Strategy of Data Selection for Adaptive Automation

Abstract

The problems of selection of logged process variables by a control computer with limited memory for continuous on-line process automation in the presence of erroneous, erratic and irrelevant data sources are considered from a strategic viewpoint. Markov and simple learning strategies are presented that allow a computer to learn about the current behavior of a process with respect to variable relevancy, resulting in a considerable increase in data selection efficiency over a random selection strategy. The case of single variable selection using a Markov strategy is solved analytically and computer simulation is used to estimate the performance of a simple learning strategy.