Abstract
•A comparative study of several moving horizon estimators using a batch cultivation process.•Moving horizon schemes outperform the constrained Extended Kalman Filter in dealing with challenges typically encountered in the operation of bioprocesses.•Smoothing arrival cost update approaches are superior in recovering from inaccurate initial conditions and covariances.•Multi-rate moving horizon schemes are the preferred choice for handling delayed multi-rate measurements compared to the Kalman filter.•The online tuning of the Q matrix improves the estimation performance and offers a reliable tuning methodology.
Advanced control and monitoring of bioprocesses are dependent on accurate state and parameter information. At the same time, bioprocesses are well known for their time-varying behavior and difficulty of obtaining online measurements of the important process states. The selection and the tuning of the estimation algorithms is therefore crucial to the design of reliable monitoring tools. In this work we discuss the design of several Moving Horizon Estimation schemes for a class of bioprocesses. We compare the algorithms in terms of the arrival cost computation and address the implementation of a multi-rate measurement structure. The tuning of the estimators and further details are illustrated using a cell culture case study, where we show that not all the estimators under investigation can cope equally well with process uncertainty and multi-rate measurements. As an outcome of the comparative study, we provide a set of guidelines for selecting an appropriate estimator for bioprocess applications.