A new hybrid approach in the estimation of end-states of a tubular plug-flow reactor by Kalman filter

The state of a chemical plug-flow reactor is estimated at the reactor outlet by designing a hybrid extended Kalman filter algorithm. Temperature, activity and six (6) compositions are considered as typical reactor states and measurements are obtained by the addition of a random noise to the deterministic values of the dynamic reactor model. The system is non-linear in the eight states and distributed in time and axial position. A hybrid approach is introduced parallel to the conventional architecture of Kalman filter, which utilizes a so called end-state derivation block (ESDB) in order to cope with Kalman filtering of a multi-dimensional model. The real-world observations of the system are combined with the values of partial derivatives of the states, obtained from deterministic simulation results at system boundary, to handle the problem of special-variation of the states. So there is no need to update Kalman filter with respect to local variation of the states, and therefore this technique significantly decreases the amount of computer burden encountered with Kalman filtering of a distributed system. However it is limited to application of Kalman filter for estimation of the end-states.