Self-adaptive robust control method for electro-hydraulic brake-by-wire system based on neural network

Abstract

The invention relates to an electro-hydraulic brake-by-wire system adaptive robust control method based on a neural network, and belongs to the field of electro-hydraulic control. The method comprises the following steps: establishing an electro-hydraulic drive-by-wire brake system mathematical model, designing a neural network observer, designing a neural network weight adaptive law based on a Lyapunov function to approach interference in an optimal weight estimation system, using an observed system state to design a control law, forming an auxiliary error signal by adopting a backstepping method and the system state, and calculating the interference in the electro-hydraulic drive-by-wire brake system. And designing an adaptive robust algorithm. And designing a self-adaptive robust controller based on a backstepping method, designing a self-adaptive law of each parameter, and realizing asymptotically stable system tracking through a virtual control law and a state variable. Aiming at the problem that the brake control performance is obviously influenced by the severe working environment of the engineering vehicle, the large inertia of the engineering vehicle and numerous unknown interferences, the response speed of the electro-hydraulic brake-by-wire system and the control precision under the interferences are improved.

Description

Technical Field

[0001] The present invention mainly relates to the technical field of control, in particular to the technical field of electronic hydraulic brake system pressure control, and specifically to an adaptive robust control method for an electro-hydraulic wire control brake system based on a neural network. Background Art

[0002] The braking system is the key to determining the safety of engineering vehicles and is developing towards real-time controllability, integration, intelligence, rapid response, etc. The braking system of engineering vehicles is a complex non-linear system. When factors such as parameter uncertainty and unknown disturbances are serious, the braking force will become unstable. The harsh working environment and its own large inertia also pose challenges to control.

[0003] At the control algorithm level, since the parameters of the PID controller remain unchanged, the control effect cannot meet all complex working conditions. The fuzzy PID control algorithm is used to deal with the situation where the mathematical model cannot be accurately established, but the pressure building speed is not as good as that of the traditional PID control algorithm. Therefore, it adopts a segmented fuzzy PI control algorithm to control the wheel cylinder hydraulic pressure according to the error size, improving the pressure building speed. However, the PID algorithm is difficult to cope with the strong non-linearity in various systems. Researchers have conducted a large number of studies on adaptive control, sliding mode control, robust control, active disturbance rejection control, etc. The sliding mode controller is widely used because it does not depend on the system model, but the chattering phenomenon brought by the sliding mode controller seriously damages the control quality. Researchers designed a control system that switches between a pressure loop and a position loop. The position loop is designed as an adaptive robust control to enable the direct drive valve to quickly pass through the valve dead zone and solve non-linear factors such as parameter uncertainty, but the control algorithm is mandatory at the switching point and is prone to mutations. By combining the adaptive control law, the robust linear feedback term, and the integral robust control term, the uncertain parameters, friction state, and modeling errors are compensated to solve the above problems. Different scholars use different algorithms to deal with a certain non-linearity of the system. However, in actual working conditions, due to the numerous non-linear factors in the system operation, high-performance control algorithms need to be further studied. Summary of the Invention

[0004] To address the deficiencies of current technologies, the present invention combines existing technologies and starts from practical applications to provide an adaptive robust control method for an electro-hydraulic wire control braking system based on a neural network, aiming to solve the problem of braking control performance of engineering vehicles in complex environments. A neural network weight adaptation law is designed using the Lyapunov function to approximate the interference in the optimal weight estimation system. Further, the observed system states are used to design the control law. An auxiliary error signal is formed by combining the backstepping method and the system states based on the Lyapunov function, and an adaptive robust algorithm is designed. Finally, the adaptation laws of each parameter are defined according to the Lyapunov stability criterion, and the parameter uncertainties and interferences are compensated through feedforward compensation to ensure the stability of the system.

[0005] The technical solution of the present invention is as follows: An adaptive robust control method for an electro-hydraulic wire control braking system based on a neural network includes the following steps: S1. Establish a mathematical model of the electro-hydraulic wire control braking system; S2. Construct a neural network observer, design a neural network weight adaptation law based on the Lyapunov function, use the observed system states to design the control law, and prove the stability based on the Lyapunov function; S3. Construct an adaptive robust controller based on the backstepping method, design the adaptation laws of each parameter, and achieve asymptotic stable system tracking through the virtual control law and state variables.

