Hybrid plant model of physical and statistical model with robust updating method

Abstract

This paper introduces an architecture of hybrid model of physical and statistical model with adaptive updating method based on Kalman Filter and just-in-time (JIT) modeling. When the real time reference output is accessible, the powerful Kalman filter is recommended to update the statistical linear model. In addition, an effective initial value calculating method is proposed to perfect the powerful Kalman filter updating mechanics. When the reference output is not available, Kalman filter can not work anymore, just-in-time (JIT) updating method is brought in to update statistical models. The validity of this architecture is clarified through a case study of Methane-Steam Reforming process.