CN113219832A - Design method of adaptive iterative learning non-uniform target tracking controller - Google Patents
Design method of adaptive iterative learning non-uniform target tracking controller Download PDFInfo
- Publication number
- CN113219832A CN113219832A CN202110507272.0A CN202110507272A CN113219832A CN 113219832 A CN113219832 A CN 113219832A CN 202110507272 A CN202110507272 A CN 202110507272A CN 113219832 A CN113219832 A CN 113219832A
- Authority
- CN
- China
- Prior art keywords
- equation
- function
- error
- substances
- following
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 238000000034 method Methods 0.000 title claims abstract description 53
- 230000003044 adaptive effect Effects 0.000 title claims abstract description 35
- 230000006870 function Effects 0.000 claims abstract description 76
- 238000013528 artificial neural network Methods 0.000 claims abstract description 27
- 239000000126 substance Substances 0.000 claims description 39
- 238000012886 linear function Methods 0.000 claims description 7
- 230000008569 process Effects 0.000 claims description 7
- 230000003247 decreasing effect Effects 0.000 claims description 3
- 241000238876 Acari Species 0.000 claims 1
- 238000004088 simulation Methods 0.000 abstract description 20
- 238000011160 research Methods 0.000 description 5
- 230000008859 change Effects 0.000 description 4
- 230000006978 adaptation Effects 0.000 description 2
- 238000000354 decomposition reaction Methods 0.000 description 2
- 210000002569 neuron Anatomy 0.000 description 2
- 101100127891 Caenorhabditis elegans let-4 gene Proteins 0.000 description 1
- 241000764238 Isis Species 0.000 description 1
- 238000012938 design process Methods 0.000 description 1
- 238000006073 displacement reaction Methods 0.000 description 1
- 210000004964 innate lymphoid cell Anatomy 0.000 description 1
- 239000011159 matrix material Substances 0.000 description 1
- 230000007246 mechanism Effects 0.000 description 1
- 230000001537 neural effect Effects 0.000 description 1
- 208000011580 syndromic disease Diseases 0.000 description 1
- 229910052723 transition metal Inorganic materials 0.000 description 1
- 150000003624 transition metals Chemical class 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G05—CONTROLLING; REGULATING
- G05B—CONTROL OR REGULATING SYSTEMS IN GENERAL; FUNCTIONAL ELEMENTS OF SUCH SYSTEMS; MONITORING OR TESTING ARRANGEMENTS FOR SUCH SYSTEMS OR ELEMENTS
- G05B13/00—Adaptive control systems, i.e. systems automatically adjusting themselves to have a performance which is optimum according to some preassigned criterion
- G05B13/02—Adaptive control systems, i.e. systems automatically adjusting themselves to have a performance which is optimum according to some preassigned criterion electric
- G05B13/04—Adaptive control systems, i.e. systems automatically adjusting themselves to have a performance which is optimum according to some preassigned criterion electric involving the use of models or simulators
- G05B13/042—Adaptive control systems, i.e. systems automatically adjusting themselves to have a performance which is optimum according to some preassigned criterion electric involving the use of models or simulators in which a parameter or coefficient is automatically adjusted to optimise the performance
Landscapes
- Engineering & Computer Science (AREA)
- Health & Medical Sciences (AREA)
- Artificial Intelligence (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Computation (AREA)
- Medical Informatics (AREA)
- Software Systems (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Automation & Control Theory (AREA)
- Feedback Control In General (AREA)
Abstract
The invention discloses a design method of a self-adaptive iterative learning non-uniform target tracking controller, which specifically comprises the following steps: the method comprises the following steps: step 1: performing system description aiming at a type of continuous time nonlinear system; step 2: describing a time-varying boundary layer and a radial basis function neural network; and step 3: designing a self-adaptive learning controller by respectively applying a Backstepping method and a radial basis function neural network in the conditions of asymmetric dead zone input and continuous nonlinear input; and 4, step 4: analyzing the stability and convergence of the designed adaptive learning controller; and 5: a simulation example is given to illustrate the feasibility and effectiveness of the method of the present invention. When the initial state error and the unknown input nonlinearity exist simultaneously, the problem of inconsistent target tracking control of self-adaptive iterative learning of a strict feedback nonlinear system is solved.
Description
Technical Field
The invention belongs to the technical field of adaptive iterative learning control, and relates to a design method of an adaptive iterative learning non-uniform target tracking controller.
Background
The self-adaptive iterative learning control method is a powerful method for solving the problem of trajectory tracking in a finite time interval of a nonlinear system, and the self-adaptive iterative learning control is combined with iterative learning control. According to the design of the adaptive learning law, the AILC theory can be divided into two main categories: along the iteration domain direction and the time domain direction, are also called discrete AILC and continuous AILC, of course. Adaptive control is often used to deal with uncertainties that can be estimated by learning, and most adaptive controllers involve some type of function approximator, such as Neural Networks (NNs) and Fuzzy Logic Systems (FLSs), in their learning mechanism. The self-adaptive NN and FLS controllers can solve the problem of stability of closed-loop systems of various control systems, and the problem of tracking variable control tasks is solved by applying the self-adaptive fuzzy ILC to an unknown nonlinear system. The AILC is designed with a variable wavelet decomposition, and the wavelet decomposition terms are increased with the increase of the iteration number, so that the tracking error is converged uniformly in a limited time interval. The AILC based on a complex energy function or a Lyapunov function plays an important role in processing the learning estimation of time-varying parameters by connecting the time domain and the iterative domain. The AILC of a non-linear parameterized system remains a disclosed problem. No research efforts have been found for the aicc of non-linear parametric systems. Although traditional D-type and P-type ILCs have good research results on this problem, it remains a public problem for AILC. Due to the physical limitations of actuators, non-linearities are always present in control inputs, such as dead band inputs and continuous non-linear inputs. The presence of non-linearities in the control input tends to undermine system performance. At present, the performance of such control systems is improved by combining different methods for some nonlinear systems such as the tracking control problem of nonlinear uncertain systems with input saturation, very general nonlinear input uncertainty, class of time-lag systems with matched nonlinear time-lag disturbance and asymmetric dead zone saturation input, and uncertain nonlinear parameterized systems with unknown input nonlinearity (NLP-systems). However, there is currently no mature research effort for AILC for nonlinear systems with unknown input nonlinearities and initial state errors. When a possibly large initial state error exists, how to use the adaptive iterative learning control to solve the problem of tracking the non-uniform target track of a nonlinear strict feedback system on [0, T ] is a challenging problem.
