A neural network interpolator for the improvement of power rectifiers control

The studied rectifier consists of an input AC second order filter, a four switch matrix and an output DC filter. The objectives of the rectifier are: regulation of the DC current; sinusoidal input current waveform; and optimal power factor. In order to satisfy these objectives, the authors have designed a control algorithm which includes an internal closed loop for the control of the AC current and an external loop respective to the control of the mean value DC current. Because of algorithm convergence problems at the start, a second control algorithm is temporarily used to obtain a nonnull DC current. The presented results show degradations of electrical quantities due to the transition between these two algorithms. To compensate these degradations, a neural network controller is designed to interpolate through the two control algorithms. Simulation results show better signal waves with a reduced implantation dimension