Three phase active rectifier power conditioning using neural network system identification

This paper describes a three phase active rectifier control technique which is capable of delivering nearly sinusoidal current waveform with unity power factor and provides regulated DC voltage. A simple on-line real time control algorithm results in fast response to load change or DC voltage reference command change and produces low harmonic currents. The boost topology is employed in this circuit in order to control the AC line currents. And the DC output voltage is controlled by regulating the AC current command. In this control scheme, switching pattern is directly determined by duty ratios which are calculated from the control equation. One disadvantage of this control scheme is that maximum magnitudes of the control voltages are limited to 1/2 Vdc. Also, the control coefficients are derived from the circuit parameters which are not exactly known and discretized with sampling interval. In order to solve the problems, parameter estimation is employed to make the proposed algorithm capable of self-adjusting. The Hopfield neural network parameter estimation is used for adjusting the parameters in the control equation. To predict the performance of the time varying system, on-line estimation is applied with adopting forgetting factor. Simulation results show the effectiveness of the rectifier performance with parameter estimation.