Disclosure of Invention
The invention aims to provide a design method of a self-adaptive iterative learning non-uniform target tracking controller, which can solve the problem of self-adaptive iterative learning non-uniform target tracking control of a strict feedback non-linear system when an initial state error and unknown input non-linearity exist simultaneously.
The technical scheme adopted by the invention is that a design method of a self-adaptive iterative learning non-uniform target tracking controller specifically comprises the following steps:
wherein the content of the first and second substances,Γ(uk) E R represents the input-output characteristics of the actuator, ykE, R refers to system output;is a smooth unknown non-linear function; k represents an iteration index;
step 2, the time-varying boundary layer network is expressed by the following formula (2):
wherein phi isi,k(t)=εi,ke-ηt,zi,kAnd ziφ,kIs a function variable of time t, epsiloni,kIs a convergence series sequence, sat is a saturation function, defined as follows:
wherein phi isi,k(t) is a time-varying boundary layer,. phii,k(t) decreasing along the time axis, selecting an initial state error phi at the kth iterationi,k(0)=εi,kThen 0 < epsiloni,ke-ηT≤φi,k(t)≤εi,k,
And 3, expressing the radial basis function neural network by adopting the following formula (4):
wherein the content of the first and second substances,is a known smooth vector function, wherein the number l of NN nodes is more than 1; radial basis functionIs a gaussian function and is expressed by the following equation (5):
wherein, mujE.omega and eta > 0 are radial basis functions, respectivelyOf (2) centerSum width, optimal weight vector W ═ ω1,...,ωl]TIs defined as:
wherein the content of the first and second substances,is an approximation error inherent to the NN;
and 4, designing the self-adaptive learning controller by respectively applying a Backstepping method and a radial basis function neural network in the conditions of asymmetric dead zone input and continuous nonlinear input.
The invention is also characterized in that:
the specific process of the step 4 is as follows:
step 4.1, use function Γ (u)k) The actuator output with an asymmetric dead band is represented as shown in equation (7) below:
wherein the parameter mrAnd mlA right slope and a left slope representing the dead zone characteristic, respectively, parameter brAnd blA breakpoint representing a dead-zone input of the actuator;
and 4.2, modeling the asymmetric dead zone of the actuator into a form of the sum of a straight line and a disturbance-like term, wherein the form is shown in the following formula (8):
Γ(uk)=m(t)uk+d(t) (8);
4.3, designing a controller for the uncertain nonlinear system with unknown asymmetric dead zone input and initial state error based on a self-adaptive iterative learning Backstepping frame;
and (4) performing a step (4.4),in the control input of a continuous-time nonlinear system of the type represented by equation (1), there is a nonlinearity which is assumed in the sector region s1 s2]A continuous non-linear function N (u)k) And N (0) ═ 0, where s is1And s2Represents two straight lines (t)1,t2) Of (2) i.e.
And 4.5, designing a controller for the condition of continuous nonlinear input.
The specific process of the step 4.3 is as follows:
step A) let z1,k=x1,k-yr,k,z2,k=x2,k-α1,kIn which α is1,kIs a virtual controller; because of the initial state error, according to the step 2 of representing the time-varying boundary layer, the following formula is adopted to respectively carry out the error function z1φ,kAnd z2φ,kTo show that:
wherein the content of the first and second substances,ε1,kand ε2,kIs a positive series of convergent numbers, η1And η2Is a designed normal number;
then z is1φ,kThe derivative with respect to time is shown in the following equation (13):
step B) according to the representation of the radial basis function neural network in the step 3, the uniform approximation performance of the RBFNN shows that the unknown nonlinear function isApproximated by RBFNN on a tight set to the following form, and with reconstruction error
Wherein the content of the first and second substances,is the approximation error, W1Is the optimal weight vector; for any real number a > 0 and positive integer l ≧ 2, takeDefinition ofThe virtual controller is taken as:
substituting equations (14) and (15) into (13) yields the following equation (16):
wherein the content of the first and second substances,andare respectively a parameter W1And N1Is estimated by the estimation of (a) a,andis the parameter estimation error, the last two terms to the right of the equal sign of equation (16) are rewritten as:
on the basis of the formula (17), the formula (16) is rewritten as the following formula (18):
in step 4.3, define: z is a radical ofn,k=xn,k-αn-1,kAccording to the step 2 representation of the time-varying boundary layer network, the following error function z is introducednφ,k:
Wherein the content of the first and second substances,εn,kis a positive series of convergent numbers, ηnIs a designed normal number;
then z isnφ,kThe derivative of (c) is:
znφ,kthe derivative with respect to time is:
according to the representation of the radial basis function neural network in the step 3, the RBFNN unified approximation performance shows that the unknown nonlinear function is representedApproximated by RBFNN on a tight set to the following form, and with reconstruction error
Wherein the content of the first and second substances,is the approximation error, WnIs the optimal weight vector;
uk=u1,k+u2,k (26);
Wherein u is2,kTo compensate for the unknown input gain m (t); substituting equations (25) and (26) into (24) yields the following equation (27):
wherein the content of the first and second substances,andare respectively a parameter WnAnd NnIs estimated by the estimation of (a) a,andis the parameter estimation error, the last two terms to the right of the equality sign of equation (27) are rewritten as:
