Neural network parameter optimization for model predictive direct speed control of PMSM

In this paper, we propose a neural network-based approach to optimize the parameters involved in the model predictive direct speed control (MPDSC) of a permanent magnet synchronous motor (PMSM). Model predictive control is a widely used technique in motor control to enhance system performance by predicting future behavior and determining control actions accordingly. However, the effectiveness of MPDSC is highly dependent on the weighting parameters of the cost function. Optimizing these parameters in a complex PMSM MPDSC control system under diverse working conditions presents a challenging task, as conventional methods may struggle to efficiently find the optimal parameters. To address this issue, we design a neural network optimization framework. Initially, an original optimization algorithm is employed to identify finite optimization parameters under various working conditions. Subsequently, the optimal parameters obtained for the finite working conditions are utilized to train a neural network. The trained network is then capable of predicting the optimal parameters across the entire range of working conditions. The proposed method is validated through simulations conducted in MATLAB, demonstrating its effectiveness in optimizing the parameters for MPDSC.