ANN-Assisted Model Predictive Control Method for IPMSM Drive Using Torque-Deviations Mapping

The high performance of conventional model predictive control (CMPC) for electric drives depends on the fidelity of the machine's parametric model. However, the parameters of the interior permanent magnet synchronous motor (IPMSM) exhibit nonlinear variations with operating conditions, such as magnetic saturation and cross-coupling effects, leading to machine-model mismatch and degradation in CMPC performance. To solve this problem, the model-free predictive concept has been investigated to increase the reliability of torque and flux predictions. Nevertheless, the computational complexity and memory requirements are the main inconveniences of this strategy. This article proposes an artificial neural network (ANN)-assisted MPC technique for the IPMSM using torque-deviations mapping. First, the raw torque and flux maps are obtained by the finite-element analysis (FEA) to directly calculate the torque deviation caused by the possible control actions, considering the motor's nonlinear characteristics. Then, these control actions are filtered to omit those unnecessary from the point of view of MPC with a duty modulator, resulting in an ANN with excellent training performance and a simple structure of trained nets. The proposed ANN-based MPC is evaluated in both simulations and hardware-in-the-loop (HIL) experiments using an 80-kW IPMSM to verify its feasibility and effectiveness.