Department of Statistics Seminar
North Carolina State University

presents

Richard D. De Veaux

Williams College

"Hybrid Neural Networks for Environmental Process Control"

ABSTRACT

In many environmental processes there is a great deal of scientific and engineering knowledge available about the system. This domain knowledge may range from a simple energy balance taking the form of a constraint, to a complex first principles models containing unmeasurable reaction rates. The challenge is to incorporate this knowledge into the data analysis and eventual control of the process. A feedforward neural network provides a particularly convenient form for folding in such prior knowledge into the estimation, prediction and control of the system. The resulting model is known as a hybrid neural network model. We will show how the neural network is trained and compare its performance to more traditional statistical forecasting techniques.

Keywords: grey box models; process optimization; nonparametric regression

Friday, November 6, 1998

3:35 - 4:35 pm

206 Cox Hall

Refreshments will be served on the second floor of Dabney Hall (left of Room 222) at 3:00 pm.