the specific process of the step 4.5 is as follows: with N (u)k) As the output of the system instead of Γ (u)k) Definition of zn,k=xn,k-αn-1,kAccording to the step 2 representation of the time-varying boundary layer network, the following error function z is introducednφ,k:
Wherein the content of the first and second substances,εn,kis a positive series of convergent numbers, ηnIs a designed normal number; then z isnφ,kIs a derivative of
then equation (32) is rewritten as equation (36) below:
according to the representation of the radial basis function neural network in the step 3, the uniform approximation performance of the RBFNN shows that the unknown nonlinear function isApproximated by RBFNN on a tight set to the following form, and with reconstruction error
Wherein the content of the first and second substances,is the approximation error, WnIs the optimal weight vector; according to the formulas (36) and (37), the following formula (38) is obtained:
the method for designing the self-adaptive iterative learning non-uniform target tracking controller has the advantages that the radial basis function neural network is introduced to learn the performance of unknown dynamics, and a typical series is used for effectively eliminating approximation errors. The problem of initial state errors is solved by applying a time-varying boundary layer. The problem of non-linearity of two types of input, namely asymmetric dead zone input and continuous input is effectively solved. It can be shown that all signals of the closed loop system are bounded over a given time interval 0, T and that the state tracking error will asymptotically converge to an adjustable set of residuals as the iteration goes to infinity.
Drawings
FIG. 1 shows the I/Z for asymmetric dead zone input in simulation research of the design method of adaptive iterative learning inconsistent target tracking controller1,kA graph of | | change with iteration index;
FIG. 2 is | | u for asymmetric dead zone input condition in simulation research of the design method of adaptive iterative learning inconsistent target tracking controller of the present inventionkA graph of | | change with iteration index;
FIG. 3 is a graph of asymmetric dead zone input conditions in simulation studies of a method for designing a non-uniform target tracking controller for adaptive iterative learning according to the present inventionA graph of variation with iteration index;
FIG. 4 shows asymmetric dead-zone input conditions in simulation studies of a method for designing a non-uniform target tracking controller according to the present inventionA graph of variation with iteration index;
FIG. 5 shows asymmetric dead-zone input conditions in simulation studies of a method for designing a non-uniform target tracking controller according to the present inventionA graph of variation with iteration index;
FIG. 6 is a graph of asymmetric dead zone input conditions in simulation studies of a method for designing a non-uniform target tracking controller for adaptive iterative learning according to the present inventionA graph of variation with iteration index;
FIG. 7 shows asymmetric dead-zone input conditions in simulation studies of a method for designing a non-uniform target tracking controller according to the present inventionA graph of variation with iteration index;
FIG. 8 is a plot of z for continuous nonlinear input conditions in simulation studies of a method of designing a non-uniform target tracking controller for adaptive iterative learning according to the present invention1,kA graph of | | change with iteration index;
FIG. 9 is a graph of | | | u for continuous nonlinear input conditions in simulation studies of a method for designing a non-uniform target tracking controller for adaptive iterative learning according to the present inventionkA graph of | | change with iteration index;
FIG. 10 is a graph of continuous nonlinear input conditions in simulation studies of a method of designing a non-uniform target tracking controller for adaptive iterative learning in accordance with the present inventionA graph of variation with iteration index;
FIG. 11 is a graph of continuous nonlinear input conditions in simulation studies of a method of designing a non-uniform target tracking controller for adaptive iterative learning in accordance with the present inventionA graph of variation with iteration index;
FIG. 12 is a graph of continuous nonlinear input conditions in simulation studies of a method of designing a non-uniform target tracking controller for adaptive iterative learning in accordance with the present inventionA graph of variation with iteration index;
FIG. 13 is a graph of continuous nonlinear input conditions in simulation studies of a method of designing a non-uniform target tracking controller for adaptive iterative learning in accordance with the present inventionA graph of variation with iteration index;
Detailed Description
The present invention will be described in detail below with reference to the accompanying drawings and specific embodiments.
The invention relates to a design method of a self-adaptive iterative learning non-uniform target tracking controller, which specifically comprises the following steps: step 1, carrying out system description on a continuous time nonlinear system; consider a class of continuous-time nonlinear systems, as shown in equation (1) below:
wherein the content of the first and second substances,is the state vector of the system that can be tested. Gamma (u)k) E R represents the input-output characteristics of the actuator, ykE R is the system output.Is a smooth unknown non-linear function. k denotes an iteration index. The target is at [0, T]Upper design adaptive iterative learning control law uk(t) such that the system output y is repeatedk(t) tracking the upper target trajectory yr,k(t) and for some small positive error bound δAnd all signals of the closed loop system are guaranteed to be bounded. y isr,k(t) represents a sufficiently smooth target trajectory.
