engleski

An Efficient Automated Learning of Qualitative Process Models

Process control system design relies in the main part on the suitable modeling and simulation of the underpinning operation of the physical system. The paper presupposes that the qualitative dependency among process variables is the fundamental model that must be generated from quantitative across time observations. Hence, a novel approach to qualitative interpretation of sensory data taken over time from the industrial plant is presented. Contrary to the most current techniques that derive a single qualitative differential equation, the described approach extracts diverse qualitative expressions (correlative knowledge) from the set of observed process variables from many distinctive and interesting time intervals. The technique is based on mapping quantitative time series data to binary coded qualitative difference vectors of process variables that significantly accelerates the modeling process. An experimental evaluation of the technique is shown.

Process modeling; Qualitative modeling; Process identification

nije evidentirano