Abstract
Soft sensors have been widely used in industrial plants to estimate process variables that are difficult to measure online. Soft sensor models predicting an objective variable should be constructed with only important explanatory variables in terms of predictive ability, better interpretation of models and lower measurement costs. Besides, some process variables can affect an objective variable with time-delays. In some studies, soft sensor models are constructed using process dynamics and in others, the selection of process variables is used to increase the predictive accuracy. No one, however, has yet realized the optimization of both considerations in process dynamics and process variable selection. We therefore have proposed the methods for selecting important process variables and optimal time-delays of each variable simultaneously, by modifying the wavelength selection methods in spectrum analysis such as stacked partial least squares (PLS), searching combination moving window PLS, and genetic algorithm-based wavelength selection. The proposed methods can select time-regions of process variables as a unit by using process data that includes process variables that are delayed for a duration ranging from 0 through some time. The case study with real industrial data confirmed that predictive, easy-to-interpret, and appropriate models were constructed using the proposed methods.
