Research On Predictive Control Technology For Large Aperture Telescope Drive Control System

In recent years,the proliferation of low-orbit micro-satellites and space debris has significantly increased the safety hazards associated with on-orbit spacecraft operations.Consequently,it has become imperative to establish a space surveillance system capable of detecting and identifying small space targets.The ground-based large-aperture widefield optical detection telescope will serve as the core equipment for the realization of this space surveillance system.However,as the telescope aperture increases,the servo control system based on brushed DC motors can no longer meet the tracking requirements of large telescopes.Due to their high power density,high torque-to-inertia ratio,and excellent low-speed tracking performance,permanent magnet synchronous motors are gradually becoming a preferred choice for the servo drive systems of large telescopes.It is noteworthy that the utilization of permanent magnet synchronous motors can significantly improve the performance and reliability of the servo systems for large telescopes.Currently,the main high-performance closed-loop control methods for permanent magnet synchronous motors(PMSMs)are field-oriented control(FOC)and direct torque control(DTC).FOC can achieve good dynamic and static performance in medium and low-power applications,but when the system’s switching frequency is low or the motor speed is high,the stability of the system may be difficult to maintain.In FOC,the control algorithm and PWM strategy are independently designed,which leaves room for further improvement of the overall performance of the system.Unlike FOC,DTC does not have an independent PWM stage and can achieve fast dynamic performance with a simple control structure.However,DTC must rely on high sampling frequency to maintain steady-state performance at an acceptable level.The model predictive control(MPC)strategy has demonstrated significant advantages in high-performance control of permanent magnet synchronous motors,owing to its fast dynamic response,multivariable control capabilities,and ability to handle nonlinear constraints with ease.Currently,MPC is widely regarded as the third high-performance control strategy that is most likely to be widely applied in the fields of power electronics and motor control,following FOC and DTC.However,the MPC strategy for permanent magnet synchronous motors is still not perfect,as it suffers from poor parameter robustness and large steady-state ripple,among other issues.In this paper,we aim to address these problems by leveraging the idea of "enabling" MPC with machine learning,and focusing on the research of model current predictive control for permanent magnet synchronous motors.In response to the problem of poor robustness of finite-set current predictive control parameters,a non-intrusive parameter identification method based on machine learning is proposed to accurately estimate motor parameters.The influence of parameter mismatch on the PMSM current prediction model is analyzed,and the current data under different parameter mismatch states are preprocessed.A function mapping relationship between the degree of inductance mismatch and the actual performance of the current is established using machine learning.The actual parameter values can be obtained when the current performance data is known.Experimental results show that the proposed method can accurately estimate the model parameters and has high realtime performance.In response to the issue of steady-state current ripple in finite-set model predictive control(MPC),the concept of data-driven approach is introduced to train a Gaussian process regression(GPR)model based on suitable features and standardized datasets.The GPR algorithm is used to learn the predictive model from the data,improving the accuracy of the predictive model and achieving better control.Joint simulation using MATLAB/Simulink and Regression Learner is conducted,and experiments with actual systems are carried out.It is found that compared to traditional model predictive control,GPR prediction has higher accuracy,reduces steady-state error,effectively suppresses the impact of model mismatch on the control system,and has better dynamic response.This study verifies the feasibility of applying data-driven methods in predictive control.