Neural-network contributions in biotechnology

On incorporating physical building parameters into neural net models of dynamic thermal behaviour

“…Fully-connected feed forward neural networks are well suited to modelling complex or only partially understood systems for several reasons -for example: due to bounded transfer functions, they can mimic highly non-linear sequences (Montague and Morris, 1994) and they are capable of adopting a wide variety of mathematical structures due to their ability to emphasise or ignore inputs. The ability to predict time-series data in the absence of a mechanistic understanding of the system is clearly useful for practical control purposes.…”

Discrete PID Controller Tuning Using Piecewise-Linear Neural Network

“…They have given possibilities to provide additional features like automatic tuning or continuous adaptation -and continuous adaptation of PID controller via neural model of controlled system (which is considered to be significantly nonlinear) is the aim of this contribution. Artificial Neural Networks have traditionally enjoyed considerable attention in process control applications, especially for their universal approximation abilities (Montague et al, 1994), (Dwarapudi, et al, 2007). In next sections, there is to be explained how to use artificial neural networks with piecewise-linear activation functions in hidden layer in controller design.…”

Identification of semi-parametric hybrid process models

“…Within this group of models, artificial neural networks (e.g. Psichogios, Ungar, 1992;Thompson, Kramer, 1994;Montague, Morris, 1994) have most often been considered. More generally, a black-box model will belong to the family of non-parametric models (Eubank, 1988;Hastie, Tibshirani, 1990;Green, Silverman, 1994).…”