High Performance Model Predictive Control for PMSM System Using Bayesian Ascent and Gaussian Process

Model predictive control has been widely developed for electrical drive systems. The weighting factors are key parameters affecting the control performance of the motor. This paper proposes a Bayesian ascent and Gaussian process method for the optimized weighting factors calculation. The root mean square error of the current in the d-q axis is taken as a criterion. A data-driven probabilistic proxy model is formed based on the Gaussian process, and the optimal weighting factor is obtained by Bayesian ascent. Grid sampling, random sampling, and sampling in the maximum possible region are integrated to improve sampling reliability. Furthermore, the maximum torque per ampere calibration and motor parameter identification can be implemented by the proposed method as well. Finally, a new perspective of global parameter optimization based on data probability is proposed. The effectiveness of the proposed methods is verified by experimental results.