CN-121704164-B - Dynamic surface control method and equipment for active vibration isolation system

CN121704164BCN 121704164 BCN121704164 BCN 121704164BCN-121704164-B

Abstract

The invention provides a dynamic surface control method and equipment for an active vibration isolation system, which belong to the technical field of vibration isolation of precise equipment, and are used for establishing a system state space model, collecting and preprocessing load position data, training an LSTM neural network to obtain a prediction model, acquiring a system state and predicting a load position in real time, designing a virtual control law with advanced compensation based on the prediction position, recursively defining a follow-up control law through the dynamic surface control method, calculating a throttle valve as power, outputting control force and updating the state for cyclic execution. The LSTM neural network is used for carrying out advanced smooth prediction on the load position, and the prediction information is deeply fused into the design of the dynamic surface controller, so that the advanced compensation on the system vibration is realized, the control precision and the dynamic response speed of the vibration isolation system are improved, and the stability and the robustness of the system are ensured.

Inventors

LIU SHUAI
Wang sanxia
WANG JUNJIE
BU SHI
LIU BIN

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. The dynamic surface control method of the active vibration isolation system is characterized by comprising the following steps of: S101, establishing a state space model of an active vibration isolation system, and defining a system state variable and a control input; S102, collecting time sequence data of a load position, and preprocessing the time sequence data to obtain smooth denoised load position data; S103, training an LSTM neural network by using the load position data after smooth denoising to obtain a load position prediction model, wherein in the training process, the input is a historical load position sequence, and the output is a Gaussian weighted future load position trend value; s104, in the real-time control process, acquiring a current system state variable, and predicting a load position by using a load position prediction model to obtain a predicted load position; S105, defining a load position tracking error according to a predicted load position and an expected load position, and designing a first virtual control law; calculating a current load position The difference from the desired load position trajectory yd (t), the difference being defined as the load position tracking error , ; For the current load position it is possible to provide, Is the desired load position; Performing analytical differentiation or numerical differentiation on a given expected load position locus yd (t) to obtain an expected load speed ; Obtaining a predicted load position from step S104 And calculates and expects the load position To obtain the predicted deviation ; Building a virtual control law The virtual control law contains the desired load speed Proportional feedback term for load position tracking error Feedforward compensation term based on prediction bias , wherein, Combining the obtained results according to the basic structure of the first virtual control law to form a mathematical expression of the first virtual control law: ; the obtained virtual control law By a time constant of Is provided with a first order low pass filter, obtaining a filtered output signal Dynamics is composed of Description, the initial conditions are set to ; S106, based on the first virtual control law, gradually defining a subsequent virtual control law and tracking errors through a dynamic surface control method; The follow-up virtual control law comprises a load speed tracking surface control law, an air cavity pressure tracking surface control law, a throttle valve core displacement tracking surface control law and a throttle valve core speed tracking surface control law; s107, calculating throttle valve acting force as final control input according to all virtual control laws and tracking errors; S108, outputting the throttle valve actuating force to a throttle valve executing mechanism, updating the system state variable and the load position prediction in real time, and repeatedly executing the control cycle.
2. The method of dynamic surface control of an active vibration isolation system according to claim 1, wherein, S101 specifically comprises the following steps: defining load displacement as a first state variable Defining the load speed as a second state variable The air cavity relative pressure, i.e. the difference between the air cavity pressure and the ambient pressure, is defined as a third state variable The throttle valve core displacement is defined as a fourth state variable The throttle valve spool speed is defined as a fifth state variable ; Based on Newton's second law, according to the dynamics equation of the load The rewriting is: ; using defined state variables, the method Replaced by Will be Replaced by Will be Replaced by Thereby obtaining the load acceleration equation expressed in terms of state variables Wherein the method comprises the steps of ; Based on mass conservation law and gas state equation, dynamic equation according to air cavity pressure Therein is provided with Expressed as% + ); Using defined state variables, z is replaced with Load speed is to Replaced by Substitution of (Pc-Pa) for Displacing the valve core of throttle valve Replaced by Thereby obtaining the pressure change rate equation expressed in terms of state variables Wherein the method comprises the steps of Based on Newton's second law, the dynamics equation of throttle valve is configured Wherein: The valve core of the throttle valve is displaced; equivalent mass of the valve core; damping coefficient of valve core movement; the spring rate of return of the valve core; The actuating force acting on the valve core; the pressure area of the valve core; Air cavity pressure And ambient pressure The difference between the two, Is at atmospheric pressure; Rewriting the dynamics equation of the throttle valve to Using defined state variables, the method Replaced by The valve core speed of the throttle valve Replaced by Will be Replaced by Will control the input Is defined as Obtaining a valve core acceleration equation expressed by state variables ; Wherein, the Integrating all the obtained equations to form a system state space model: Wherein, the A vector representing the i-th state.
3. The method of dynamic surface control of an active vibration isolation system according to claim 1, wherein, S102 specifically comprises the following steps: configuring a data acquisition system and setting sampling parameters; Continuously collecting the load position of the vibration isolation object in the running process through a data collection system to form an original time sequence data sequence Wherein A load position value representing an i-th sampling time; Carrying out Min-Max normalization calculation on the obtained original data set Z, and linearly mapping the Min-Max normalization calculation to a [0,1] interval; smoothing and denoising the normalized data by adopting robust quadratic regression; and defining a Huber loss function, performing iterative weighted fitting, and outputting load position data after smooth denoising.
