A neural-network-based space vector PWM of a five-level voltage-fed inverter

Abstract

The paper describes an artificial neural network (ANN) based space vector pulse width modulation (SVM) for a five-level voltage-fed inverter. Basically, it uses two multilayer perceptron (MLP) type neural networks. The first ANN uses the amplitude and angle of the reference voltage vector to determine the nearest three vectors (NTV) of the inverter by identifying the triangle wherein the reference vector lies. The second ANN is used to calculate the duty cycles of the three space vectors. An erasable programmable logic device (EPLD) synthesizes the PWM waves. The main advantages of this approach are the fast and simple implementation of the highly complex SVM algorithm for multilevel inverters without loosing precision compared to the conventional DSP-based SVM algorithm. Performance of the inverter using the proposed ANN-based SVM has been investigated extensively, and the results are found to be excellent. The principle described in the paper can be easily extended to an inverter with higher number of levels.