Nonlinear Filters for State Estimation of UV Flash Processes

We describe four algorithms for state estimation of stochastic differential-algebraic equations. We consider the extended Kalman filter, the unscented Kalman filter, the particle filter, and the ensemble Kalman filter. The differential-algebraic equations that we consider are in a semi-explicit index-l form. Models of dynamic UV flash processes are in such a form. The UV flash is relevant to rigorous models of many chemical phase equilibrium processes because it is a mathematical representation of the second law of thermodynamics. State estimation is relevant to model predictive control, model identification, fault detection, monitoring, and prediction. State estimation of UV flash processes is therefore important to safe and economical operation of processes such as flash separation, distillation, multiphase flow in pipelines, and oil production. We compare the accuracy and efficiency of the four filters using a numerical example that involves a UV flash separation process. Furthermore, we demonstrate that the filters can be used as soft sensors that estimate the vapor-liquid composition of the separation process based on temperature and pressure measurements.