SPEAKER:   Derrick Rollins, Iowa State University, Departments of Statistics
                                             and Chemical Engineering

TITLE:   Beyond ARIMA, Beyond ARMAX, and Addressing Serially Correlated
                                            Errors in Dynamic Processes
 

ABSTRACT:

ARIMA modeling of dynamic systems can be practical when the
objective is prediction and not process understanding between input
(explanatory) and output (response) variables. However, when this
understanding is required, such as in model based control, ARIMA models
will not be useful. NARMAX modeling, or transfer function modeling,
expresses input/output relationships, and is a popular, if not exclusive,
method among statisticians. Regardless, this methodology cannot capture
nonlinear dynamic behavior. This work seeks to broaden ones modeling tool
box in treating complex non-linear dynamic systems with an approach called
block-orientedmodeling that is able to treat a wide range of dynamic
behavior. We will demonstrate the ability of one of its structures, called
Wiener,to accurately model in the presence of general ARMA serially
correlated errors for simulated and physical processes. We will also show
that, for physical systems in the presence of serially correlated errors,
the model developed under the assumption of no serial correlation is better
than the model developed under the opposite assumption (but meeting the
requirements of i.i.d. errors for parameter estimation) when prediction is
based solely on the inputs.

COFFEE: 3:45 p.m., 104 Snedecor Hall