Visualization of process data by use of evolutionary computation

Successful exploratory analysis of process data often depends on the extrac tion and visualization of compact representative features of the data. This is usually accomplished via the construction of a model that relates the o riginal higher-dimensional set of variables to a set of lower-dimensional f eatures. In complex process systems, non-linear models such as neural netwo rks are often the only way of extracting compact (2D or 3D) variable sets. By making use of evolutionary Computation, a population of mapping function s are constructed, which provides a more natural approach to deal with the large number of local minima in the error surfaces associated with the opti mization of the mapping functions. In addition. relatively simple, explicit mapping functions can be extracted which may be more useful in application s such as the monitoring of multivariate process systems. (C) 2001 Elsevier Science Ltd. All rights reserved.