Reinforcement-Learning-Based Parameter Look-Up Table Generating Method for Optimal Torque Control of Induction Motors

Abstract

In the optimal control of induction motors, it is a challenging task to maintain the optimal torque over the varying operation conditions. This paper proposes a parameter look-up table generating method, that can achieve an optimal torque over a wide range of currents and speeds, even though the commands of current are not set correctly. Based on the motor’s testing data, this method uses a reinforcement-learning algorithm to generate parameter look-up tables iteratively. Experimental results show that the proposed method can learn appropriate parameters from the running data to output an optimal torque. The comparative studies show that the proposed method can generate 5%-25% more torque than traditional model-based parameter estimation methods, over a wide range of currents and speeds. Furthermore, the proposed method has a faster convergence feature and a higher identification resolution than many conventional search-based methods.