Satellite attitude super-agile scanning neural network minimum parameter learning sliding mode control method

Abstract

The application discloses a satellite attitude super-agile wobble scanning neural network minimum parameter learning sliding mode control method and belongs to the technical field of micro-satellite attitude control. The application discloses a satellite attitude super-agile wobble scanning neural network minimum parameter learning sliding mode control method and belongs to the technical field of micro-satellite attitude control. The method is used for converging the system state to zero within a limited time through the fast terminal sliding mode control, and the switching term in the sliding mode is approximated through the radial basis function neural network, so that the purpose of weakening the chattering is achieved. Meanwhile, the minimum parameter learning method is designed to replace the weight value adjustment of the radial basis function neural network, the weight value matrix is prevented from being operated in a large amount, the controller efficiency is effectively improved, and the method has the characteristics of good convergence, imaging and control effect.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of micro-satellite attitude control, and particularly relates to a satellite attitude super-agile scanning neural network minimum parameter learning sliding mode control method. BACKGROUND

[0002] A high-performance and low-cost micro-satellite platform is one of the future directions of space development. As the core part of the micro-satellite, the attitude control system is the most important part in the development of the universal commercial satellite platform, and at the same time, the attitude control system is also the part with the most faults. Therefore, it is very important to improve the reliability of the attitude and orbit control subsystem as much as possible under the constraints of the volume, mass, power and cost of the small satellite.

[0003] A magnetic torque device is a commonly used attitude control actuator. The magnetic torque device uses the electric energy on the satellite to generate a dipole magnetic moment through the coil on the magnetic rod, and the geomagnetic field and the generated magnetic moment interact to generate a magnetic control torque to control the satellite attitude. It has the advantages of not consuming working medium, low power consumption, low cost, small volume and mass, and has been applied to about two-thirds of the satellites. The satellite will be affected by various torques on the orbit, which will affect the attitude of the satellite and further affect the normal work of the satellite. Since the magnetic torque device needs to rely on the geomagnetic field to generate a control torque, only the magnetic torque device is used for active control, and the three-axis of the satellite cannot be controlled in real time, so the control results of different control algorithms are very different. How to design a good control algorithm is the main problem faced by the satellite active magnetic control.

[0004] Therefore, in the design, in order to ensure that the low-orbit micro-satellite attitude super-agile scanning maneuvering high-quality imaging can be seamlessly spliced to realize the non-missing detection of the regional target, the satellite attitude controller is required to realize high-precision tracking and fast error convergence of the planned attitude trajectory, and at the same time, the ability to stably point to the target is required. At present, there are few studies on the attitude super-agile maneuvering control of micro-satellites, and in the studies on the attitude control algorithm for other application scenarios, only one or two of the tracking accuracy, convergence speed and stable pointing are generally concerned, which is difficult to meet the control requirements of the attitude super-agile maneuvering stable imaging. For the application scenario of the present application, in order to realize the super-agile attitude maneuvering of the micro-satellite and ensure the tracking imaging quality, the attitude control algorithm is required to have the abilities of high-precision tracking, fast convergence and stable pointing at the same time.

[0005] Therefore, in view of the above problems, a satellite attitude super-agile scanning neural network minimum parameter learning sliding mode control method is urgently needed to solve the problems existing in the prior art. SUMMARY

[0006] In view of the above problems, the present application aims to provide a satellite attitude super-agile wobble neural network minimum parameter learning sliding mode control method, which uses fast terminal sliding mode control to make the system state converge to zero in a limited time, and uses a radial basis function neural network to approximate the switching term in the sliding mode, thereby achieving the purpose of weakening chattering; at the same time, a minimum parameter learning method is designed to replace the weight adjustment of the radial basis function neural network, avoiding a large amount of weight matrix calculation, effectively improving the efficiency of the controller, and having the characteristics of good convergence, imaging and control effect.

[0007] In order to achieve the above purpose, the technical scheme adopted by the present application is as follows:

[0008] A satellite attitude super-agile wobble neural network minimum parameter learning sliding mode control method, comprising the steps of

[0009] S1. In the process of satellite control, an actual output model of the neural network is established;

[0010] S2. On the basis of the actual output model of the neural network, a minimum parameter learning method is introduced to replace the weight adjustment of the radial basis function neural network, and the calculation of the neural network weight matrix W is converted into the solution of a single parameter , thereby realizing the approximation of the sign function term sgn(s) in the sliding mode controller.

