CN-121704576-B - Active vibration isolation method and system based on prediction and self-adaptive control

CN121704576BCN 121704576 BCN121704576 BCN 121704576BCN-121704576-B

Abstract

The invention belongs to the technical field of active vibration isolation, and particularly discloses an active vibration isolation method and an active vibration isolation system based on prediction and self-adaptive control. The method comprises the steps of performing multi-degree-of-freedom dynamics modeling on a vibration isolation system, taking the pose of a load as system output, taking the control quantity of an execution mechanism as system input, uniformly representing unknown dynamics caused by structural nonlinearity and disturbance as nonlinear items, building a self-adaptive neural network state observer based on the dynamics model, estimating a state which cannot be directly measured, collecting historical operation data of the vibration isolation system, preprocessing the data, constructing an input time sequence and an output time sequence, training a long-term and short-term memory neural network prediction model to generate a prediction signal representing future position information, designing a self-adaptive control law according to a state estimation result and the prediction signal, obtaining a feedback control quantity, and superposing the feedback control quantity and a feedforward control quantity obtained based on the prediction signal to form a control instruction for driving the execution mechanism.

Inventors

LIU SHUAI
Wang sanxia
CHEN GUOXING
BU SHI
WANG JUNJIE

Assignees

山东大学
北京半导体专用设备研究所（中国电子科技集团公司第四十五研究所）

Dates

Publication Date: 20260512
Application Date: 20260213

Claims (9)

