Control Enhancement of Traction Electric Drives Using Neural Network Predictive Controller

This paper investigates the use of a hybrid Recurrent Neural Network to reproduce the behavior of a nonlinear long-horizon Model Predictive Controller (MPC) used in traction motor drive systems. The goal is to assess the operational validity and control performances of such neural network based predictive controller (further referred to as Full Neural Network MPC), and to compare against Field Oriented Control (FOC), the current industry standard. Based on simulation results using a Model in the Loop (MiL) environment, it is shown that the proposed FNN-MPC can properly learn the behavior and characteristics of the nonlinear long-horizon MPC while being more computationally implementable, with still better control performance and quality than FOC.