Frequency-Adaptive Neural Network Based Virtual Flux Estimation for Sensorless Control of PWM Converters Under Unbalanced Conditions

Abstract

In this paper, sensorless control of three-phase pulse-width modulation converters under unbalanced grid conditions is proposed. This method exploits a new virtual flux (VF) estimator based on a frequency-adaptive neural network (FANN) configured as a quadrature signal generator (QSG). The inherent frequency-adaptive characteristic of the FANN-QSG is obtained using new developed frequency-locked loop. Two parallel FANNQSGs are utilized for simultaneous VF estimation and symmetrical components sequence separation in the proposed FANN-VF estimator. This approach avoids cascaded delays related to cascaded VF estimation and sequence separation structures. The proposed FANN-VF estimator has been tested as part of VF-based predictive current control (VF-PCC) system. Simulation results show that the FANN-VF estimator presented fast and accurate estimation of the grid frequency and the positive and negative sequence VF components. Moreover, the obtained VF-PCC achieved better current waveforms, under unbalanced conditions, than those obtained by the conventional PCC which uses grid voltage measurement.