Inverter switching patterns using a neural network

The authors present a method of switching mode selection for a digital current controller using a neural network. The neural network was trained to act as a vector selector for the inverter. The desired voltage vector was calculated from parameters of the induction motor, stator currents, and electromotive force. The backpropagation technique for training neural networks was used based on input/output data obtained from previous simulation studies. The results show excellent correlation between the output of the neural vector selector and the vector selection based on analytical methods. The neural network approach was faster than the previously used analytical methods, making it possible to increase the switching frequency. >