Research On Finite-Control-Set Model Predictive Control For Permanent Magnet Synchronous Motor Drives

Permanent Magnet Synchronous Motor(PMSM)achieves the frequency conversion control by adopting the power electronic frequency converter.PMSM has the characteristics of compact structure,reliable and efficient operation.It is widely used in defense industry,equipment manufacturing,transportation and any other high performance applications.The speed control system with demand of frequent start and stop,repeated acceleration and deceleration,and clockwise and counterclockwise running pursuits the dynamic performance,multivariable and multi-constraint control,such as numerically-controlled machine tool,robot motion servo,and electric vehicle.Finite-control-set model predictive control(FCS-MPC)is a nonlinear controller which can realize the simultaneous control of multiple variables and contain multiple nonlinear constraints.The exhaustive forecasting in limited number of control states is adopted to choose the optimal control output.This is consistent with the discrete characteristics of power electronic systems.Besides,the most suitable switching state is decided in control period according to the cost function design and the constraint of modulator is eliminated.Thus FCS-MPC expresses the excellent dynamic control performance and is received extensive attention in high dynamic AC motor drive field.However,there are still some issues to be solved in the conventional FCS-MPC for PMSM drives,such as multitimescale subdivision of prediction sequence,multi-target collaborative optimization of cost function,improvement of steady-state performance and digital system real-time implementation.Therefore,the two-level voltage source inverter and surface mounted PMSM are chosen as research object and the research of finite-control-set model predictive control(FCS-MPC)of PMSM drives is studied in this paper.Based on the fundamental principle of FCS-MPC,the working mechanisms of prediction model and cost function are discussed.The internal relation of control variable and system performance in each part of MPC is clarified.The performance that need to be optimized are clear.The complete prediction model and control scheme that contain the issues to be optimized are conducted,which provides the unified architecture and theoretical support for the next study and analysis.Aiming at the simplified of multi-timescale coupling of prediction sequences and the dynamic prediction accuracy is reduced,based on the typical cascaded structure of speed loop and current loop,a MPC with multi-timescale optimization for PMSM drives is proposed in this paper.The complete analysis of all prediction sequences over full timescale for fast and slow sampling models is conducted.The multi-timescale prediction model for PMSM is built.The influence mechanism of coupling prediction sequences in different timescale subsystems is studied.The relation between coupling prediction sequences and system performance is clarified.Since the coupling prediction sequences are unable to calculate accurately,an estimation method based on linear function and virtual prediction instants is proposed to repair the lacking of multi-timescale coupling prediction sequence.The prediction accuracy is improved and the dynamic performance is optimized.Aiming at the coordination issues of multi-target in cost function,the dynamic cost function with multi-objective collaborative optimization is proposed in this paper.The influence of current tracking target,current limiting constraint and switching frequency optimization on voltage vector selecting in cost function is discussed.The relation of weighting assignment and system performance is clarified.Considering the dynamic process of the speed-regulation,the MPC based on fuzzy dynamic cost function is proposed to solve the fixed weighting assignment in conventional method.The proposed method can adjust the weighting assignment dynamically according to the speed deviation and variation based on the fuzzy logic.The dynamic response is improved and the average switching frequency is reduced at the same time.The system realizes the comprehensive coordinated operation.The steady-state performance is weaken in the conventional method because the prediction is conducted by prediction model only and the historical value of the control error is lack.An aided prediction method based on statistical knowledge is proposed in this paper and the historical data in feedback is considered into the control system.The analysis that prediction sequences select the voltage vector in the cost function is conducted to clarify the feature that prediction sequences affect the steady-state performance.Considering the computational burden of real-time implementation,the real-time first-order grey model that need small amount of data is introduced to conduct the first step aided prediction.The model based prediction is the second step prediction.The unnecessary switch motion in steady-state is optimized.The steady-state performance is improved while ensuring the dynamic performance of system.Besides,the switching frequency is optimized in some degree.The exhaustive forecasting for all switching states of inverter is adopted in FCS-MPC to conduct the online optimization,leading to the high algorithm complexity and making it difficult to realize the real-time implementation.This paper proposes a matrix optimal operation based on parallel unrolling and maximized parallel pipeline architecture.The key procedure of MPC that influences the real-time performance is analyzed and optimized.The proposed method is designed to optimize the computational time of proposed method at the cost of additional digital resources and on-chip power.The issue of excessive consumption of computational time for iteration of multiple voltage vectors in the conventional method is solved.The real-time performance and operation efficiency optimization are achieved.