately uniformly bounded. Therefore, the boundedness and convergence of the system are proved.

This embodiment is implemented through the cooperation of software and hardware of the suspension system.

As shown in Figure 4, the levitation system includes a gap sensor, a chopper, a magnetic levitation controller and a levitation electromagnet. The gap sensor is connected to the maglev controller through a chopper, the suspension electromagnet is connected to the magnetic levitation controller through the peripheral hardware, and the gap sensor is installed on the suspension electromagnet. Further, the gap sensor is connected to the chopper through the gap processing board, the control board and the interface conversion board.

The magnetic levitation controller includes computer hardware and algorithm software. The algorithm software of the magnetic levitation controller is compiled and stored in the computer hardware. When the magnetic levitation controller is working, it is responsible for executing the operation of the neural network approximation algorithm, obtaining the set physical parameters of the levitation system input, and obtaining the gap data between the track and the car body in real time, and calculating and controlling the output control signal.

The working principle of this embodiment is:

The gap sensor measures and collects air gap spacing data in real time and at high speed. After analog-to-digital conversion, filtering and modulation, the air gap spacing data is transmitted to the computer equipment carrying the magnetic levitation controller through the communication line. The computer equipment carrying the magnetic levitation controller outputs the control quantity obtained by the magnetic levitation controller to the peripheral hardware, drives the levitation electromagnet to work, and suspends the train. The error between the air gap distance and the target position is calculated by the magnetic suspension controller to obtain the control output. With the real-time and uninterrupted work of the gap sensor, computer equipment, and peripheral hardware of the suspension electromagnet, the train will move to the target position within a limited time, and keep it within the limit of the position error to achieve a stable and reliable suspension control effect. .

The preferred embodiments of the present invention have been described above in detail. It should be understood that those skilled in the art can make many modifications and changes according to the concept of the present invention without creative efforts. Therefore, any technical solutions that can be obtained by those skilled in the art through logical analysis, reasoning or limited experiments on the basis of the prior art according to the concept of the present invention shall fall within the protection scope determined by the claims.

Claims (8)

1. a suspension system control method for maglev train, is characterized in that, specifically comprises the following steps: S1. Construct a second-order sliding mode surface based on the suspension control dynamics model of the maglev train, and introduce a nonlinear bounded function that is related to the positioning error signal and approximated by an online real-time training neural network to obtain the final sliding mode variable structure control law model, Magnetic levitation controllers for building levitation systems; S2. Input the set physical parameters of the suspension system in the magnetic suspension controller; S3. The magnetic levitation controller obtains the gap data between the track and the car body in real time and outputs a control signal; S4. After receiving the control signal, the peripheral hardware of the suspension system drives the suspension electromagnet to move to the target position within a limited time, and keeps it within the limit range of the position error. 2. a kind of suspension system control method for maglev train according to claim 1, is characterized in that, in described step S1, the expression of final sliding mode variable structure control law model is:

where sgn(·) is the sign function, s is the dynamic sliding mode surface, e is the system error, c ₁ , c ₂ , η, μ are the control gain parameters,

and

are the neural network approximations of the unknown nonlinear bounded functions f( ) and g( ), respectively, x is the network input, j is the jth hidden layer node, W ^* , L ^* are f( ) and g ( ) ideal network weights, h _f (x) and h _g (x) are the Koski equations of the neural network, and r is the ideal tracking instruction; According to the minimum parameter learning method, the adaptation rates of f( ) and g( ) are defined as a single parameter

and

Among them, the parameters γ ₁ , γ ₂ , Ω ₁ , Ω ₂ ∈R ⁺ . 3. a kind of suspension system control method for maglev train according to claim 1, is characterized in that, in described step S1, final sliding mode variable structure control law model construction method comprises: S11. Establish an affine nonlinear mathematical model; S12. Design the sliding film control law of the sliding mode controller according to the affine nonlinear mathematical model, and obtain a sliding mode variable structure control law model and two nonlinear bounded functions f(·) and g(·); S13. Approximate the nonlinear bounded functions f(·) and g(·) in the sliding mode variable structure control law model through online learning of the RBF neural network, and obtain the final sliding mode variable structure control law model. 4. a kind of suspension system control method for maglev train according to claim 3 is characterized in that, in described step S11, the expression of affine nonlinear mathematical model is:

in,

In the formula, z ₁ represents the air gap spacing, z ₂ represents the change speed of the air gap spacing, z ₃ represents the air gap spacing acceleration, m is the body mass, μ ₀ is the vacuum permeability, N _m is the number of coil windings, A _m is the cross-sectional area of the magnet, and R _m represents the electromagnet winding resistance. 5. a kind of suspension system control method for maglev train according to claim 4, is characterized in that, in described step S12, the expression of sliding mode variable structure control law model is:

In the formula, u _SMC (x, t) is the control output, η, μ ∈ R ⁺ represent the constant arrival coefficient and exponential arrival coefficient, respectively, r is the ideal tracking command, c ₁ , c ₂ ∈ R ⁺ is the positive position control gain, e is the systematic error and s is the dynamic sliding surface. 6. a kind of suspension system control method for maglev train according to claim 1 is characterized in that, described suspension system comprises gap sensor, chopper, maglev controller and electromagnet, described gap sensor The magnetic levitation controller is connected through a chopper, the electromagnet is connected to the magnetic levitation controller through peripheral hardware, and the gap sensor is installed on the electromagnet. 7 . The suspension system control method for a maglev train according to claim 6 , wherein the gap sensor is connected to the chopper through a gap processing board, a control board and an interface conversion board. 8 . 8. a kind of suspension system control method for maglev train according to claim 1, is characterized in that, described maglev controller comprises computer hardware and algorithm software, is used for executing the operation of neural network approximation algorithm, obtaining input The physical parameters of the suspension system are set, and the gap data between the track and the car body is obtained in real time and the control signal is output.

Images