Step 2, a time-varying boundary layer and a radial basis function neural networkThe description is carried out; 1) a time-varying boundary layer; to overcome the uncertainty of the initial state error, a new function z is introduced as followsiφ,k:
Wherein z isi,kAnd ziφ,kIs a function variable of time t, epsiloni,kIs a convergence series sequence, sat is a saturation function, defined as follows:
wherein phi isi,k(t) is a time-varying boundary layer. Notice phii,k(t) decreasing along the time axis, selecting an initial state error phi at the kth iterationi,k(0)=εi,kThen 0 < epsiloni,ke-ηT≤φi,k(t)≤εi,k,
Assuming 1, since the initial state error for each iteration of the initial state error is not necessarily zero, small and fixed, epsilon is constant for some known normality i,k1, m, assuming that the initial state error satisfies | zi,k(0)||≤εi,kWherein z isi,k(0) Is the initial state error of the state error at each iteration. Since z is easily obtained from equation (2) if it is assumed that 1 holdsiφ,k(0)=0,During the whole time interval [0, T]When η is increased properly, phii,k(t) will be as small as possible. If can prove thatThen the error function will satisfy that when k → ∞, whereinIs an arbitrarily small normal number, i.e., the control target is completed.
2) A radial basis function neural network; unknown smooth non-linear functionThe approximation on the tight set Ω will be done by the following Radial Basis Function Neural Network (RBFNN):
wherein the content of the first and second substances,is a known smooth vector function, where the number of NN nodes, l > 1. Radial basis functionIs selected as a commonly used gaussian function andwherein mujE.omega and eta > 0 are radial basis functions, respectivelyThe center and the width of (c). The optimal weight vector W ═ ω1,...,ωl]TIs defined as:
wherein the content of the first and second substances,is an approximation error inherent to the NN, which can be arbitrarily reduced by increasing the value of the NN network node number l. Following approximation errorThe following assumptions are made.
Suppose 2, on tight set Ω, the error is approximatedIs assumed to be bounded andwherein the unknown parameter thetai(1. ltoreq. i. ltoreq.n) representsA minimum upper bound of where θi≥0。
And 3, respectively applying a Backstepping method and a radial basis function neural network to design the self-adaptive learning controller under the conditions of asymmetric dead zone input and continuous nonlinear input. Firstly, asymmetric dead zone input and continuous nonlinear input are described, and then corresponding controllers are respectively designed. 1) Asymmetric dead zone input; actuator dead band is common in mechanical connections, hydraulic servo valves, piezoelectric sensors, and electric servomotors. The presence of such non-linearities often undermines system performance. Step 3 will introduce an asymmetric dead zone input characteristic. Function Γ (u)k) Representing actuator output with asymmetric dead band, as expressed below
Wherein the parameter mrAnd mlRespectively representing the right and left slopes of the dead zone characteristic. Parameter brAnd blRepresenting a breakpoint of the actuator dead band input. The asymmetric dead zone of the actuator can be modeled as a straight line sumThe sum of one class perturbation term is in the form of:
Γ(uk)=m(t)uk+d(t) (7);
2) Continuous non-linear input; due to the physical limitations of the actuators, the output of the actuators acting on the system is not accurate, i.e., there is a non-linearity in the control input to the system (equation 1). These non-linearities are defined by the fact that the field of the device is a sector [ s ]1 s2]A continuous non-linear function N (u)k) And N (0) ═ 0, where s is1And s2Represents two straight lines (t)1,t2) Of (2) i.e.
Let 4, parameter s1And s2Is an unknown non-zero normal number.
3) Designing a controller under the condition of asymmetric dead zone input; and designing a controller for the uncertain nonlinear system with unknown asymmetric dead zone input and initial state error based on the adaptive iterative learning Backstepping framework.
Wherein epsilon1,kAnd ε2,kIs a positive series of convergent numbers, η1And η2Is a design normal number. Because:
according to the description of the radial basis function neural network in the step 2, the uniform approximation performance of the RBFNN shows that the unknown nonlinear function isCan be approximated by RBFNN as the following form on a tight set, and has reconstruction error
Wherein the content of the first and second substances,is an approximation error and W1Is the optimal weight vector. For any real number a > 0 and positive integer l ≧ 2, takeDefinition ofThe virtual controller is taken as:
substituting equations (13) and (14) into (12) can give:
wherein the content of the first and second substances,andare respectively a parameter W1And N1Is estimated.Andis the parameter estimation error. The last two terms on the right of the equation of equation (15) can be rewritten as:
according to equation (16), equation (15) is rewritten as follows:
suppose 5, the residue term ω1Is bounded and | ω1|≤ωM1Wherein ω isM1Is a normal number. Assuming 5 is reasonable, there are two reasons: 1) according to the assumption of 2, the method,is bounded; 2) as long as eta2Is suitably large and is suitably large enough to be,it can be small enough. Take the following non-negative function:
wherein, gamma is11And Γ21Is a symmetric positive definite matrix. V is given below1,kAlong the derivative of system equation (18) with time:
step i (i is more than or equal to 2 and less than or equal to n-1), defineAs will be given later. Let zi+1,k=xi+1,k-αi,kAnalogously to step 1, the following error function z is introduced according to the description of step 2 for the time-varying boundary layeriφ,kAnd z(i+1)φ,k:
Wherein epsiloni,kAnd ε(i+1),kIs a positive series of convergent numbers, ηiAnd ηi+1Is a design normal number. z is a radical ofiφ,kThe derivative with respect to time is as follows:
equation (23) can be rewritten as:
according to the description of the radial basis function neural network in the step 2, the uniform approximation performance of the RBFNN shows that the unknown nonlinear function isCan be approximated by RBFNN as the following form on a tight set, and has reconstruction error
Wherein the content of the first and second substances,is an approximation error and WiIs the optimal weight vector. The virtual controller is taken as:
substituting equation (25) and equation (26) into equation (24) yields the following equation (27):
wherein the content of the first and second substances,andare respectively a parameter WiAnd NiIs estimated.Andis the parameter estimation error. The last two terms to the right of the equal sign of equation (27) can be rewritten as:
suppose 6, the residue term ωiIs bounded and | ωi|≤ωMiWherein ω isMiIs an unknown normal number. The following non-negative function was chosen:
v is given belowi,kDerivative with respect to time along system equation (29):
step n, define zn,k=xn,k-αn-1,kThe following error function z is introduced according to step 2 description of the time-varying boundary layernφ,k:
Wherein epsilonn,kIs a positive series of convergent numbers, ηnIs a design normal number. Then znφ,kThe derivative of (c) is:
then znφ,kThe derivative with respect to time is:
according to the description of the radial basis function neural network in the step 2, the uniform approximation performance of the RBFNN shows that the unknown nonlinear function isCan be approximated by RBFNN as the following form on a tight set, and has reconstruction error
Wherein the content of the first and second substances,is an approximation error and WnIs the optimal weight vector.