[0011] Further, the process of establishing the actual output model of the radial basis function neural network in step S1 comprises

[0012] (1) In the process of satellite control, the sign function estimation value of the sliding mode surface s output of the sliding mode controller is:

[0013] u sgn =ρf(x) (1)

[0014] Wherein, f(x) is the ideal output of the sign term approximated by the radial basis function neural network;

[0015] (2) has

[0016] f=W *T h(x)+e (2)

[0017] Wherein, x=[q ev ω e ] T is the input of the neural network, h represents the Gaussian kernel function, W * represents the ideal value of the weight from the hidden layer to the output layer, and e represents the approximation error of the neural network;

[0018] (3) the actual output of the neural network is:

[0019]

[0020] wherein, are the estimated weights of the neural network.

[0021] Further, the step S2 of introducing the least parameter learning method instead of the weight adjustment process of the radial basis function neural network based on the actual output model of the neural network comprises

[0022] (1) According to the least parameter learning method of the neural network, let φ is a constant, and φ > 0, let φ represents the estimation of φ, and the error is:

[0023]

[0024] (2) The new control law after introducing the radial basis function neural network based on the least parameter learning method is:

[0025]

[0026] The role of the robust term Ksgn(s) in the control law is to overcome the disturbance and the neural network approximation error to ensure the stability of the system

[0027]

[0028] In the formula, W is the weight matrix of the neural network, h is the Gaussian kernel function, e is the neural network approximation error, I 3×3 is a 3x3 unit matrix, α, β, g, h, m, n are all sliding mode surface design parameters, q ev is an error quaternion, ω e is an error angular velocity, J is a satellite moment of inertia, ω b is a satellite body rotation angular velocity, ω d is a desired angular velocity, H CMG is a control moment gyro (CMG) angular momentum, is a desired coordinate system to body coordinate system conversion matrix, u is a control input, d0 is a disturbance torque.

[0029] Further, the sliding mode control method further comprises a step S3 of proving the stability of the new control law after the radial basis function neural network based on the least parameter learning method.

[0030] Further, the step S3 of proving the stability of the new control law after the radial basis function neural network based on the least parameter learning method comprises

[0031] S31. Define a Lyapunov function

[0032]

[0033] wherein, gamma > 0;

[0034] Derivation of V, obtains:

[0035]

[0036] S32. Substituting the control law of formula (1) into formula (5), obtains:

[0037]

[0038] Derivation obtains:

[0039]

[0040] Then obtains:

[0041]

[0042] Wherein, Derivation knows Take adaptive law as:

[0043]

[0044] In the formula, kappa > 0;

[0045] There is:

[0046]

[0047] Take Substitute into formula (8), there is

[0048]

[0049] In the formula,

[0050] Solve the inequality:

[0051]

[0052]

[0053] Therefore, the convergence accuracy of s is determined by phi, gamma and eta jointly.

[0054] The beneficial effects of the present application are: the present application discloses a satellite attitude super-agile swing-scan neural network minimum parameter learning sliding mode control method, compared with prior art, the improvement of the present application lies in:

[0055] 1. This invention designs a satellite attitude ultra-agile sweeping neural network minimum parameter learning sliding mode control method. This method uses minimum parameter learning to adaptively estimate parameters instead of adjusting the weights of the radial basis function neural network, transforming the calculation of the neural network weight matrix W into the calculation of individual parameters. The adaptive solution method is simpler, faster, and more efficient, and it can also achieve the best approximation effect of radial basis function neural network. It is also beneficial to apply it to actual satellite control systems.

[0056] 2. This method utilizes fast terminal sliding mode control to converge the system state to zero within a finite time, and uses a radial basis function neural network to approximate the switching term in the sliding mode, thereby reducing chattering.

[0057] 3. This method replaces the weight adjustment of the radial basis function neural network with a minimum parameter learning method, avoiding a large number of calculations of the weight matrix, effectively improving the controller efficiency, and has the advantages of good convergence, imaging and control effects. Attached Figure Description

[0058] Figure 1 This is a block diagram of the neural sliding mode control based on minimum parameter learning in this invention.