1. An active vibration isolation method based on prediction and self-adaptive control is characterized by comprising the following steps: Performing multi-degree-of-freedom dynamics modeling on the vibration isolation system, taking the pose quantity of the load as system output, taking the control quantity of the execution mechanism as system input, uniformly representing unknown dynamics as nonlinear items, and defining a state which cannot be directly measured in the system; a self-adaptive neural network state observer is designed based on a multi-degree-of-freedom dynamics model, and a neural network is used for approximating nonlinear items to obtain an estimation result of an undetectable state of the system; constructing a long-term and short-term memory neural network prediction model, collecting historical operation data of a vibration isolation system, and preprocessing the historical operation data to obtain an input time sequence and an output time sequence; Training a long-term and short-term memory neural network prediction model based on the input time sequence and the output time sequence, so that the prediction model can output a prediction signal used for representing the future position information of the load; Constructing a feedback control law by adopting a self-adaptive control strategy according to a state estimation result and a prediction signal which are given by a state observer, and compensating the non-linear item and the parameter uncertainty to obtain a feedback control quantity; Superposing the feedback control quantity and the feedforward control quantity obtained based on the prediction signal to generate a control instruction for driving an executing mechanism of the vibration isolation system, and implementing active vibration isolation control on the load; Tracking error from output by back-stepping frame method Initially, as Designing virtual control laws Defining a second error Is that Designing virtual control laws ; Recursion is performed in this way until the actual control law is designed in the last step ; At each step, an unknown nonlinear combination term RBF neural network Approximation, approximation error Has an upper bound; control law sum is neural network weight And The adaptive law of design is designed as follows: Wherein the method comprises the steps of , , , , , All are normal numbers.
2. The active vibration isolation method based on prediction and adaptive control of claim 1, wherein the step of performing multi-degree of freedom dynamics modeling of the vibration isolation system comprises: The position and the gesture of the load in the space are expressed as output variables of a multi-degree-of-freedom dynamic model, and corresponding state dimensions are determined according to the structural characteristics of the system; Taking the opening of a throttle valve, electromagnetic driving force or air film regulating quantity as control input of a dynamic model, and modeling and representing the force or moment of an actuating mechanism acting on a load; unifying unknown dynamics caused by air film pressure change, friction, structure flexibility change and external disturbance in a vibration isolation system into nonlinear function items and introducing the nonlinear function items into a model structure; defining the speed, the attitude change rate or the internal dynamic variable which cannot be directly obtained by a sensor in the dynamic model, so that the speed, the attitude change rate or the internal dynamic variable and the measurable output form a system state vector together; the multi-degree-of-freedom dynamics model is as follows: Wherein, the Is in a system state in which Is the measurable load pose information; And Respectively a control input and a system output, Is an unknown smooth nonlinear function, and the multiple-degree-of-freedom dynamics model is shared Degree of freedom of The subsystems are represented separately.
3. The active vibration isolation method based on prediction and adaptive control according to claim 2, wherein the step of designing the adaptive neural network state observer based on the multiple degree of freedom dynamics model comprises: Constructing a state observation structure comprising measurable output quantity and unmeasurable state quantity according to the multi-degree-of-freedom dynamics model; Selecting a radial basis function as a basis function form of a neural network, and constructing a neural network input vector for approximating a nonlinear term according to the input quantity and the measurable output quantity of the system; Dynamically meeting a convergence condition by setting the gain of the observer, wherein the observer error is driven by the measurable output quantity and the approximation result of the neural network; and in the running process of the observer, the weight of the neural network is adaptively updated according to the observation error so as to obtain real-time estimation of the state of the system which can not be directly measured.
4. The active vibration isolation method based on prediction and adaptive control according to claim 3, wherein when Time state Not measurable, a neural network state observer is designed: Wherein the method comprises the steps of Is RBF neural network for approximating unknown nonlinear function in system ; By selecting observer gain So that For a Hulvitz matrix, the observer is adopted to asymptotically estimate the total state of the system 。
5. The active vibration isolation method based on prediction and adaptive control according to any one of claims 1 to 4, wherein when constructing the long-term memory neural network prediction model, it comprises: For a pair of Subsystem, design self-adaptive neural network controller separately Design of And a predictive controller.
6. The active vibration isolation method based on predictive and adaptive control of claim 5, comprising, when designing the predictive controller: For the ith subsystem Standardization: ; Adopting Huber loss-based quadratic regression, solving by an iterative weighted least square method, and automatically distributing low weight to abnormal values to obtain a smooth curve which can reflect the data core trend and is not interfered by noise ; Training data structure with lead effect: Input device For a section of history data ; Output of Is to the future Individual points Weighted summation is performed by applying a gaussian window: ; the center of the Gaussian window is arranged at the rear position of the window; A plurality of using the above construction For a training deep LSTM network.
7. The method of claim 6, wherein the trained deep LSTM network receives the latest position sequence and outputs a filtered position prediction value with a lead amount ; The position prediction controller is defined as 。
8. The predictive and adaptive control-based active vibration isolation method of claim 7, wherein superimposing the feedback control amount with the feedforward control amount based on the predictive signal to generate the control command for driving the vibration isolation system actuator comprises: First virtual control signal of adaptive neural network controller Compensation signal with position prediction controller Fusing to form final control instruction 。
9. An active vibration isolation system based on prediction and adaptive control for implementing the active vibration isolation method based on prediction and adaptive control as claimed in claim 1, characterized in that the system comprises: The load position sensor is configured to acquire pose information of a load in a space and output a load pose measurement signal; the vibration isolation device comprises an air spring assembly and an actuating mechanism connected with the air spring, wherein the actuating mechanism comprises a high-speed throttle valve and a Lorentz motor and is configured to adjust the air film pressure or apply electromagnetic force according to a control signal; The multi-degree-of-freedom dynamics model construction unit is configured to establish a multi-degree-of-freedom dynamics model based on a load pose measurement signal, define the spatial position and the pose of a load as model output, take the opening degree of a high-speed throttle valve and electromagnetic driving force as control input, represent unknown dynamics formed by air film pressure change, friction, structure flexibility change and external disturbance as nonlinear items, and define speed, pose change rate and internal dynamics variables which can not be directly measured to form a system state vector; the neural network state observer is configured to construct a state observation structure based on a multi-degree-of-freedom dynamics model, approximates a nonlinear term by using a radial basis function neural network, updates the neural network weight on line according to observer errors, and obtains an estimated quantity of an unmeasurable state of the system; The long-period memory neural network prediction module is configured to collect historical operation data of the vibration isolation system, normalize and smooth the historical operation data to form an input time sequence and an output time sequence, train the long-period memory neural network to generate a prediction model capable of outputting a future position signal of a load, and receive the latest position sequence in the operation process to generate a position prediction value with a lead in real time; The self-adaptive back-step control module is configured to construct a multi-layer virtual control quantity according to a back-step method based on a state estimation result of the state observer and a prediction signal output by the prediction module, approximate an unknown nonlinear combination item by using a neural network and update weights through a self-adaptive law, so as to generate a feedback control quantity; The control signal fusion module is configured to superimpose the feedback control quantity and the feedforward control quantity generated by the prediction signal to form a composite control instruction for driving the execution mechanism; The real-time control unit comprises a real-time target machine and a control model execution environment, and is configured to receive a load pose measurement signal, a prediction signal and a state estimation result, run a state observer, a self-adaptive backstepping control module and a prediction module, and output a composite control instruction to an execution mechanism so as to realize active vibration isolation control.

