TABLE OF CONTENTS by Paul G. Luebbers [*]
Artificial neural networks (ANNs) are networks composed of many nonlinear computing nodes. These computing nodes were originally inspired by biological neurons. Since these computing nodes are a rough, approximate model of biological neurons, the networks are often called ANNs to distinguish them explicitly from true neural networks(such asthe human brain for example). Since these ANNs should be easy to distinguish from their biological counterparts, they are referred to simply as neural networks (NNs).
NNs, for our purposes, are nonlinear mathematical approximations. These NNs have various characteristics that make them very convenient for modeling functions and processes:
1. NNs can easily be used to develop linear and nonlinear empirical models. It has been shown that NNs can be used to model any nonlinear function to any degree of accuracy, providing enough computing nodes are used. These models, or relationships, may be static or time varying.
2. Input/output data can be used to construct these empirical models. Often the underlying principles that govern a process or plant are not known well enough to construct a first-principle model; however, input/output data sets that describe the process can almost always be collected.
3. The resulting models can often be used to generalize and predict outputs for input sets that have not been used in the construction or training of the NN model.
The human brain is arranged in a massively parallel network of neurons, which permit...
