Adaptive soft sensor for online prediction based on moving window Gaussian process regression

Very often important process variables cannot be measured online due to low sampling rate of sensors or because their values have to be obtained by laboratory analysis. In order to enable continuous process monitoring and efficient process control in such cases, soft sensors are usually used to estimate these difficult-to-measure process variables. Most industrial processes exhibit some kind of time-varying behavior. To ensure that soft sensor retains its precision, adaptation mechanism has to be implemented. In this paper adaptive soft sensor based on Gaussian Process Regression (GPR) is presented. To make GPR model training more efficient, algorithm for variable selection based on Mutual Information is proposed. Prediction capabilities of the proposed method are examined on real industrial data obtained at an oil distillation column.