State and covariance estimation of a semi-batch reactor for bioprocess applications

Abstract

In this work, extended Kalman filter (EKF) and moving horizon estimation (MHE)-based approaches are introduced and applied to a Galacto-oligosaccharides (GOS) bioprocess system considering a recently developed enzymatic model. Using this model, plant data is simulated by varying the kinetic parameters and applying white noise to the resulting output. In terms of EKF applications, both the canonical and parametric (Dual EKF) formulations have their covariances specified using a modified direct optimization (DO) algorithm to reduce estimation error. This work also outlines the development and application of a novel Parameter-based Moving Horizon Estimation (P-MHE) approach and directly compares it to traditional MHE formulations. The proposed P-MHE method reduces both the estimation error and computational time of traditional MHE approaches. When compared to EKF-based approaches, P-MHE can compete in terms of estimation error while exhibiting excellent robustness characteristics such as guaranteed feasibility. Despite the increased computational time of P-MHE when compared to the EKF formulations, state estimates can be obtained in under 3 seconds once a measurement arrives, allowing this algorithm to be applied to real-time process monitoring and other process systems.