uk=u1,k+u2,k (38);
Wherein u is2,kTo compensate for the unknown input gain m (t). Substituting equation (37) and equation (38) into equation (36) yields the following equation (39):
wherein the content of the first and second substances,andare respectively a parameter WnAnd NnIs estimated.Andis the parameter estimation error. The last two terms to the right of the equality sign of equation (39) can be rewritten as:
suppose 7, the residue term ωnIs bounded and | ωn|≤ωMn1Wherein ω isMn1Is an unknown normal number. The following adaptive iterative learning law is taken:
wherein, gamma is1i,Γ2i,Γ3Is a learning law that needs to be designed. The following non-negative function was chosen:
v is given belown,kAlong with the derivative of system equation (42) with respect to time, in conjunction with equations (43) through (45), the following equation (47) results:
for the setting of the initial estimate at each iteration, the following assumptions are given.
that is to say that the position of the first electrode,then when k → ∞ is reached,whereinIs an arbitrarily small normal number. Here, the whole time interval [0, T]If let η1Suitably large then phi1,∞(t) may be arbitrarily small.
4) Designing a controller for the condition of continuous nonlinear input; in selecting a controller in an actual physical system, there is nonlinearity in the control input due to the limitations of the actuator. For controller design, the following arguments are used:
znφ,kN(uk)≤sznφ,kuk,s∈{s1,s2} (49);
and (3) proving that: if u is multiplied simultaneously on both sides of equation (9)kThen, the following formula (50) is obtained:
further, the inequality in the formula (50) is multiplied by each sideThe following formula (51) is obtained:
the formula (51) is rewritten as the following formula (52):
s1(znφ,kuk)2≤znφ,k(znφ,kuk)N(uk)≤s2(znφ,kuk)2 (52);
if the controller ukSatisfies znφ,kuk0 or less, the following inequality holds:
znφ,kN(uk)≤s1znφ,kuk (53);
if the controller ukSatisfies znφ,kukGreater than or equal to 0, the following inequality holds:
znφ,kN(uk)≤s2znφ,kuk (54);
obviously, the conclusion of lemma 1 holds. After the syndrome is confirmed. During the whole design process, use N (u)k) As the output of the system instead of Γ (u)k). Step 1, step i, i 2, n-1 is the same as in part 3) of step 3.
Step n, define zn,k=xn,k-αn-1,kThe following error function z is introduced according to step 2 description of the time-varying boundary layernφ,k:
Wherein epsilonn,kIs a positive series of convergent numbers, ηnIs a design normal number. Then z isnφ,kThe derivative of (c) is:
equation (56) is rewritten as:
according to the description of the radial basis function neural network in the step 2, the uniform approximation performance of the RBFNN (radial basis function neural network) shows that the unknown nonlinear function isCan be approximated by RBFNN as the following form on a tight set, and has reconstruction error
Wherein the content of the first and second substances,is an approximation error and WnIs the optimal weight vector. From equation (57), equation (58) is rewritten as:
the following non-negative function was chosen:
wherein the content of the first and second substances,andare respectively a parameter Wn、NnAndis estimated. Andis the parameter estimation error. Gamma-shaped1i,Γ2i,Γ4Is a learning gain that needs to be designed.
V is given belown,kAlong the derivative of system equation (59) with respect to time, and substituting equation (49), then there is:
uk=u1,k+u2,k (62);
suppose 9, the residue term ωnIs bounded and | ωn|≤ωMn2Wherein ω isMn2Is an unknown normal number. The following adaptive iterative learning law is taken:
in order to set the initial estimation at each iteration, the following assumptions are specifically given: assuming 10, for any k, when t is 0,
theorem 2, if hypothesis 2, hypothesis 4-hypothesis 6, hypothesis 9, and hypothesis 10 are satisfied, and the controller equation (62) and the parametric adaptation rate equations (65) - (67) are applied to the system equation (1), then all signals of the closed-loop system are bounded at [0, T ] and have:
that is to say that the position of the first electrode,then when k → ∞ is reached,whereinIs an arbitrarily small normal number. Here, the whole time interval [0, T]If let η1Suitably large then phi1,∞(t) may be arbitrarily small.
Step 4, analyzing the stability and convergence of the self-adaptive learning controller designed in the step 3; 1) proof theorem 1;
it is proved that according to the description of the time-varying boundary layer in step 2, there areWherein z isφ,k=[z1φ,k,z2φ,k,...,znφ,k]T. From equation (46), the following holds:
byTo obtainThen V0,kIs bounded. And due to the fact thatBy the formula (71) toIs bounded, thereforeIs also bounded. And alsoTherefore, the method comprises the following steps:
from the formula (73),is bounded. Due to DeltakIs bounded, and T ∈ [0, T ]]Thus, therefore, it isIs also bounded. From the above discussion, for any k, Vn,k(t) is bounded, then there is xi,k, Andis bounded. From equation (39), uk is bounded. As can be seen from the formula (29),is bounded, therefore ziφ,kIs consistently continuous, then by the Barbalt theorem, we get when k → ∞, ziφ,k→ 0, that is, when k → ∞, z1φ,k→ 0, by definition of Limit, forSuch that:
when k is greater than N, the number of the transition metal,and is also provided with Then when k → ∞ is reached,so theorem proves that the method is good. 2) Theorem 2 is the same as the certification process of theorem 1.