[0006] Furthermore, in step S1, a three-position three-way electromagnetic direct-acting valve is adopted for the pressure regulation part of the wire control braking system. The main driving part of the electromagnetic direct-acting valve is an electromagnetic linear actuator. By controlling the spool displacement, the braking pressure of the wheel cylinder is controlled. The hydraulic execution part of the electronic hydraulic braking system also includes a brake wheel cylinder. The mathematical model of the electro-hydraulic wire control braking system is constructed based on the electromagnetic linear actuator, steady-state hydraulic pressure, brake wheel cylinder motion equation, and flow equation.

[0007] Furthermore, step S2 specifically includes: S21. Construct a neural network observer to feedforward compensate the observed system states to the control system. In the neural network observer, the spool displacement, velocity, hydraulic pressure, and hydraulic pressure change rate are regarded as state variables. For the total disturbance observed by the neural network observer, a three-layer neural network is used, and the output of the neural network is used to fit the total disturbance of the system; S22. Construct a neural network weight adaptation law to estimate the ideal weight; S23. Construct a Lyapunov function to prove the stability and verify the mathematical stability of the observer and the adaptation law.

[0008] Furthermore, step S3 specifically includes: S31. Design a virtual control law based on the backstepping method, which is coupled with the disturbance estimation value and the observed state; S32. Construct an adaptive robust controller by using the Lyapunov function combined with the parameter adaptation rate.

[0009] Furthermore, the digital-analog model expression of the electro-hydraulic line control braking system is as follows: , where, is the piston cross-sectional area, is the brake fluid pressure, is the mass of the piston, is the displacement of the brake wheel cylinder piston, is the equivalent damping coefficient of the piston, is the initial length of the spring, is the spring stiffness in the piston cylinder, is the brake fluid flow rate, is the bulk modulus of elasticity of the liquid, is the volume of the brake wheel cylinder, is the pressure-flow coefficient including leakage, is the total mass including the electromagnetic linear actuator, spool valve, connecting parts, and spring, is the spool valve movement speed, is the input voltage, is the motor constant of the electromagnetic actuator, is the displacement of the spool valve, is the coil resistance, is the damping coefficient, is the spring stiffness, is the frictional force, is the steady-state liquid pressure, is the transient liquid pressure.

[0010] Furthermore, the calculation method of the steady-state liquid pressure is as follows: ; The calculation method of the transient liquid pressure is as follows: ; In the formula, is the valve port area gradient, is the valve port pressure difference, is the flow coefficient, is the actual flow length of the hydraulic oil in the valve cavity, is the oil density.

[0011] Furthermore, the expression of the neural network observer is as follows: ; In the formula, , representing the brake fluid pressure; , representing the hydraulic pressure change rate; , representing the displacement of the spool valve, , representing the spool valve movement speed, ; ; , is the oil source pressure, , , , , , is the positive gain of the observer, - are system parameters, is related to the associated disturbance, is related to the associated disturbance; The weight adaptation law expression is as follows: ; In the formula, , are respectively related to -related, and related to the weight adaptation laws, , are respectively related to -related, and related to the weight adaptation law gain coefficients, is the estimated value of the radial basis function set in the neural network related to , is the estimated value of the radial basis function set in the neural network related to ; ; .

[0012] Further, the controller expression is as follows: ; The parameter adaptation law expression is as follows: ; Among them, is a given positive number, , , , , , are respectively the virtual control laws defined corresponding to , , , is the defined positive number, , 。

[0013] Advantages of the present invention: For the adaptive robust control method of the electro-hydraulic wire control braking system based on neural network of the present invention, an RBF neural network observer is designed. The unknown disturbance is nonlinearly approximated by the ideal weight through the weight adaptive law, and the observed state and the adaptive law are combined based on the backstepping method to form a complete adaptive robust control law to stabilize the system. The verification results prove that the proposed improved segmented active disturbance rejection controller can improve the response speed of the electro-hydraulic system. In the presence of disturbances, the designed controller has good anti-disturbance ability and control accuracy, and as the white noise power increases, the performance of the designed controller basically does not deteriorate. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] Figure 1 It is the overall principle block diagram of the adaptive robust control method of the electro-hydraulic wire control braking system based on neural network of the present invention.

[0015] Figure 2 It is the performance comparison diagram of the adaptive robust control method of the electro-hydraulic wire control braking system based on neural network of the present invention and other control methods during step response.

