Gaussian Process-Based Inferential Control System

Gaussian process are emerging as a relatively new soft sensor building technique with promising results. This paper proposes a Gaussian Process Inferential Control System (GP-ICS) to control infrequently-measured variables in industrial processes. This is achieved by utilising an adaptive Gaussian process-based soft sensor to provide accurate reliable and continuous online predictions of difficult to measure variables and feeding them back to a PI controller. The contributions of the paper are i) the introduction of Gaussian process-based soft sensors in building inferential control systems, ii) we empirically show that the Gaussian process based inferential controller outperforms the ANN-based controller.