Example (b): a simulation example is given to illustrate the feasibility and effectiveness of the method of the present invention.
Considering a class of mass-spring mechanical systems to illustrate what is presented in this sectionThe effectiveness of the controller is learned adaptively and iteratively. A massAttached to the wall by a spring and a slide on a horizontal smooth surface, i.e. the resistance caused by friction is assumed to be zero. The mass being influenced by an external force ukCan be regarded as a control variable. Let ykIs a displacement from a reference position. In the presence of asymmetric dead-zone inputs, the dynamic equations for the system are given belowWherein t is ∈ [0,1 ]],Fms(. is) the restoring force of the spring. k denotes an iteration index. Definition of x1,k=yk,And isConverting the system into a state space form:
the restoring force of the spring can be modeled as:in the system, the selected parameter k is 1, a0=0,a1=a2=a3=a4=1,q=4。
1. The case of asymmetric dead-zone input; the asymmetric dead zone is described as follows:
the control target is such that the output of the system (formula 74) is at [0, π → ∞ when k → ∞]Upper trace upper reference track yr,k. Choose to haveThe same amplitude reference track yr,k=gksin (2 π t), choosing g when k is even for the case of non-uniform tracesk-0.1, when k is an odd number, gk0.1. According to theorem 1, the adaptive iterative learning controller is selected as:
wherein the content of the first and second substances,the parameter adaptive iterative learning law is given by equation (43), where η1=20,η2=50,Γ21=1,Γ22=1,Γ 31. The neural network consists of 31 neurons, with the centers of the basis functions uniformly covered [ -1,1 ]]The width of the basis function is chosen to be 1. The following parameters and initial values of the states and estimated parameters are selected: the iteration number k is 30, and the simulation result is shown in fig. 1-fig. 7. It can be seen from fig. 1 that the tracking error can converge to zero. Further, fig. 2-7 show the control signal | | | uk||,In the interval [0, pi]Upper is bounded. The simulation results shown in fig. 1 to fig. 7 further prove the effectiveness of the control method of the controller designed by the design method of the adaptive iterative learning non-uniform target tracking controller provided by the invention.
2. Continuous non-linear outputThe case of entry; here, a continuous non-linear input N (u) is usedk) Replacing asymmetric dead-zone input Γ (u)k). Function N (u)k) Is described as N (u)k)=(0.5+0.1sin(uk))uk. Selecting reference tracks y having different amplitudesr,k=gksin (2 π t), choosing g when k is even for the case of non-uniform tracesk-0.1, when k is an odd number, gk0.1. According to theorem 2, the adaptive iterative learning controller is selected as:
wherein the content of the first and second substances,the parameter adaptive iterative learning law is given by equation (65), where η1=20,η2=50,Γ21=1,Γ22=1,Γ 31. The neural network consists of 31 neurons, with the centers of the basis functions uniformly covered [ -1,1 ]]The width of the basis function is chosen to be 1. The following parameters and initial values of the states and estimated parameters are selected:
the simulation result is shown in fig. 8-14 by taking the iteration number k as 30. It can be seen from fig. 8 that the tracking error can converge toAnd (4) zero. Further, fig. 9-14 show the control signal | | | uk||, In the interval [0,1]Upper is bounded. The simulation results shown in fig. 8-fig. 14 further prove the effectiveness of the control method of the controller designed by the design method of the adaptive iterative learning non-uniform target tracking controller provided by the invention. In summary, the invention solves the problem of non-uniform target tracking control of adaptive iterative learning of a strict feedback nonlinear system when an initial state error and unknown input nonlinearity coexist. A backstepping method and a radial basis function neural network are utilized to solve the problem of trajectory tracking of a strict feedback nonlinear system. A radial basis function neural network is introduced to learn the performance of unknown dynamics, a typical series is utilized to effectively cancel approximation errors, and the problem of inconsistent target tracking is solved. The problem of initial state errors is solved by applying a time-varying boundary layer. The problem of non-linearity of two types of input, namely asymmetric dead zone input and continuous input is effectively solved. It can be shown that in a given time interval 0, T]All signals of the upper closed loop system are bounded and the state tracking error will asymptotically converge to an adjustable set of residuals as the iteration goes to infinity.
Claims (5)
1. A design method of a self-adaptive iterative learning non-uniform target tracking controller is characterized by comprising the following steps: the method specifically comprises the following steps:
step 1, the following formula (1) is adopted to represent a continuous time nonlinear system:
wherein the content of the first and second substances,Γ(uk) E R represents the input-output characteristics of the actuator, ykE, R refers to system output;is a smooth unknown non-linear function; k represents an iteration index;
step 2, the time-varying boundary layer network is expressed by the following formula (2):
wherein phi isi,k(t)=εi,ke-ηt,zi,kAnd ziφ,kIs a function variable of time t, epsiloni,kIs a convergence series sequence, sat is a saturation function, defined as follows:
wherein phi isi,k(t) is a time-varying boundary layer,. phii,k(t) decreasing along the time axis, selecting an initial state error phi at the kth iterationi,k(0)=εi,kThen 0 < epsiloni,ke-ηT≤φi,k(t)≤εi,k,
And 3, expressing the radial basis function neural network by adopting the following formula (4):
wherein the content of the first and second substances,is known lightA sliding vector function, wherein the number l of NN nodes is more than 1; radial basis functionIs a gaussian function and is expressed by the following equation (5):
wherein, mujE.omega and eta > 0 are radial basis functions, respectivelyThe optimal weight vector W ═ ω1,...,ωl]TIs defined as:
wherein the content of the first and second substances,is an approximation error inherent to the NN;
and 4, designing the self-adaptive learning controller by respectively applying a Backstepping method and a radial basis function neural network in the conditions of asymmetric dead zone input and continuous nonlinear input.
