Improved neural network current regulator for VS-PWM inverters

In the paper the application of an off-line trained neural network (NN) working as a current regulator for a three phase PWM inverter is presented. It is shown that this type of NN regulator has operation characteristics similar to delta modulators which cannot apply the zero voltage vector. When the NN output signals are sampled not at the same instant but with a phase shift, i.e. 2/spl pi//3, the regulator can select zero vectors and always chooses adjacent active vectors. This guarantees the lower RMS error for the same mean switching frequency. Finally, two different learning techniques for such current regulators are discussed and compared.<>