Description

Active vibration isolation method and system based on prediction and self-adaptive control Technical Field The invention belongs to the technical field of active vibration isolation, and particularly relates to an active vibration isolation method and system based on prediction and self-adaptive control. Background Vibration, as a ubiquitous environmental disturbance factor, can have a significant impact on the performance of high-end equipment such as precision manufacturing, nanoscale measurement, optical imaging, aerospace testing, and the like. Especially in the application scenes such as a scanning electron microscope, an atomic force microscope and the like which are extremely sensitive to micro-displacement, the imaging blurring, the measuring error or the processing precision reduction can be caused by the micro-vibration, so that the isolation of the environment vibration becomes an indispensable key link in the operation process of the precision equipment. The active vibration isolation technology forms a closed loop system through the sensor, the controller and the executing mechanism, and can realize wide-band vibration suppression in theory, so that the active vibration isolation technology is widely researched and applied in precision equipment. The existing industrial system generally adopts a proportional-integral-derivative (PID) linear control method, although the structure is simple and the implementation is easy. Although the existing prediction method based on the neural network or data driving can model complex dynamics to a certain extent, most of the prediction methods only perform short-time prediction, cannot form stable and controllable phase advance, and lack deep fusion with a real-time robust control strategy. Meanwhile, in the multi-degree-of-freedom vibration isolation system, the coupling relation, the unmeasurable state estimation and the unknown nonlinear term compensation among a plurality of directions need to be processed at the same time, and the prediction capability, the robustness and the instantaneity are often difficult to be combined in the existing scheme. Disclosure of Invention Aiming at the problems in the prior art, the invention provides an active vibration isolation method and an active vibration isolation system based on prediction and self-adaptive control, so as to solve the problems that in the multi-degree-of-freedom vibration isolation system in the prior art, the coupling relation between a plurality of directions, the undetectable state estimation and the compensation of unknown nonlinear items are required to be processed simultaneously, and the problems of prediction capability, robustness and instantaneity are difficult to be considered in the existing scheme. The technical scheme adopted by the invention is as follows: In a first aspect, the present application provides an active vibration isolation method based on predictive and adaptive control, the method comprising the steps of: Performing multi-degree-of-freedom dynamics modeling on the vibration isolation system, taking the pose quantity of the load as system output, taking the control quantity of the execution mechanism as system input, uniformly representing unknown dynamics as nonlinear items, and defining a state which cannot be directly measured in the system; a self-adaptive neural network state observer is designed based on a multi-degree-of-freedom dynamics model, and a neural network is used for approximating nonlinear items to obtain an estimation result of an undetectable state of the system; constructing a long-term and short-term memory neural network prediction model, collecting historical operation data of a vibration isolation system, and preprocessing the historical operation data to obtain an input time sequence and an output time sequence; Training a long-term and short-term memory neural network prediction model based on the input time sequence and the output time sequence, so that the prediction model can output a prediction signal used for representing the future position information of the load; Constructing a feedback control law by adopting a self-adaptive control strategy according to a state estimation result and a prediction signal which are given by a state observer, and compensating the non-linear item and the parameter uncertainty to obtain a feedback control quantity; And superposing the feedback control quantity and the feedforward control quantity obtained based on the prediction signal to generate a control instruction for driving an executing mechanism of the vibration isolation system, and executing active vibration isolation control on the load. Further, the step of performing multi-degree-of-freedom dynamics modeling on the vibration isolation system includes: The position and the gesture of the load in the space are expressed as output variables of a multi-degree-of-freedom dynamic model, and corresponding state dimensions are determined according t