CN-122018335-A - Method and system for introducing compound learning into nonlinear system output feedback self-adaptive control

CN122018335ACN 122018335 ACN122018335 ACN 122018335ACN-122018335-A

Abstract

The invention belongs to the technical field of nonlinear system control, and discloses a method and a system for introducing compound learning into nonlinear system output feedback self-adaptive control. The invention directly multiplexes the existing K-filter in the standard output feedback back-step control to construct the extended prediction error without establishing an additional observer or state estimation model (such as a serial-parallel estimation model, a fuzzy observer and the like) parallel to the K-filter. The function multiplexing design eliminates the calculation overhead of an additional dynamic system, reduces the storage requirement and the real-time calculation burden of a controller, and obviously reduces the hardware realization cost and the system debugging complexity. Meanwhile, through the compound drive of tracking errors and extended prediction errors, the accumulated information is continuously introduced by utilizing the historical memory of regression quantity, the self-adaption and self-learning capacity of a control system is obviously enhanced, and the parameter estimation convergence and the system response performance are improved.

Inventors

ZHU RUSONG
ZHANG SHUANGXI
WANG SHENGLI
HU JUN

Assignees

中国空气动力研究与发展中心设备设计与测试技术研究所

Dates

Publication Date: 20260512
Application Date: 20260410

Claims (7)

1. A method for introducing compound learning into nonlinear system output feedback adaptive control, comprising the steps of: For a nonlinear controlled object with an output feedback form, estimating an internal unmeasurable state of a system by using a K-filter to generate a filtering signal and a regression matrix, wherein the K-filter directly multiplexes a state observer structure in a standard output feedback control law without establishing an additional observer or a state estimation model parallel to the K-filter; constructing an extended prediction error by moving time window integral operation by utilizing the regression matrix, wherein the extended prediction error represents accumulated prediction deviation based on regression amount history memory in an integral period; The tracking error generated in the backstepping control process and the extended prediction error jointly drive an updating law of an unknown parameter, a composite parameter self-adaptive law is constructed, and an estimated value of the unknown parameter is updated by utilizing the composite parameter self-adaptive law; Based on the updated parameter estimation value, generating an actual control law through a backstepping recursion design, and driving a controlled object to realize the asymptotic tracking of the system output to the reference signal.
2. The method for introducing compound learning into nonlinear system output feedback adaptive control according to claim 1, wherein the need to build additional observers in parallel with the K-filter is eliminated, in particular comprising the need to build a serial-parallel estimation model, a fuzzy state observer or a neural network observer.
3. The method of introducing compound learning into nonlinear system output feedback adaptive control of claim 1 wherein the design of the K-filter comprises: Designing a filter dynamic equation to generate the filtering signals and a regression matrix, wherein the parameter vector of the K-filter is selected so that the filter dynamic matrix is a Hulvitz matrix; and reconstructing the system state by using the filtering signals and the regression matrix to obtain a state estimation expression containing unknown parameters, wherein the state estimation error index converges.
4. The method for introducing compound learning into nonlinear system output feedback adaptive control in accordance with claim 1, wherein said constructing an extended prediction error by a moving time window integral operation comprises: A static regression relation between the system output and the unknown parameters established by the K-filter is utilized; Decomposing the current value and the historical value of the regression matrix to construct a regression matrix containing a control gain component and a residual parameter component; Integrating the product of the regression matrix and the transpose thereof in a moving time interval to construct an expanded regression matrix; and calculating the extended prediction error based on the extended regression matrix, the integral value output by the system and the parameter estimation value.
5. The method for introducing compound learning into nonlinear system output feedback adaptive control as claimed in claim 1, wherein said double estimating the control gain parameter using an over-parameterized scheme comprises: In a first step of the back-step control, calculating a stabilizing function using the first control gain estimate; In a subsequent back-step, processing a tracking error coupling term using a second control gain estimate that is independent of the first control gain estimate; The first control gain estimated value is updated by the projection mechanism to ensure that the sign of the first control gain estimated value is known and the absolute value of the first control gain estimated value is larger than a preset lower limit, and the second control gain estimated value is independently updated with the first control gain estimated value as a part of a parameter vector.
6. The method of introducing compound learning into nonlinear system output feedback adaptive control in accordance with claim 1, said method further comprising the step of stability verification: Constructing a Lyapunov functional containing tracking error, parameter estimation error and observation error, wherein the Lyapunov functional contains a double integral term used for processing the integral of a moving time window so as to eliminate the influence of the residual state estimation error in the extended prediction error on the integral of the time window; By selecting appropriate gain parameters, the time derivative of the Lyapunov functional is negatively determined, thereby proving that all signals of the closed loop system are globally consistent and bounded and tracking errors are asymptotically converged.
7. A system for introducing compound learning into nonlinear system output feedback adaptive control, comprising: The system comprises an estimation module, a feedback control module and a feedback control module, wherein the estimation module is used for estimating an internal unmeasurable state of a system by using a K-filter aiming at a nonlinear controlled object with an output feedback form to generate a filtering signal and a regression matrix, and the K-filter directly multiplexes a state observer structure in a standard output feedback control law without establishing an additional observer or a state estimation model parallel to the K-filter; the first construction module is used for constructing an extended prediction error by utilizing the regression matrix through moving time window integral operation, wherein the extended prediction error represents accumulated prediction deviation based on regression quantity history memory in an integral period; The second construction module is used for jointly driving an updating law of an unknown parameter by a tracking error and the extended prediction error generated in a backstepping control process, constructing a composite parameter self-adaptive law, and updating an estimated value of the unknown parameter by using the composite parameter self-adaptive law, wherein the control gain parameter is subjected to double estimation by adopting a parameterized scheme, and the estimated value is kept away from zero and a symbol is kept unchanged by a projection mechanism; and the driving module is used for generating an actual control law through a backstepping recursion design based on the updated parameter estimation value and driving the controlled object to realize the asymptotic tracking of the system output to the reference signal.

