Data Selection and Regression Method and Its Application to Softsensing Using Multirate Industrial Data

Abstract

The estimation of difficult and infrequently measured variables (composition, melt flow index, viscosity, etc.) using easily and frequently measured variables (temperatures, flow rates, pressure, etc.) is of industrial interest. From such multirate data (data available at different sampling rates), a mathematical model that relates the frequently measured variables to the infrequently measured variable is developed—this model is often referred to as the soft sensor. This work considers the development of soft sensors to predict the concentration of a hydrocarbon species R at the exit of a two-reactor train. Specifically, we examine the development of soft sensors (one for each reactor) using optimal window size and demonstrate the efficacy of multiple model based prediction.