Generalized switching table of the DTC of an induction motor determined by reinforcement learning

This paper proposes a new method of reinforcement learning. (RL) method based on Q learning algorithm applied to the direct torque control (DTC) of an induction machine powered by a multilevel inverter. The Q-learning algorithm is used in direct torque control with multilevel inverters to find optimal actions at data states off-line among several available discrete actions, updating the action (voltage vector) awards received. The results obtained by the reinforcement learning (RL) method are validated by simulation using Matlab/Simulink. The generalization of this approach to N levels without any difficulty (increase of the level of inverter and therefore of the number of voltage vectors to be selected) gives the reinforcement learning method an appreciable advantage. It makes it possible to determine the switching table automatically regardless of the number of voltage levels of the inverter used.