Control Set Reduction for PMSM Predictive Controller via Assisted Learning Algorithm

Abstract

This paper introduces innovative methods for reducing the control set in finite control set model predictive control of the Permanent Magnet Synchronous Motor powered by a 3-level voltage source inverter. The primary objective of this reduction is to address a crucial factor in the computational burden of the control algorithm-the exponential growth in the number of potential switching state combinations forming the controller’s control set with an increasing prediction horizon length. The proposed methods aim to decrease the number of switching states necessary for evaluation, mitigating the aforementioned exponential growth. These methods leverage information about the controller’s behavior. The first method relies solely on the count of transitions between individual switching states. Additionally, the second method incorporates information about the states of the controlled motor to construct a decision tree, forming the new control set. The behavior of the controllers with reduced and complete control sets is compared in the simulation experiment, emphasizing the proper tracking of the requested angular speed and their overall computational complexity.This paper introduces innovative methods for reducing the control set in finite control set model predictive control of the Permanent Magnet Synchronous Motor powered by a 3-level voltage source inverter. The primary objective of this reduction is to address a crucial factor in the computational burden of the control algorithm-the exponential growth in the number of potential switching state combinations forming the controller’s control set with an increasing prediction horizon length. The proposed methods aim to decrease the number of switching states necessary for evaluation, mitigating the aforementioned exponential growth. These methods leverage information about the controller’s behavior. The first method relies solely on the count of transitions between individual switching states. Additionally, the second method incorporates information about the states of the controlled motor to construct a decision tree, forming the new control set. The behavior of the controllers with reduced and complete control sets is compared in the simulation experiment, emphasizing the proper tracking of the requested angular speed and their overall computational complexity.