4. The method of dynamic surface control of an active vibration isolation system according to claim 1, wherein, S103 specifically comprises the following steps: setting a sliding window length L for smoothing the denoised load position data sequence Setting a predicted total step length P and Gaussian window parameters mu and sigma; Using a sliding window of length L, sequences after smoothing denoising Sequentially sliding, intercepting continuous historical load position subsequences to form training input feature set Wherein each sample ; For each input sample Normalized raw data of corresponding future P points Gaussian weighted summation is carried out to calculate an output target Wherein the weights are ; And constructing an LSTM neural network model structure, configuring model training parameters, starting a training process, completing model training and storing a final model.
5. The method of dynamic surface control of an active vibration isolation system according to claim 1, wherein, S104 specifically comprises the following steps: configuring a first-in first-out data buffer with a length L in a controller; When each control period starts, reading the measured values of the load position sensor, the pressure sensor and the valve core displacement sensor, updating the current value of the state variable x 1 ,x 3 ,x 4 , and updating the current value of x 2 ,x 5 through differential calculation or direct measurement; From the current load position x 1 , normalization calculation is performed according to the normalization parameters exactly the same as the training phase, i.e. min (Z) and max (Z) based on the training dataset by the following formula: Adopting a robust quadratic regression algorithm which is the same as the training stage to carry out real-time smoothing treatment on the normalized value to obtain ; The obtained real-time load position data after smoothing processing Pushing the initialized FIFO buffer, discarding the oldest data point in the buffer, thereby forming the latest historical load position sequence with length L As a current input to the load position prediction model; The constructed current input sequence The load position prediction model executes forward propagation calculation, sequentially passes through an LSTM layer, a Dropout layer and a full connection layer, and finally generates a Gaussian weighted future load position trend predicted value at a regression output layer ; Outputting the obtained prediction on the normalized scale Performing inverse normalization calculation, mapping the calculated value back to the original physical dimension to obtain the predicted load position in the final physical world 。
6. The method of dynamic surface control of an active vibration isolation system according to claim 1, wherein, S106 specifically comprises the following steps: calculating load speed And the first filter output Is defined as the load speed tracking error ; Based on error, system dynamics function Controlling gain Designing a second virtual control law Wherein the method comprises the steps of Control gain being positive; will virtually control law The passing time constant is A first order low pass filter of (2) to obtain a filtered output The dynamics of which are described as The initial conditions are set as ; Calculating the relative pressure of air chambers And a second filter output Is defined as the air cavity pressure tracking error Based on the error and system dynamics function Controlling gain Design of a third virtual control law Wherein the method comprises the steps of Control gain being positive; will virtually control law The passing time constant is A first order low pass filter of (2) to obtain a filtered output The dynamics of which are described as The initial conditions are set as ; Calculating the displacement of the valve core of a throttle valve And a third filter output Is defined as the spool displacement tracking error Designing a fourth virtual control law based on the error Wherein the method comprises the steps of Control gain being positive; will virtually control law The passing time constant is A first order low pass filter of (2) to obtain a filtered output Dynamics are described as The initial conditions are set as 。
7. The method of dynamic surface control of an active vibration isolation system according to claim 1, wherein, S107 specifically comprises the following steps: Calculating the speed of the valve core of the throttle valve And the fourth filter output Is defined as the spool speed tracking error ; All elements required for calculating the final control law are prepared, including the system dynamics function Differentiation of the fourth filter output Spool speed tracking error Tracking error of valve core displacement Gain control Positive control gain ; Based on current system state variables According to the formula Real-time computing system dynamics functions Is a numerical value of (2); Obtaining differential signals output by a filter ; From the fourth first order low pass filter's kinetic equation And are currently known And (3) with Through algebraic operation of the values of (2) Real-time calculation to obtain differential signal ; All elements prepared in steps S1072 to S1074 are combined according to the control law structure, and the final throttle valve actuating force is calculated ; Actuating the calculated throttle valve And carrying out amplitude limiting processing to ensure that the numerical value is constrained within a preset reasonable range.
8. The method of dynamic surface control of an active vibration isolation system according to claim 1, wherein, S108 specifically comprises the following steps: converting the calculated throttle valve actuating digital signal into an analog driving signal; The analog driving signal is amplified by power and then output to a throttle valve executing mechanism; Updating the system state variable acquired in the current control period to a storage unit; Updating the load position data after the current smoothing treatment to a prediction model input buffer area; Starting a timer after the current control period is ended, and triggering the next control period after waiting for a fixed time interval; the system signal is monitored during execution of the control loop and a safety protection mechanism is enabled upon an exception.
9. An electronic device comprising a memory, a processor and a computer program stored on the memory and executable on the processor, characterized in that the processor implements the steps of the active vibration isolation system dynamic surface control method according to any one of claims 1 to 8 when the program is executed by the processor.

