GP-based Soft Sensor Modeling

Soft sensors are wildly employed as an effective alternative for physical sensors in industrial processes. And the key problem is to set the model of soft sensors. We propose a novel modeling approach using Gaussian processes(GP). GPs are probabilistic kernel machines. Theoretic analysis and simulation experiment show that GP-based soft sensor is moderately simple to implement and use without loss of performance compared with artificial neural networks (ANN) and support vector machines (SVM), which lays solid basis for advanced control system.