Application of Kohonen´s self-organizing artificial neural networks to PWM inverter drives

When considering PWM waveform generation, the engineer has at his disposal a wealth of knowledge retrievable from the pertinent literature. Implementation of pulse width modulators appears to be simple in theory. In practice, there are some detrimental effects which result in a waveform degradation when not compensated for. The purpose is not solely to develop physical insights into these effects in order to design compensators for a waveform correction on a deterministic basis. The approach taken is to apply a self-organizing neural Kohonen network which supplies self-adapted inputs to the PWM inverter. Empirical testing of these inputs enables the net to fade out the waveform degradation. Computer studies showed that the proposed ANN is in a position to generate exact waveforms even with an unsupervised learning algorithm