Description

Dynamic surface control method and equipment for active vibration isolation system Technical Field The invention belongs to the technical field of vibration isolation of precision equipment, and particularly relates to a dynamic surface control method and equipment of an active vibration isolation system, which combine long-term and short-term memory (LongShort-TermMemory, LSTM) neural network position prediction and dynamic surface control (DynamicSurfaceControl, DSC). Background In the fields of precision manufacturing, precision measurement, aerospace inertial navigation and the like, micro-vibration of a working environment is one of key factors influencing equipment performance and yield. The active vibration isolation system counteracts vibration by applying reverse control force, and is a core device for ensuring stable operation of the precise equipment. Conventional active vibration isolation systems commonly employ proportional-integral-derivative (PID) control. The PID controller has simple structure and convenient parameter setting, but when dealing with a system with essential nonlinearity, time-lag characteristic and complex external disturbance such as an air floatation vibration isolation platform, the control performance of the PID controller tends to catch the front of a circle and break through the elbow, and the PID controller is characterized by lag response, obvious overshoot and insufficient vibration suppression capability for specific frequency. Although compensation can be performed by means of feedforward control, notch filters, etc., this increases the complexity and debugging difficulty of the system. The related technology adjusts the control action according to the current or historical load position error, when external disturbance force exists, the control response has time difference, vibration diffusion cannot be restrained in time, the amplitude of the load position deviating from the expected track is increased, and the working stability of precision equipment is affected. The single-degree-of-freedom air flotation vibration isolation system relates to multi-component coupling of a throttle valve, an air cavity, a load and the like, has complex dynamic characteristics, is difficult to construct an accurate model which is fit with reality by a control method, and has difficult control precision to meet the requirements of high-end precision equipment. The position prediction model of the related technology mostly adopts a single-point prediction mode, the output result is easy to be interfered by noise, the randomness is strong, the smooth trend reflecting the position change in a future period cannot be provided, and the position prediction model is difficult to be directly used for control strategy design. The control amount calculated by the control strategy often does not fully consider the physical constraint of the executing mechanism, and the condition that the amplitude exceeds the rated range or the change rate is too fast can occur, so that the executing mechanism cannot accurately respond to the instruction, and even the system is interrupted to continuously run due to long-term overload damage. Disclosure of Invention The invention provides a dynamic surface control method of an active vibration isolation system, which solves the problems of response lag and nonlinear compensation deficiency of the traditional method by establishing an accurate throttle valve-air cavity-load dynamic model and combining data-driven advanced prediction with model-driven robust control, and improves the dynamic performance and control precision of the system. The method comprises the following steps: S101, establishing a state space model of an active vibration isolation system, and defining a system state variable and a control input; S102, collecting time sequence data of a load position, and preprocessing the time sequence data to obtain smooth denoised load position data; S103, training an LSTM neural network by using the load position data after smooth denoising to obtain a load position prediction model, wherein in the training process, the input is a historical load position sequence, and the output is a Gaussian weighted future load position trend value; s104, in the real-time control process, acquiring a current system state variable, and predicting a load position by using a load position prediction model to obtain a predicted load position; S105, defining a load position tracking error according to a predicted load position and an expected load position, and designing a first virtual control law; s106, gradually defining a follow-up virtual control law and tracking errors through a dynamic surface control method based on the first virtual control law, wherein the follow-up virtual control law comprises a load speed tracking surface control law, an air cavity pressure tracking surface control law, a throttle valve core displacement tracking surface control law and a thr