[0059] Figure 2 This is a comparison diagram of the expected attitude quaternion and the radial basis function neural sliding mode controller tracking attitude quaternion of the present invention.

[0060] Figure 3 This is a graph showing the tracking attitude error results of the radial basis function neural sliding mode controller of the present invention.

[0061] Figure 4 The diagram shows the tracking angular velocity results of the radial basis function neural sliding mode control of this invention. Detailed Implementation

[0062] To enable those skilled in the art to better understand the technical solutions of the present invention, the technical solutions of the present invention will be further described below in conjunction with the accompanying drawings and embodiments.

[0063] Example 1: Refer to Appendix Figures 1-4 This paper presents a satellite attitude ultra-agile sweeping neural network minimum parameter learning sliding mode control method. This method utilizes the characteristic that radial basis function neural networks can infinitely approximate any nonlinear function. A radial basis function neural network is used to approximate the sign function term sgn(s) in the sliding mode controller, thereby ensuring a reduction in the sliding surface switching gain and weakening the chattering of the sliding mode controller. The method includes the following steps:

[0064] S1. Establish the actual output model of the radial basis function neural network during satellite control.

[0065] The sign function estimation value of the sliding mode controller sliding mode surface s output is:

[0066] u sgn = p f(x) (1)

[0067] Where f(x) is the ideal output of the sign term approximated by the radial basis function neural network, then:

[0068] f = W *T h(x) + e (2)

[0069] Where x = [q ev ω e ] T is the input of the neural network, h represents the Gaussian kernel function, W * represents the ideal value of the hidden layer to the output layer weight, and e represents the approximation error of the neural network;

[0070] The actual output of the neural network is:

[0071]

[0072] Where is the estimated weight of the neural network;

[0073] S2. On the basis of the actual output model of the neural network, the minimum parameter learning method is introduced to replace the weight adjustment of the radial basis function neural network, and the calculation of the neural network weight matrix W is converted into the solution of a single parameter

[0074] According to the minimum parameter learning method of the neural network, let φ is a constant, and φ > 0, let represent the estimation of φ, then the error can be represented as:

[0075]

[0076] The new control law after introducing the radial basis function neural network based on the minimum parameter learning method can be represented as:

[0077]

[0078] The role of the robust term Ksgn(s) in the control law is to overcome the disturbance and the neural network approximation error to ensure the stability of the system, where

[0079]

[0080] S3. The stability of the new control law based on the minimum parameter learning method radial basis function neural network shown in equation (6) is proved​

[0081] S31. Define Lyapunov function

[0082]

[0083] where γ > 0;

[0084] Taking the derivative of V, we have:

[0085]

[0086] S32. Substitute the control law of formula (1) into formula (5), we have:

[0087]

[0088] It can be deduced that:

[0089]

[0090] Then we can get:

[0091]

[0092] where, and the derivation can be known

[0093] Taking adaptive law as:

[0094]

[0095] In the formula, κ > 0.

[0096] Therefore, we have:

[0097]

[0098] Taking Substitute into formula (8), then we have

[0099]

[0100] In the formula,

[0101] Solve the inequality:

[0102]

[0103] From this, we can see that the convergence accuracy of s is determined by φ, γ and η together.

[0104] Example 2: Unlike the example, in order to verify the effectiveness of the control algorithm as proposed in Example 1, this example is designed to simulate the micro-satellite attitude hypersensitive swing-scan sliding mode control method based on the minimum parameter learning adaptive radial basis function neural network proposed in Example 1. The simulation basic conditions are set as follows, Table 1 is the satellite parameter table, and Table 2 is the controller parameter table.

[0105] Table 1: Satellite and orbit parameters in radial basis function sliding mode control

[0106]

[0107] Table 2: Radial basis function sliding mode controller parameter table

[0108]

[0109] The simulation results are shown in Figures 2-4 .

[0110] By simulating the micro-satellite attitude hypersensitive swing-scan sliding mode control method based on the minimum parameter learning adaptive radial basis function neural network proposed in this patent, Figure 2 Figures (a) and (b) respectively show the planned expected attitude quaternion and the attitude quaternion obtained by tracking the expected attitude under the control algorithm proposed in this patent, and the expected attitude and the tracking attitude curve still maintain the same change trend.

