Multilevel Neural Network DTC with Balancing Strategy of Sensorless DSSM Using Extended Kalman Filter

Abstract

This paper presents direct torque control based on artificial neural networks of a double star synchronous machine without mechanical speed and stator flux linkage sensors. The estimation is performed using the extended Kalman filter, which is known for its ability to process noisy discrete measurements. The proposed approach consists of replacing the switching tables with one artificial neural network controller. The output vector of the artificial neural network controller is directed to a multilevel switching table to decide which reference vector should be applied to control the two five-level diode-clamped inverters. This inverter topology has the inherent problem of DC-link capacitor voltage variations. Multilevel direct torque control based on a neural network with balancing strategy is proposed to suppress the unbalance of DC-link capacitor voltages. The simulation results presented in this paper highlight the improvements offered by the proposed control method based on the extended Kalman filter under various operating conditions.