CN-121979053-A - Vibration active control method based on PPO-HAFxLMS algorithm

CN121979053ACN 121979053 ACN121979053 ACN 121979053ACN-121979053-A

Abstract

The invention discloses a vibration active control method based on a PPO-HAFxLMS algorithm, and relates to the field of vibration active control. The method comprises the steps of determining a target vibration frequency band to be suppressed and a vibration attenuation standard, determining configuration of related hardware in a vibration active control system, capturing external disturbance signals received by a controlled object, and generating a required control voltage signal in real time by using a PPO-HAFxLMS algorithm based on time domain data of the external disturbance signals to achieve vibration suppression. According to the method, a feedback channel is added to form HAFxLMS algorithm based on the traditional FxLMS algorithm, HAFxLMS algorithm is used for optimizing filter coefficient and step length update, step length and filter weight are dynamically adjusted according to environment change through PPO reinforcement learning algorithm according to the correlation of error signals and input signals, so that convergence speed and steady-state error of vibration control are optimized, and vibration suppression effect is remarkably improved.

Inventors

GAO DAYONG
YAO HONGLIANG
YANG JINBO
LI JIANLEI

Assignees

东北大学

Dates

Publication Date: 20260505
Application Date: 20260129

Claims (8)

1. The vibration active control method based on the PPO-HAFxLMS algorithm is characterized by comprising the following steps of: determining a target vibration frequency band to be suppressed and a standard of vibration attenuation, thereby determining the configuration of related hardware in the vibration active control system; Capturing an external disturbance signal received by a controlled object; Based on the time domain data of the external disturbance signals, the vibration active control system generates the required control voltage signals in real time by using a PPO-HAFxLMS algorithm, so that vibration suppression is realized.
2. The vibration active control method according to claim 1, wherein the configuration of the related hardware includes a sensor including one acceleration sensor installed near an external disturbance signal as a reference sensor and another acceleration sensor installed in a vibration suppressing area as an error sensor, and the type, number, and layout of actuators installed in the vibration suppressing area of the controlled object.
3. The vibration active control method according to claim 2, wherein the method of capturing the external disturbance signal received by the controlled object is to capture the external disturbance signal received by the controlled object from the acceleration signal acquired by the reference sensor.
4. The vibration active control method according to claim 1, wherein the vibration active control system comprises a channel P (Z) from an external disturbance signal x (n) received by the controlled object at any n time to an error sensor measuring vibration response signals, a channel model S (Z) from a secondary channel, namely a control voltage signal, to the vibration response signals collected by the sensor, a feedforward filter W 1 (Z) for generating a feedforward control signal y 1 (n) and a feedback filter W 2 (Z) for generating a feedback control signal y 2 (n), wherein Z represents a complex variable in Z transformation, y 1 (n) is used for responding to vibration caused by external disturbance, and y 2 (n) is used for correcting the response of the vibration active control system by adjusting the error signal e (n) of vibration response.
5. The vibration active control method according to claim 4, wherein the vibration active control system generates a desired control voltage signal in real time using a PPO-HAFxLMS algorithm based on time domain data of an input vibration signal, so that the vibration signal is suppressed, comprising: Step 4.1, according to external disturbance signals x (n) received by a controlled object at any n time, actual vibration response time domain data d (n) of the controlled object is obtained through HAFxLMS algorithm calculation, and then error signals e (n) of vibration response are obtained through calculation, wherein the calculation process is as follows: (1) (2) In the formula, Representing the vibration control amount of the actuation voltage signal y (n) acting on the controlled object via the secondary channel S (z); Step 4.2, updating the feedforward filter W 1 (z) and the feedback filter W 2 (z) through HAFxLMS algorithm according to the calculated error signal e (n) so as to optimize the generation of a control voltage signal and reduce the vibration error of a controlled object; And 4.3, generating a control voltage signal by utilizing HAFxLMS algorithm, and adaptively optimizing the step length of the feedforward filter W 1 (z) and the step length of the feedback filter W 2 (z) by utilizing PPO reinforcement learning algorithm in the process of generating the control voltage signal.
6. The vibration active control method according to claim 5, characterized in that the feedforward filter W 1 (z) is updated by HAFxLMS algorithm as follows: (3) wherein W 1 (n) is the coefficient of the feedforward filter at the time n, and W 1 (n+1) is the coefficient of the feedforward filter at the time n+1; the step length parameter is used for controlling the updating speed of the feedforward filter coefficient, and x' (n) is a signal of an input signal x (n) after being identified by a secondary channel; the feedback filter W 2 (z) is updated by HAFxLMS algorithm as follows: (4) wherein W 2 (n) is the coefficient of the feedback filter at the time n, and W 2 (n+1) is the coefficient of the feedback filter at the time n+1; step length parameters of the feedback filter are used for controlling the updating rate of the feedback filter coefficients; recognition of the actuation voltage signal, i.e. the control input y (n), via the secondary channel S (z) The control input signal identified by the error sensor is numerically 。
7. The vibration active control method of claim 6, wherein generating the control voltage signal using HAFxLMS algorithm comprises: First, a feedforward control signal y 1 (n) is generated based on an input signal x (n) and a feedforward filter W 1 (z) according to equation (5): (5) In the formula, Is that Is a transpose of (2); Then, based on the error signal e (n) and the secondary channel S (z), the feedback control signal y 2 (n) is calculated according to equation (6): (6) In the formula, Is that Is a transpose of (2); Representing the identification of the secondary channel S (z); Then, the feedforward control signal y 1 (n) and the feedback control signal y 2 (n) are synthesized to obtain a total control voltage signal y (n) acting on the controlled object: (7) Finally, after the total control voltage signal y (n) is processed by the FIR filter, the driving actuator generates a control force, acts on the controlled object and generates a final control signal y' (n) through the secondary channel S (z): (8) Where S 1 (n) is the impulse response of the secondary channel S (z).
8. The method according to claim 7, wherein the step size of the feedforward filter W 1 (z) and the step size of the feedback filter W 2 (z) are adaptively optimized by the PPO reinforcement learning algorithm in the process of generating the control voltage signal in step 4.3, and the method comprises the steps of designing the state space s (n) to be: (10) wherein e (n-1) is a historical vibration error; the action space a (n) is defined as: (11) Wherein, Δmu 1 (n) and Δmu 2 (n) respectively represent the adjustment amount of the step length of the feed-forward channel and the adjustment amount of the step length of the feedback channel generated by the PPO reinforcement learning algorithm, which are used for optimizing the weight update formula of the filter: (12) Wherein W 11 (n+1) and W 22 (n+1) are coefficients of the feedforward filter and the feedback filter at the time of n+1 after PPO reinforcement learning compensation adjustment respectively; the bonus function r (n) is designed to: (14) Where λ is a trade-off factor for balancing error minimization and step size stability; the optimization objective function of the PPO reinforcement learning algorithm is: (16) In the formula, Is the importance sampling ratio; is an advantage function for measuring the goodness of the current action, and epsilon is a clipping parameter for limiting the updating amplitude of the strategy.

