Weighting factors design in model predictive direct torque control based on cascaded neural network

An innovative control strategy is introduced in this paper for model predictive direct torque control (MPDTC) of a permanent magnet synchronous motor (PMSM) inverter, which is based on a cascaded neural network for the automatic selection and online tuning of weighting factors. Specifically, the lower network in the proposed two-stage cascaded network is trained by utilizing the weighting factors and the corresponding converter parameters under a specific working condition to establish data substitution models under that working condition and determine the fitness function for obtaining the optimal weighting factors. The upper network of the cascaded network is trained using various working conditions and their corresponding optimal weighting factors calculated by the lower network, with the goal of achieving real-time tuning of the weighting factors under variable working conditions. By employing a data-driven method to design the weighting factors at each working condition offline, this method reduces online tuning to a single calculation in the upper network and minimizes the online calculation burden. This approach enables the weighting factors to be adjusted in real-time according to changes in the working conditions, leading to enhanced accuracy of weighting factor optimization. Experimental results demonstrate that the proposed strategy improves the overall performance of the inverters.