[0016] Figure 3 It is the performance comparison diagram of the adaptive robust control method of the electro-hydraulic wire control braking system based on neural network of the present invention and other control methods under the cyclic working conditions when being disturbed. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0017] In combination with the drawings and specific embodiments, the present invention will be further described. It should be understood that these embodiments are only used to illustrate the present invention and not to limit the scope of the present invention. In addition, it should be understood that after reading the content taught by the present invention, those skilled in the art can make various changes or modifications to the present invention, and these equivalent forms also fall within the scope defined by this application.

[0018] This embodiment provides an adaptive robust control method for an electro-hydraulic wire control braking system based on a neural network. The main steps are as follows: S1. Establish the mathematical model of the electro-hydraulic wire control braking system. The pressure regulation part of the wire control braking system uses a three-position three-way electromagnetic direct-acting valve, and the main driving part of the electromagnetic direct-acting valve is an electromagnetic linear actuator; S2. Design a neural network observer, design the neural network weight adaptive law based on the Lyapunov function, use the observed system state to design the control law, and prove the stability based on the Lyapunov function; S3. Design an adaptive robust controller based on the backstepping method, design the adaptive laws of each parameter, and achieve asymptotic stable system tracking through the virtual control law and state variables.

[0019] Among them, in step S1, the construction vehicle has a stable hydraulic source. The pressure regulating part of the wire-controlled braking system adopts a three-position three-way electromagnetic direct-acting valve. By applying a PWM signal to the valve coil, the displacement of the valve core is controlled to achieve the control of the wheel cylinder braking pressure. In addition, the hydraulic execution part of the electronic hydraulic braking system also includes a brake wheel cylinder and corresponding brake pipelines. A pressure sensor is set in the brake pipeline, which can measure the wheel cylinder pressure in real time and feedback it to the brake pressure controller. The mathematical model of the electromagnetic linear actuator is as follows: (1), In the formula: is the current; is the coil resistance; is the coil inductance; is the damping coefficient; is the spring stiffness; is the total mass including the electromagnetic linear actuator, valve core, connecting piece, and spring; is the motor constant of the electromagnetic actuator; is the input voltage; is the moving speed; is the displacement of the valve core; is the steady-state hydraulic pressure; is the transient hydraulic pressure; is the frictional force. Among them: (2), (3); In the formula: is the brake fluid flow rate; is the flow coefficient, usually taken as 0.61; is the valve port area gradient. is the actual flow length of the hydraulic oil in the valve cavity; is the valve port pressure difference; is the oil density; (4); In the formula, is the oil source pressure, is the brake fluid pressure.

[0020] Taking the brake wheel cylinder and the brake caliper as the research objects, the motion formula can be expressed as: (5); According to the liquid volume elasticity modulus formula, the brake wheel cylinder flow equation is: (6), (7); Wherein: is the mass of the piston; is the displacement of the piston of the brake wheel cylinder; is the cross-sectional area of the piston; is the spring stiffness in the piston cylinder; is the equivalent damping coefficient of the piston; is the initial length of the spring; is the volume of the brake wheel cylinder; is the pressure flow coefficient including leakage.

[0021] Finally, the total mathematical model of the system can be obtained through the direct drive valve model including the electromagnetic linear actuator and the steady-state hydraulic pressure, and the brake system equations including the motion equation and the flow equation of the brake wheel cylinder.

[0022] For step S2, the neural network observer is designed as follows: (8), wherein, ; ; , respectively represent the spool displacement and velocity. ; ; ; are system parameters.

[0023] Design a neural network observer to observe the total disturbance. A three-layer neural network is used, and the output of the neural network is used to fit the total disturbance of the system. According to the approximation characteristics of the neural network, there exists an ideal weight vector W , such that: (9); wherein, is the unmodeled dynamics; is the external disturbance; , representing the error between the true value and the estimated value; the input is when the disturbance including unmodeled dynamics; is the input is when the disturbance including unmodeled dynamics; is the set of radial basis functions in the RBF neural network, is the former estimated value; is the approximation error; when represents related to , when represents related to .

