Development of model-based iterative learning control of batch processes

In this contribution, a framework for modeling of batch and semi-batch processes using a set of either finite impulse response models or autoregressive models with exogenous inputs is extended to include initial conditions and measurement noise. For identification of the resulting high dimensional model sets, an identification scheme has been developed which uses regularization to constrain excessive degrees of freedom. The regularization constraints are based on desired model structure. Utilizing the data-driven model sets, iterative learning control may conveniently be set up in a model predictive framework. Implementing iterative learning control in such a framework offers in-batch disturbance rejection, which will improve from batch to batch. The above mentioned identification scheme and control algorithm are validated on simulated fed-batch yeast fermentations with promising results.