2. The design method of the adaptive iterative learning non-uniform target tracking controller according to claim 1, characterized in that: the specific process of the step 4 is as follows:
step 4.1, use function Γ (u)k) The actuator output with an asymmetric dead band is represented as shown in equation (7) below:
wherein the parameter mrAnd mlA right slope and a left slope representing the dead zone characteristic, respectively, parameter brAnd blA breakpoint representing a dead-zone input of the actuator;
and 4.2, modeling the asymmetric dead zone of the actuator into a form of the sum of a straight line and a disturbance-like term, wherein the form is shown in the following formula (8):
Γ(uk)=m(t)uk+d(t) (8);
wherein:
4.3, designing a controller for the uncertain nonlinear system with unknown asymmetric dead zone input and initial state error based on a self-adaptive iterative learning Backstepping frame;
step 4.4, there is a non-linearity in the control input of a class of continuous time non-linear systems represented by equation (1), which is taken to be in the sector region [ s ]1 s2]A continuous non-linear function N (u)k) And N (0) ═ 0, where s is1And s2Represents two straight lines (t)1,t2) Of (2) i.e.
And 4.5, designing a controller for the condition of continuous nonlinear input.
3. The design method of the adaptive iterative learning non-uniform target tracking controller according to claim 2, characterized in that: the specific process of the step 4.3 is as follows:
step A) let z1,k=x1,k-yr,k,z2,k=x2,k-α1,kIn which α is1,kIs a virtual controller; time ticks according to step 2 because of initial state errorsThe expression of the varying boundary layer is applied to the error function z by the following formula1φ,kAnd z2φ,kTo show that:
wherein the content of the first and second substances,ε1,kand ε2,kIs a positive series of convergent numbers, η1And η2Is a designed normal number;
because:
then z is1φ,kThe derivative with respect to time is shown in the following equation (13):
step B) according to the representation of the radial basis function neural network in the step 3, the uniform approximation performance of the RBFNN shows that the unknown nonlinear function isApproximated by RBFNN on a tight set to the following form, and with reconstruction error
Wherein the content of the first and second substances,is the approximation error, W1Is the optimal weight vector;
for any real number a > 0 and positive integer l ≧ 2, takeDefinition ofThe virtual controller is taken as:
substituting equations (14) and (15) into (13) yields the following equation (16):
wherein the content of the first and second substances,andare respectively a parameter W1And N1Is estimated by the estimation of (a) a,andis the parameter estimation error, the last two terms to the right of the equal sign of equation (16) are rewritten as:
on the basis of the formula (17), the formula (16) is rewritten as the following formula (18):
4. the design method of the adaptive iterative learning non-uniform target tracking controller according to claim 3, characterized in that: in the step 4.3:
defining: z is a radical ofn,k=xn,k-αn-1,kAccording to the step 2 representation of the time-varying boundary layer network, the following error function z is introducednφ,k:
Wherein the content of the first and second substances,εn,kis a positive series of convergent numbers, ηnIs a designed normal number;
then z isnφ,kThe derivative of (c) is:
wherein:
defining:
znφ,kthe derivative with respect to time is:
according to the representation of the radial basis function neural network in the step 3, the RBFNN unified approximation performance shows that the unknown nonlinear function is representedApproximated by RBFNN on a tight set to the following form, and with reconstruction error
Wherein the content of the first and second substances,is the approximation error, WnIs the optimal weight vector;
uk=u1,k+u2,k (26);
wherein u is2,kTo compensate for the unknown input gain m (t);
substituting equations (25) and (26) into (24) yields the following equation (27):
wherein the content of the first and second substances,andare respectively a parameter WnAnd NnIs estimated by the estimation of (a) a,andis the parameter estimation error, the last two terms to the right of the equality sign of equation (27) are rewritten as:
then:
getWhereinIs an uncertain parameter(ii) an estimate of (d); order toThen, the formula (29) becomes the following formula (30):
5. the design method of the adaptive iterative learning non-uniform target tracking controller according to claim 4, characterized in that: the specific process of the step 4.5 is as follows:
with N (u)k) As the output of the system instead of Γ (u)k) Definition of zn,k=xn,k-αn-1,kAccording to the step 2 representation of the time-varying boundary layer network, the following error function z is introducednφ,k:
Wherein the content of the first and second substances,εn,kis a positive series of convergent numbers, ηnIs a designed normal number; then z isnφ,kIs a derivative of
Wherein:
defining:
then equation (32) is rewritten as equation (36) below:
according to the representation of the radial basis function neural network in the step 3, the uniform approximation performance of the RBFNN shows that the unknown nonlinear function isApproximated by RBFNN on a tight set to the following form, and with reconstruction error
Wherein the content of the first and second substances,is the approximation error, WnIs the optimal weight vector;
according to the formulas (36) and (37), the following formula (38) is obtained:
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202110507272.0A CN113219832A (en) | 2021-05-10 | 2021-05-10 | Design method of adaptive iterative learning non-uniform target tracking controller |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202110507272.0A CN113219832A (en) | 2021-05-10 | 2021-05-10 | Design method of adaptive iterative learning non-uniform target tracking controller |