[0024] In this embodiment, an RBF neural network is used to approximate the interference. The first layer of the neural network is the input layer; the second layer is the hidden layer, and the third layer is the output layer. W is the ideal weight of the linear combination of the non - linear interference functions. Under the ideal weight, the unknown interference is , and the radial basis activation function uses the Gaussian function as: (10); In the formula, is the control input, is the center point of the th neuron. Its input has the same vector structure form, represents the th radial basis function width; according to the properties of the Gaussian function, it can be obtained that the distance between the sample input and the center point determines the value of the non - linear function obtained by the linear combination of each radial basis function; The observer is designed as follows: (11); In the formula, , , , , , is the positive gain of the observer.

[0025] Let , and the weight adaptation rate is defined as: (12); In the formula, ; .

[0026] The observation error is defined as: (13); A Lyapunov function is defined as: (14); is the error between the weight estimate value and the true value. Taking the derivative of the above formula gives: (15); Assume , According to the inequality theorem, there are: (16), (17), (18), (19); Wherein , , is a given positive gain.

[0027] There exists a matrix (20) that satisfies the positive definite condition of the matrix: (20); Wherein: ; ; ; ; ; ; ; ; ; ; .

[0028] Substituting the matrix (20) into equations (15)-(19) gives: (21); Wherein, ; ; .

[0029] Then, according to the second criterion, it can be proven that the observer is stable, i.e., , , , , and the weight estimate can finally reach the ideal weight value through the adaptive law.

[0030] For step S3, an adaptive robust control method is adopted. Through online parameter estimation and real-time adjustment, the influence of these uncertainty factors on the control performance can be effectively suppressed. The virtual control law is defined as , , .

[0031] Let , be the target value. The Lyapunov function 1 is defined as follows: (22); Wherein, is a defined positive number. Differentiating equation (22) gives: (23); To make equation (23) less than 0 under the action of the controller, let .

[0032] Let , substituting the above equation into Equation (23) gives: (24); It can be seen from Equation (24) that the control objective is transformed into making .

[0033] Define the Lyapunov function 2 as shown below. Define , = 1, 2, 3, 4 are given positive numbers.

[0034] (25); In order to make Equation (25) less than 0 under the action of the controller, let: (26); Differentiate Equation (25), let , substituting Equation (26) gives: (27); Where: , .

[0035] It can be seen from Equation (27) that the control objective is transformed into making ; Define the Lyapunov function 3 as shown below: (28); In order to make Equation (28) approach 0 under the action of the controller, let: (29); Differentiate Equation (28), , substituting the above equation gives: (30); Where: , according to Equation (30), it can be seen that the control objective is transformed into making . Define the Lyapunov function 4 as shown below: (31); Differentiate the above equation, define the control law as , substituting the above equation gives: (32); Where: , , from the above results, define the parameter adaptation law as follows: (33).

[0036] To test the response speed of the control algorithm (ARCNN) of the present invention, auto-disturbance rejection control (ADRC), and adaptive robust control (ARC), the target is set to 4 Mpa, and the tracking performance curve is as follows Figure 2 shown. The pressure rise time of ADRC is 104 ms; the pressure rise time of ARC is 80 ms; the rise time of ARCNN is 60 ms, which is 25% and 42.3% shorter than that of ADRC and ARC respectively. From the above analysis, it can be seen that the response speed of ARCNN is faster than that of other control methods. The mean square error values of the steady-state errors of ADRC and ARC are 0.14 Mpa and 0.03 Mpa respectively, and that of ARCNN is 0.014 Mpa. Therefore, ARCNN has a faster response speed and control accuracy.

[0037] To test the anti-interference ability of the control algorithm under interference, an interference signal with a certain power density is given, and the response curves of this control algorithm and other control algorithms are as follows Figure 3 shown. The target hydraulic pressure is set as the signal of the pressure increase and decrease cycle working condition, which simulates the irregular braking when the braking system of an engineering vehicle works. According to the results, the pressure rise lag time of ADRC is 20 ms under the condition of interference input; the pressure rise lag times of ARC and ARCNN are 5 ms; under the interference input, the mean square error value of ADRC is 0.14 Mpa, the mean square error value of ARC is 0.06 Mpa, and the mean square error value of ARCNN is 0.03 Mpa. From the above data, it can be seen that ARCNN can track the braking instruction of the cycle working condition with better control effect under the condition of interference input, thus proving the effectiveness of the proposed method.