[0111] By calculating the error between the expected attitude quaternion and the attitude tracking quaternion, we can get Figure 3 , Figure (a) is the attitude tracking error represented by Euler angle, and Figure (b) is the attitude tracking error represented by quaternion. The maximum attitude angle error on the three axes is 0.009°, 0.008° and 0.009°, respectively, and the error accuracy is basically around 10 -3 orders of magnitude. Figure 4 Figure (a) shows the angular velocity tracking error results of the micro-satellite attitude hypersensitive swing-scan sliding mode control method based on the minimum parameter learning adaptive radial basis function neural network. The maximum angular velocity error on the three axes is 0.14° / s, 0.17° / s and 0.17° / s, respectively. The simulation results verify that the algorithm proposed in Example 1 of the present application can meet the control requirements of micro-satellite attitude hypersensitive maneuvering stable imaging.

[0112] The above shows and describes the basic principles, main features and advantages of the present application. Those skilled in the art should understand that the present application is not limited to the above-mentioned embodiments, and the above-mentioned embodiments and descriptions in the specification are only to illustrate the principles of the present application. Without departing from the spirit and scope of the present application, various changes and improvements can be made to the present application, and these changes and improvements all fall within the scope of the claimed present application. The scope of protection of the present application is defined by the appended claims and their equivalents.

Claims

1. A satellite attitude hyperswift wobble neural network minimum parameter learning sliding mode control method, characterized in that: Comprising the steps of S1. establishing an actual output model of a radial basis function neural network in a satellite control process; S2. On the basis of the actual output model of the neural network, the minimum parameter learning method is introduced to replace the weight adjustment of the radial basis function neural network, which converts the calculation of the neural network weight matrix to the solution of a single parameter , and realizes the approximation of the sign function item in the sliding mode controller; The process of establishing the actual output model of the radial basis function neural network in step S1 comprises (1) In the process of satellite control, the sliding mode controller is set to the sliding mode surface The output sign function estimate value is: （1） wherein, is the ideal output of the symbolic term approximated by a radial basis function neural network; (2) there are （2） wherein, is an input to the neural network, denotes a Gaussian kernel function, denotes an ideal value of the hidden layer to output layer weight, denotes an error of the neural network approximation; (3) the actual output of the neural network is obtained as: （3） wherein, are estimated weights of the neural network; The process of introducing the minimum parameter learning method to replace the weight adjustment of the radial basis function neural network on the basis of the actual output model of the neural network in step S2 comprises (1) According to the minimum parameter learning method of neural network, let , be a constant, and , let represent the estimate of , and the error is: （5） (2) the new control law of the radial basis function neural network based on the minimum parameter learning method is: （6） Robust term in control law The role is to overcome the interference and neural network approximation error, to ensure system stability （7）； In the formula, W is a weight matrix of a neural network, h is a Gaussian kernel function, e is a neural network approximation error, I 3×3 is a 3×3 unit matrix, , , g, h, m, n are all sliding mode surface design parameters, is an error quaternion, is an error angular velocity, is a satellite rotational inertia, is a satellite body rotation angular velocity, is a desired angular velocity, is a control moment gyro (CMG) angular momentum, is a desired coordinate system to body coordinate system conversion matrix, is a control input, is a disturbance torque.

2. The neural network minimum parameter learning sliding mode control method for satellite attitude hyperswift scan according to claim 1, characterized in that: The sliding mode control method further comprises a process of proving the stability of the new control law of the radial basis function neural network based on the minimum parameter learning method in step S3.

3. The neural network minimum parameter learning sliding mode control method for satellite attitude hyperswift scan according to claim 2, characterized in that: The process of proving the stability of the new control law of the radial basis function neural network based on the minimum parameter learning method in step S3 comprises S31. defining a Lyapunov function （8） wherein ; Deriving V, we have: (9) S32. substituting equation (5) into equation (9), we have: （10） It is concluded that: (11) Then we get: (12) wherein , it is derived that ; Taking the adaptive law as: (13) In the formulae, ; There are: (14) Take Substituting into equation (8), we have (15) In the formulae, ; Solving the inequality: (16) (17) Thereby, The convergence accuracy is determined by and together.

Images