Publications (1)
Publication Number | Publication Date |
---|---|
CN113219832A true CN113219832A (en) | 2021-08-06 |
Family
ID=77094383
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202110507272.0A Pending CN113219832A (en) | 2021-05-10 | 2021-05-10 | Design method of adaptive iterative learning non-uniform target tracking controller |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN113219832A (en) |
Cited By (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN114371616A (en) * | 2021-12-09 | 2022-04-19 | 上海工程技术大学 | Tracking control method of dead zone nonlinear time-lag system |
Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN109828528A (en) * | 2019-01-21 | 2019-05-31 | 河北工业职业技术学院 | Robot trace tracking method and device |
CN110134011A (en) * | 2019-04-23 | 2019-08-16 | 浙江工业大学 | A kind of inverted pendulum adaptive iteration study back stepping control method |
CN111290276A (en) * | 2020-02-23 | 2020-06-16 | 西安理工大学 | Fractional order integral sliding mode control method for neural network of hydraulic position servo system |
CN112102366A (en) * | 2020-09-24 | 2020-12-18 | 湘潭大学 | Improved algorithm for tracking unmanned aerial vehicle based on dynamic target |
CN112255920A (en) * | 2020-11-04 | 2021-01-22 | 浙江理工大学 | Self-adaptive optimal iterative learning control method based on data driving |
CN112462608A (en) * | 2020-11-18 | 2021-03-09 | 大连交通大学 | Discrete sliding mode track and speed tracking control method for high-speed train |
-
2021
- 2021-05-10 CN CN202110507272.0A patent/CN113219832A/en active Pending
Patent Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN109828528A (en) * | 2019-01-21 | 2019-05-31 | 河北工业职业技术学院 | Robot trace tracking method and device |
CN110134011A (en) * | 2019-04-23 | 2019-08-16 | 浙江工业大学 | A kind of inverted pendulum adaptive iteration study back stepping control method |
CN111290276A (en) * | 2020-02-23 | 2020-06-16 | 西安理工大学 | Fractional order integral sliding mode control method for neural network of hydraulic position servo system |
CN112102366A (en) * | 2020-09-24 | 2020-12-18 | 湘潭大学 | Improved algorithm for tracking unmanned aerial vehicle based on dynamic target |
CN112255920A (en) * | 2020-11-04 | 2021-01-22 | 浙江理工大学 | Self-adaptive optimal iterative learning control method based on data driving |
CN112462608A (en) * | 2020-11-18 | 2021-03-09 | 大连交通大学 | Discrete sliding mode track and speed tracking control method for high-speed train |
Non-Patent Citations (1)
Title |
---|
张春丽: "几类非线性系统的自适应迭代学习控制研究", 《信息科技辑》 * |
Cited By (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN114371616A (en) * | 2021-12-09 | 2022-04-19 | 上海工程技术大学 | Tracking control method of dead zone nonlinear time-lag system |
CN114371616B (en) * | 2021-12-09 | 2023-09-12 | 上海工程技术大学 | Tracking control method of dead zone nonlinear time-lapse system |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
Wen et al. | Optimized backstepping for tracking control of strict-feedback systems | |
Cunha et al. | Output-feedback model-reference sliding mode control of uncertain multivariable systems | |
Wu et al. | Practical adaptive fuzzy tracking control for a class of perturbed nonlinear systems with backlash nonlinearity | |
Rivals et al. | Black-box modeling with state-space neural networks | |
CN113359445B (en) | Distributed output feedback asymptotic consistent control method for multi-agent hysteresis system | |
CN108733031B (en) | Network control system fault estimation method based on intermediate estimator | |
CN111474922B (en) | Method for constructing controller of continuous nonlinear system | |
CN114509949A (en) | Control method for presetting performance of robot | |
CN113219832A (en) | Design method of adaptive iterative learning non-uniform target tracking controller | |
Azhdari et al. | Adaptive robust tracker design for nonlinear sandwich systems subject to saturation nonlinearities | |
CN113325717A (en) | Optimal fault-tolerant control method, system, processing equipment and storage medium based on interconnected large-scale system | |
Xie et al. | Observer based control of piezoelectric actuators with classical Duhem modeled hysteresis | |
Shiev et al. | Trajectory control of manipulators using type-2 fuzzy neural friction and disturbance compensator | |
Choi et al. | Neural network-based Smith predictor design for the time-delay in a tele-operated control system | |
Wang et al. | Smart neural control of uncertain non‐linear systems | |
CN112346342A (en) | Single-network self-adaptive evaluation design method of non-affine dynamic system | |
Psillakis et al. | Unifying adaptive control with the nonlinear PI methodology: designs for unknown strict‐feedback nonlinear systems with nonsmooth actuator nonlinearities | |
Li et al. | Neural sliding mode control for systems with hysteresis | |
CN117335957B (en) | Secret communication method for BAM memristor neural network correction function projection synchronization | |
Shao et al. | T‐S modelling‐based anti‐disturbance finite‐time control with input saturation | |
Wang et al. | Direct Neural Network Adaptive Tracking Control for Uncertain Non-Strict Feedback Systems With Nonsymmetric Dead-Zone | |
JP3872457B2 (en) | Learning adaptive controller | |
Efe | Identification and control of nonlinear dynamical systems using neural networks | |
Du | Fuzzy mixed h2/h? sampled-data control design for nonlinear dynamic systems | |
Yu et al. | Adaptive Fuzzy Event-Triggered Control for Nonlinear Systems with Asymmetric Hysteresis |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
RJ01 | Rejection of invention patent application after publication | ||
RJ01 | Rejection of invention patent application after publication |
Application publication date: 20210806 |