Prediction-Error-Driven Finite-Control-Set Model Predictive Control For PMSM Drives

Finite-control-set model predictive control(FCS-MPC)is an emerging candidate for the control of permanent magnet synchronous machine(PMSM).The FCS-MPC has the merits of intuitive design,fast response,modulator-free,and easy inclusion of multiple objectives and nonlinearity.However,the performance of the FCS-MPC relies on the accuracy of modeling.The performance will be deteriorated when the parameters mismatch.Meanwhile,parameter identification and sensorless control are essential in many PMSM drive applications nowadays due to the ever-increasing demand for lower cost,higher efficiency and higher reliability,whereas most researches on the FCS-MPC of the PMSMs are restricted within the tracking of the current and torque references.Once the parameter or position estimation is demanded,the traditional methods which incorporates the modulators are combined with the FCS-MPC mechanically,which on one hand compromises the benefits of both the FCS-MPC and the estimation methods,and on the other hand brings difficulties in implementation and aggravates the already heavy computational burden.Therefore,a prediction-error-driven FCS-MPC(PED-MPC)method is proposed to tackle the problem of parameter dependency by adaptation.Then,the solutions to the problems in three different PMSM control scenarios,i.e.,the parameter-mismatch disturbance rejection in the current control,the parameter identification in the high-efficiency control and position-sensorless control,are proposed.The proposed methods,which are concluded below,compose the PED-MPC methodology in the PMSM control.1.To solve the problem of performance degradation in the case of parameter mismatch,a disturbance-rejecting FCS-MPC(D-MPC)and a nonparametric FCS-MPC(N-MPC)are proposed based on the PED-MPC.The D-MPC decreases the control error by accounting all modeling mismatches into the lumped voltage disturbances and compensating the mean prediction error.The N-MPC uses lumped-parameter model instead of the conventional voltage model,so that the real machine parameters are not needed to approach the performance of the FCS-MPC with accurate model.2.The PED-MPC-based two-parameter and three-parameter identification methods are proposed to estimate the inductances and magnet flux linkage online in order to optimize the maximum-torque-per-Ampere and flux-weakening models.As a result,the efficiency of the PMSM and the maximum speed under the same current and voltage constraints are improved.Also,the dead-time compensation and vector delay compensation methods are proposed to improve the estimation accuracy.3.Based on the discrete switching nature of the FCS-MPC,which is without modulators,the full-speed-range PMSM sensorless control strategies are proposed in both cases when system parameters are known or unknown.When the parameters are known,a d-axis and a q-axis current prediction-error-based position estimation method(named d-PED and q-PED respectively)that extracts position information from the back EMF and the inherent excitation of the discrete switching actions respectively,are proosed.position estimation of the whole speed range can be realized by the combination of the d-PED and q-PED methods.Compared with the conventional modulator-based methods,the proposed strategy achieves the identical performance with less signal injection,lower ripple currents and acoustic noises,better disturbance-rejecting ability and less complexity in the filter design.In the case when the parameters are unknown,a q-NPED method is proposed based on the q-PED,so as to estimate the position in the whole speed range without using the real parameters and without switching between two methods in different speed regions.Different from the conventional prediction control methods,the PED-MPC uses the model information for prediction and control,and uses the prediction error to update the model.In this way,the control process and the model identification process are integrated.Furthermore,the inherent excitation within the discrete switching actions can be extracted to drive the estimation,so that the estimation accuracy is guaranteed without additional signal injections.All those PED-MPC methods are discussed and verified by analyses,simulations and experiments.Comparative experiments are also designed to compare the PED-MPC methods with the conventional methods and other methods in the literature.The results verify the feasibility of the methods,and demonstrate the features and benefits of the PED-MPC.