Research On Model Predictive Torque Control Of Permanent Magnet Synchronous Motor Used In Vehicle Based On Data Drive

With the development of automobile electrification,the application of electric motors in automobiles has become more and more extensive.Permanent Magnet Synchronous Motor(PMSM)is one of the main choices for automotive motors.The direct torque control of the PMSM adopts the switching table to realize the qualitative selection of the voltage vector,but the torque ripple is large.Model predictive control(MPC)can realize the quantitative optimization of voltage vector selection,but the algorithm is complicated and the calculation burden is heavy.Data drive only needs the input and output information of the system to achieve the approximation of complex control laws,and can convert the time-consuming traversal optimization into offline model training and online fast inference,which can be used for learning and replacing MPC and help solve its problem of poor real-time performance.Therefore,this paper studies a data-driven model predictive control system,and realizes the function of replacing model predictive control by training a deep neural network(DNN),and the control performance is basically equivalent.The method to improve the DNN control performance by optimizing DNN parameters is studied,and two suppression strategies are proposed for the data-driven out-of-control problem,which enhances the robustness of datadriven control.Finally,the data-driven real-time performance is verified through the single chip microcomputer.The main research content of the article is as follows:Aiming at the disadvantages of DTC’s large torque ripple,this paper introduces model predictive torque control(MPTC),and carries out a non-dimensionalization cost function design.By data analysis and data excavating,one simplified set of candidate voltage vectors based on the zero-voltage vector and the switching table’s output is further proposed,which reduces the number of candidata voltage vectors from 7 to 2 or 0 and significantly reduces the computational burden of the system on the basis of ensuring the control performance.Aiming at the problem of slow dynamic tracking caused by zero voltage vector,an improved adaptive switching strategy was further proposed,which achieved a significant improvement in dynamic response performance.In order to further improve the performance of the PMSM-MPTC system,this paper introduces the concept of data-driven control.By selecting appropriate features and training a deep neural network(DNN)based on the constructed data set,the mapping law of MPTC’s selecting the optimal voltage vectors is established and learned.Through the co-simulation of MATLAB/Simulink and Python/Pytorch,it is verified that the control performance of the DNN is basically equivalent to MPTC with a certain control robustness.Analyzing the voltage vector sequence output by DNN,it is found that DNN has a very high selection accuracy for feasible solutions and high selection accuracy for optimal solutions,which verifies the feasibility of data driving in replacing predictive control,and reveals the reason why DNN can realize the normal operation of the motor system and the control performance is close to MPTC.In order to further improve the performance of data-driven control,the classification accuracy and generalization performance of the model are improved by optimizing the batch size parameter,which is conducive to accurately predicting the first term of the optimal voltage vector sequence.By increasing the training weight of the zero voltage vector,the goal of optimizing system switching frequency is achieved.This article further reveals the hidden dangers of out-of-control of data-driven control in dynamic state.In response to this problem,this paper proposes two out-of-control suppression strategies to enhance data-driven robustness and stability.And through the STM32H743 single-chip microcomputer,it is verified that the data drive has a great real-time advantage in replacing the multi-step predictive control.