High-throughput screening of optimal process conditions using model predictive control.

Modern biotechnological laboratories are equipped with advanced parallel mini-bioreactor facilities that can perform sophisticated cultivation strategies (e.g., fed-batch or continuous) and generate significant amounts of measurement data. These systems require not only optimal experimental designs that find the best conditions in very large design spaces, but also algorithms that manage to operate a large number of different cultivations in parallel within a well-defined and tightly constrained operating regime. Existing advanced process control algorithms have to be tailored to tackle the specific issues of such facilities such as: a very complex biological system, constant changes in the metabolic activity and phenotypes, shifts of pH and/or temperature, and metabolic switches, to name a few. In this study we implement a model predictive control (MPC) framework to demonstrate: (1) the challenges in terms of mathematical model structure, state, and parameter estimation, and optimization under highly nonlinear and stiff dynamics in biological systems, (2) the adaptations required to enable the application of MPC in high throughput bioprocess development, and (3) the added value of MPC implementations when operating parallel mini-bioreactors aiming to maximize the biomass concentration while coping with hard constrains on the dissolved oxygen tension profile.

Efficient Representation and Approximation of Model Predictive Control Laws via Deep Learning.

We show that artificial neural networks with rectifier units as activation functions can exactly represent the piecewise affine function that results from the formulation of model predictive control (MPC) of linear time-invariant systems. The choice of deep neural networks is particularly interesting as they can represent exponentially many more affine regions compared to networks with only one hidden layer. We provide theoretical bounds on the minimum number of hidden layers and neurons per layer that a neural network should have to exactly represent a given MPC law. The proposed approach has a strong potential as an approximation method of predictive control laws, leading to a better approximation quality and significantly smaller memory requirements than previous approaches, as we illustrate via simulation examples. We also suggest different alternatives to correct or quantify the approximation error. Since the online evaluation of neural networks is extremely simple, the approximated controllers can be deployed on low-power embedded devices with small storage capacity, enabling the implementation of advanced decision-making strategies for complex cyber-physical systems with limited computing capabilities.