Hybrid Gaussian Process Models for continuous time series in bolus fed-batch cultures

Hybrid modeling, meaning the integration of data-driven and knowledge-based methods, is quickly gaining popularity in many research fields, including bioprocess engineering development. Recently, part hybrid methods have been largely extended with machine learning algorithms (e.g., artificial neural network, support vector regression), while mechanistic typically based on differential equations to describe dynamics process its current state. In this work we present an alternative model formulation that merges advantages Gaussian Process State Space Models numerical approximation equation systems through full discretization . The use complex bioprocesses batch, fed-batch, continuous has reported several applications. Nevertheless, handling states system, known a time-dependent evolution governed by implicit dynamics, proven be major challenge. Discretization matching sampling steps source complications, as are: 1) not being able handle multi-rate date sets, 2) step-size derivative defined frequency, 3) high sensitivity addition errors. We coupling polynomial regression representation right-hand side ordinary system demonstrate typical fed-batch cultivation for monoclonal antibody production.