Description

Vibration active control method based on PPO-HAFxLMS algorithm Technical Field The invention belongs to the technical field of vibration active control, and particularly relates to a vibration active control method based on a PPO (Proximal Policy Optimization, near-end strategy optimization) reinforcement learning algorithm and a HAFxLMS (Hybrid Adaptive FxLMS, hybrid self-adaptive FxLMS) algorithm. Background With the increasing demand for vibration control technology, vibration problems are becoming one of the key considerations in many engineering applications. The effectiveness of vibration control not only directly affects the stability and life of the device, but also affects the overall performance of the system. In many fields, control of vibration has become a critical factor in improving product quality and performance. Particularly in the control of low frequency vibrations, due to its complexity and difficulty, is one of the main challenges faced by vibration control techniques. In the field of vibration active control, the FxLMS algorithm is widely applied to a vibration active control system due to the characteristics of simple and efficient structure and easy realization, and the algorithm has been applied to a plurality of fields such as vehicle vibration, industrial equipment, building structures and the like by updating a filter coefficient through a minimized error signal, but has the key defects that a fixed step updating filter is adopted, a larger step can realize rapid convergence but easily causes larger steady-state error, a smaller step can reduce steady-state error but causes slow convergence speed, and finally causes the performance of the vibration active control system to be reduced. Aiming at the fixed step length problem of the FxLMS algorithm, partial improvement schemes (such as an adaptive control method based on the LMS algorithm) exist in the prior art, but the problems of insufficient step length adjustment, poor instantaneity and the like still exist in the schemes, and the control requirement of a complex vibration scene is difficult to meet. The prior related patent technology also has the defects that the method still depends on fixed step length and lacks a self-adaptive regulation mechanism for dynamic change environment, thus limiting the adaptability and precision of a control algorithm, the method is characterized in that the self-adaptive ship vibration control method based on differential calculation is used for adjusting control parameters in real time through a differential algorithm to optimize vibration suppression effect, the algorithm has higher calculation complexity, is easy to influence the real-time response and control precision of the control algorithm, and is difficult to migrate and applied to different dynamic vibration scenes under the high-frequency change environment. In summary, when the existing vibration active control technology is applied to a complex dynamic vibration source, the problems of insufficient step length self-adaptive adjustment, missing multi-channel control, difficulty in balancing the calculation complexity and the real-time performance and the like generally exist, and the control requirements of high precision and high real-time performance cannot be met, so that research and development of a vibration active control algorithm capable of being adjusted in real time are particularly necessary. Disclosure of Invention In view of this, the invention provides a vibration active control method based on PPO-HAFxLMS algorithm. According to the method, on the basis of a traditional FxLMS (Filtered-x LEAST MEAN Square) algorithm, a feedback channel is added to form a HAFxLMS (Hybrid Adaptive FxLMS, hybrid self-adaptive FxLMS) algorithm, and the step length and the filter weight are dynamically adjusted through a PPO reinforcement learning algorithm, so that the convergence speed and steady-state error of a vibration control system are optimized, and the vibration suppression effect is remarkably improved. The technical scheme of the invention is as follows: A vibration active control method based on PPO-HAFxLMS algorithm comprises the following steps: determining a target vibration frequency band to be suppressed and a standard of vibration attenuation, thereby determining the configuration of related hardware in the vibration active control system; Capturing an external disturbance signal received by a controlled object; Based on the time domain data of the external disturbance signals, the vibration active control system generates the required control voltage signals in real time by using a PPO-HAFxLMS algorithm, so that vibration suppression is realized. Optionally, according to the vibration active control method, the configuration of the related hardware comprises types, numbers and layout modes of sensors and actuators, wherein the sensors comprise one acceleration sensor which is installed near an external disturbance si