Application of Neural Network in Parameters Optimization of Permanent Magnet Synchronous Motor Model Predictive Control

This paper proposes a method to realize the parameters optimization of permanent magnet synchronous motor (PMSM) model predictive control (MPC) using neural network (NN). The first step of the method is to use different parameter combinations to perform multiple simulations (or experiments) of the MPC algorithm, and extract the key performance indicators (such as average switching frequency of the inverter, total harmonic distortion, etc.) of the system. Then, train the NN with the acquired data. The trained NN performs as a substitute for the simulation model, and the performance indicators of the system can be estimated quickly and accurately corresponding to arbitrary parameter combinations. Therefore, user can define any fitness function composed of performance indicators, and the optimal parameter combination minimizing the fitness function can be found automatically. Finally, the parameter combinations designed for three different fitness function were verified by simulation, and the predicted performance indicators turned out to be close to the simulation model, with error less than 4%.