Chapter 11 - Neural Network and Applications

Abstract

This chapter describes the basic principles of neural networks and some of their applications in the power electronics and motor drive area. Neural networks (NNWs) are a very important subject in artificial intelligence (AI), and the technology has advanced tremendously in recent years with the expansion of the applications into various areas. The idea of NNWs comes from the biological neuron and biological neural network; in practice, however, a NNW's performance may be far inferior. NNWs help in solving many problems efficiently that are difficult to solve by the traditional methods. A NNW has linear transfer characteristics with a linear activation or transfer function (TF), whereas a non-linear TF can provide NNWs with a non-linear mapping property. Most NNW-based projects require non-linear TFs. Many NNW topologies have been proposed in the literature that can be classified as feed forward and feedback or recurrent NNWs (RNN). Feedforward networks have static non-linear mapping property, whereas recurrent networks have dynamical or temporal mapping property. Therefore, feedforward NNWs are used for pattern recognition type applications, whereas RNNs are utilized for emulating dynamical systems.