Description

Method and system for introducing compound learning into nonlinear system output feedback self-adaptive control Technical Field The invention relates to the technical field of nonlinear system control, in particular to a method and a system for introducing compound learning into nonlinear system output feedback self-adaptive control. Background Adaptive control techniques have become an important means of dealing with non-linear system parameter uncertainty by adjusting controller parameters on-line to account for system uncertainty. Aiming at an actual engineering system with incompletely measurable state, the self-adaptive output feedback control reconstructs the internal state through a designed state observer, and realizes closed-loop control by only using output signals. In order to further improve the parameter estimation performance, the composite self-adaptive control method drives parameter updating together by introducing prediction error information (or identification error) and tracking error so as to enhance the parameter learning capability and convergence performance. However, in engineering practice where compound learning is applied to adaptive output feedback control, the following objective technical problems still remain: (1) The system has high implementation complexity and heavy real-time calculation burden. To construct the prediction error, the existing compound adaptive output feedback control method generally needs to build an additional state estimation model (such as a serial-parallel estimation model, a fuzzy logic observer or a neural network observer, etc.). These additional dynamic systems run concurrently with the original state observer, resulting in a controller that needs to maintain real-time solutions for multiple sets of differential equations, significantly increasing computational resource consumption and storage requirements. More importantly, independent setting of multiple sets of observer parameters brings heavy burden to engineering debugging, and the improvement of the complexity of the system structure directly promotes hardware realization cost and reliability risk. (2) The parameter learning efficiency is limited, and the convergence speed is slow. The existing compound self-adaptive method generally only uses the instantaneous prediction error at the current moment to update parameters, and cannot effectively use the continuous excitation information in the historical data. Under the scene of slow system parameter change or insufficient excitation of a reference signal, the updating mode which only depends on instantaneous information leads to slow convergence speed of parameter estimation, so that the convergence time of tracking errors is prolonged, the overshoot is increased, and the dynamic response quality of the system is affected. (3) The control signal is easy to be singular, and the system safety is not enough. In the output feedback design based on the back-step control, the estimated value of the control gain parameter directly appears at the denominator position of the control law. The prior art lacks an effective mechanism to ensure that the estimated value is always far from zero, and the estimated value can approach zero or even change number in the process of improper parameter initialization or transient state, so that the control signal generates singular jump. The method has the advantages that the tracking performance is seriously influenced, unmodeled dynamic state of the controlled object is more likely to be excited, system instability or equipment damage is caused, and the application of the method in the safety critical field is limited. (4) The applicable system is limited in type and has insufficient universality. The existing compound self-adaptive method is mostly aimed at a nonlinear system in a strict feedback form, and compound learning research aimed at an output feedback form (Output Feedback Form) system is relatively insufficient. In output feedback form systems, direct application of the prior art faces structural compatibility barriers due to dynamic coupling of control gain and system, resulting in difficulty in applying compound learning to enhance performance for many practical physical systems (e.g., certain aircraft, mechanical systems) that only output a measurable quantity. In view of this, the present invention has been made. Disclosure of Invention The invention aims at solving at least any one of the technical problems, and provides a method and a system for introducing compound learning into nonlinear system output feedback self-adaptive control. In order to achieve the above object, the first technical scheme adopted by the present invention is as follows: A method of introducing compound learning into nonlinear system output feedback adaptive control, comprising the steps of: For a nonlinear controlled object with an output feedback form, estimating an internal unmeasurable state of a system by using a K-fi