Recurrent neural networks basic concepts and applications for process monitoring and modelling

ANNs are computational systems with an architecture and operation inspired from our knowledge of biological neural cells (neurons) in the brain. They can be described either as mathematical and computational models for static and dynamic (time-varying) non-linear function approximation, data classification, clustering and non-parametric regression or as simulations of the behavior of collections of model biological neurons. These are not simulations of real neurons in the sense that they do not model the biology, chemistry, or physics of a real neuron. They do, however, model several aspects of the information combining and pattern recognition behavior of real neurons in a simple yet meaningful way. Neural modelling has shown incredible capability for emulation, analysis, prediction, and association. ANNs are able to solve difficult problems in a way that resembles human intelligence [34]. What is unique about neural networks is their ability to learn by examples. ANNs can and should, however, be retrained on or off line whenever new information becomes available.

There exist many different ANN structures. Among them there are two main categories in use for control applications: feedforward neural network (FNN) and recurrent (feed back) neural network (RNN) [35,36]. FNN consists of only feed-forward paths, its node characteristics involve static nonlinear functions. An example of a FNN is shown in Figure 1.3. In contrast to FNNs,

layer layer layer

Fig. 1.3. Topological structure of a FNN.

layer layer layer

Fig. 1.3. Topological structure of a FNN.

the topology in RNNs consists of both feed-forward and feedback connections, its node characteristics involve nonlinear dynamic functions and can be used to capture nonlinear dynamic characteristics of non-stationary systems [7,37]. An example of RNN is illustrated in Figure 1.4.

Input Hidden Output layer layer layer

Fig. 1.4. Topological structure of a RNN.

Input Hidden Output layer layer layer

Fig. 1.4. Topological structure of a RNN.

0 0

Post a comment