Research On Control Strategy Of Permanent Magnet Linear Synchronous Motor Based On Machine Learning

In recent years,China’s industrialization and urbanization have accelerated,and motors play an essential role in the development process.Compared to rotary motors,linear motors do not require intermediate transmission links such as gears and screws and directly carry out linear reciprocating movements,which have the advantages of high speed,high precision,and low noise.Among them,the permanent magnet linear synchronous motors(PMLSM),which use permanent magnets as rotors,have been widely used as a new type of drive in modern industries such as lift traction,logistics and transportation and high-precision machine tool machining because of their small size,high efficiency,low mechanical losses and fast dynamic response.As PMLSM has a simplified mechanical structure,it is more susceptible to factors such as motor parameters and load variations.At the same time,PMLSM is a non-linear,strongly coupled,multi-variable system,and its non-linear modelling errors and position signal detection noise can have a large impact on the control performance of PMLSM.Traditional control methods cannot better cope with the effects of these situations,so this paper takes the PMLSM system as the research object and introduces machine learning algorithms into the field of PMLSM control,and the relevant research is as follows.Firstly,this paper describes the basic structure and operation principle of PMLSM.and establishes its mathematical model.To facilitate the analysis and reduce the complexity of the control algorithm design,coordinate transformation is introduced to complete the order reduction and electromagnetic decoupling transformation of the mathematical model of the motor.According to the basic principle of coordinate change,the mathematical model of PMLSM in the d-q coordinate system is obtained,the basic principle of vector control is introduced,and the vector control structure of PMLSM is built in this way.Secondly,the incremental width learning controller is designed to address the problem of poor self-adaptation of traditional PID controllers by utilizing the excellent non-linear approximation capability of incremental broad learning(IBL).Compared to fixed-gain PID controllers,IBL controllers are insensitive to parameter changes and load disturbances and can better meet control requirements without frequent adjustment of controller parameters.The design process of IBL controllers can be carried out without the knowledge of the mathematical model of the PMLSM system and other a priori knowledge and treats the whole system as a black box,reducing the difficulty of controller design.Then,using the MATLAB/Simulink simulation platform,the effectiveness of these control methods is verified.Then,to improve the PMLSM control performance,finite control set model prediction control was applied to the motor control field.To solve the problem of the tremendous computational effort of a multi-step prediction algorithm,stochastic configuration networks(SCNs)were combined with model prediction control,and a stochastic configuration networks-based model prediction controller was designed.The multi-step model prediction uses an exhaustive method to select the optimal switching vector.Its computational effort increases geometrically as the number of cycles grows,significantly increasing the burden on the hardware equipment.SCNs have excellent general approximation capabilities and can fit the multi-step prediction controller and achieve multi-step predictive control effects while reducing computational effort.Then,using the MATLAB/Simulink simulation platform,the effectiveness of these control methods is verified.Finally,a hardware-in-the-loop simulation platform of PMLSM system is constructed based on d SPACE,and the proposed control methods are experimentally verified.The experimental results show that the control method proposed in this paper also has good stability and control effects in practical applications.