Research On Model Predictive Current Control Of Vehicle Permanent Magnet Synchronous Motor Based On Neural Network

In the face of the global energy and environmental crisis,the development of new energy vehicles is crucial,Permanent Magnet Synchronous Motor(PMSM)as the mainstream control motor of new energy vehicles,its control algorithm has an important impact on vehicle energy consumption,driving efficiency,safety and many other performances.Model Predictive Current Control(MPCC)can meet the performance requirements of low switching frequency,multi constraint control,and high dynamic response of automotive PMSM by integrating various nonlinear constraint conditions.However,there are still problems such as complex control laws and long calculation time.Through data learning and network training,neural networks can approximate complex nonlinear mapping relationships,and have the advantages of large-scale parallel processing and fast computing speed,which can replace complex control algorithms.This paper conducts research on neural network-based model predictive current control,and the main contents are as follows:(1)In order to meet the requirements of fast dynamic response and multi-constraints control for automotive PMSM,the Finite Control Set Model Predictive Current Control(FMPCC)model based on seven alternative voltage vector control sets is established under the synchronous rotation coordinate system.Through MATLAB/Simulink and the single-chip microcomputer platform,it is verified that the F-MPCC algorithm has the advantages of fast dynamic response and simple implementation process.It is also pointed out that the F-MPCC algorithm has the problem of large current pulsation in the steady-state control process.(2)Aiming at the problem of large current pulsation of F-MPCC,the alternative voltage vector control set is expanded,and the three-phase duty cycle of extended alternative voltage vector control set is processed into an offline table form by combining the spatial vector pulse width modulation technology.When system is running,the on-off signal of each switching device can be obtained by simply looking up the table,which effectively avoids the real-time calculation process of vector modulation.Simulation results show that compared with F-MPCC algorithm,Extensional MPCC(E-MPCC)algorithm based on the extension of the alternative voltage vector control set can significantly reduce current ripple in the control system.(3)Aiming at the cumbersome and time-consuming problem of finite set MPCC traversal solving process,the deep neural network multi-classification model is introduced.The F-MPCC,E-MPCC,and multi-step MPCC models based on deep neural networks are established by using a large amount of data generated during the operation process of the above systems.Through the joint simulation platform of MATLAB and python,it is verified that the neural network can accurately learn the control laws of the finite set MPCC mentioned above and has comparable performance.The real-time verification results of the microcontroller indicate that neural network is more suitable for alleviating the time-consuming problem of finite set multi-step MPCC with complex control laws.Compared to single-step MPCC,the real-time nature of neural networks is not superior.In order to ensure the stability of the multi-step MPCC of the neural network,the F-MPCC algorithm is used as an alternative for optimal voltage vector determination.A multi-step model predictive current control algorithm based on the combination of deep neural network and seven-voltage vector control set is proposed.The simulation results show that the algorithm can effectively suppress the runaway problem of multi-step control of neural network.(4)In order to further reduce the current pulsation of the MPCC control system,a Continuous Control Set Model Predictive Current Control(C-MPCC)model is established under the synchronous rotating coordinate system.A continuous set model predictive current control(NFC-MPCC)algorithm based on the combination of deep neural network and FMPCC is proposed to address the complex iterative solution process of C-MPCC optimal control variables,utilizing the advantages of simple implementation process and low computational complexity of deep neural network and F-MPCC.The simulation and microcontroller verification results show that the NFC-MPCC algorithm can effectively simplify the implementation process of the C-MPCC system under the premise of ensuring the high precision and high reliability of the control system.