Claims

1. An adaptive robust control method for an electro-hydraulic brake-by-wire system based on a neural network, characterized in that: The steps include: S1. Establish a mathematical model of the electro-hydraulic brake-by-wire system; S2. Construct a neural network observer, design a neural network weight adaptation law based on the Lyapunov function, use the observed system state to design the control law, and prove the stability based on the Lyapunov function; S3. Construct an adaptive robust controller based on the backstepping method, design the adaptive laws of each parameter, and achieve asymptotically stable system tracking through virtual control laws and state variables.

2. The method for adaptive robust control of an electro-hydraulic brake-by-wire system based on a neural network according to claim 1, characterized in that: In step S1, the pressure regulation part of the wire control brake system adopts a three-position three-way electromagnetic direct-drive valve. The main driving part of the electromagnetic direct-drive valve is the electromagnetic linear actuator. By controlling the displacement of the valve core, the wheel cylinder brake pressure is controlled. The hydraulic execution part of the electronic hydraulic brake system also includes a brake wheel cylinder. The mathematical model of the electro-hydraulic wire control brake system is constructed based on the electromagnetic linear actuator, steady-state hydraulic pressure, brake wheel cylinder motion equation and flow equation.

3. The method for adaptive robust control of an electro-hydraulic brake-by-wire system based on a neural network according to claim 2, characterized in that: Step S2 specifically includes: S21, constructing a neural network observer to feedforward the observed system state to the control system, wherein the valve core displacement, speed, hydraulic pressure and hydraulic pressure change speed are used as state variables in the neural network observer, and a three-layer neural network is used to fit the total disturbance of the system with the output of the neural network when the neural network observer observes the total disturbance; S22, constructing a neural network weight adaptive law to estimate the ideal weight; S23. Construct Lyapunov function to prove stability and verify the mathematical stability of observer and adaptive law.

4. The method for adaptive robust control of an electro-hydraulic brake-by-wire system based on a neural network according to claim 3, characterized in that: Step S3 specifically includes: S31, design a virtual control law based on the backstepping method, coupled with the disturbance estimate and the observed state; S32. An adaptive robust controller is constructed using Lyapunov function combined with parameter adaptation rate.

5. The method for adaptive robust control of an electro-hydraulic brake-by-wire system based on a neural network according to claim 1, characterized in that: The digital model expression of the electro-hydraulic brake-by-wire system is as follows: ， in, is the cross-sectional area of ​​the piston, is the brake fluid pressure, is the mass of the piston, is the displacement of the wheel cylinder piston, is the piston equivalent damping coefficient, is the initial length of the spring, is the spring stiffness in the piston cylinder, is the brake fluid flow rate, is the bulk elastic modulus of the liquid, is the volume of the brake wheel cylinder, is the pressure flow coefficient including leakage, is the total mass including the electromagnetic linear actuator, valve core, connectors, and springs. is the valve core movement speed, is the input voltage, is the motor constant of the electromagnetic actuator, is the displacement of the valve core, is the coil resistance, is the damping coefficient, is the spring stiffness, is the friction force, is the steady-state fluid pressure, is the transient fluid pressure.

6. The method for adaptive robust control of an electro-hydraulic brake-by-wire system based on a neural network according to claim 5, characterized in that: The steady-state hydraulic pressure is calculated as follows: ， The transient hydraulic pressure is calculated as follows: ， In the formula, is the valve port area gradient, is the valve port pressure difference, is the flow coefficient, is the actual flow length of the hydraulic oil in the valve chamber, is the oil density.

7. The method for adaptive robust control of an electro-hydraulic brake-by-wire system based on a neural network according to claim 5, characterized in that: The neural network observer expression is as follows: ， In the formula, , indicating the brake fluid pressure; , indicating the speed of change of hydraulic pressure; , represents the displacement of the valve core, , indicating the valve core movement speed, ; ; , is the oil source pressure, , , , , , is the observer positive gain, - is the system parameter, For Related interference, For Related interference; The weight adaptation law expression is as follows: ， In the formula, , Respectively Related to The relevant weight adaptation law, , Respectively Related to The related weight adaptive law gain coefficient, is with Radial basis function ensemble estimates in correlated neural networks, is with Radial basis function ensemble estimates in correlated neural networks, ; .

8. The method for adaptive robust control of an electro-hydraulic brake-by-wire system based on a neural network according to claim 7, characterized in that: The controller expression is as follows: ， The parameter adaptation law expression is as follows: ， in, is a given positive number, , , , , , They are respectively , , The virtual control rate defined, is a defined positive number, , .

Images