Adaptation of Dynamic Data-Driven Models for Real-Time Applications: From Simulated to Real Batch Distillation Trajectories by Transfer Learning

In the absence of knowledge about challenging dynamic phenomena involved in batch distillation processes, e.g., complex flow regimes or appearing and vanishing phases, generation of accurate mechanistic models is limited. Real plant data containing this missing information is scarce, also limiting the use of data-driven models. To exploit the information contained in measurement data and a related but inaccurate first-principles model, transfer learning from simulated to real plant data is analyzed. For the use case of a batch distillation column, the adapted model provides more accurate predictions than a data-driven model trained exclusively on scarce real plant data or simulated data. Its enhanced convergence and lower computational cost make it suitable for optimization in real-time.     

Kernel-Based State-Space Kriging for Predictive Control

In this paper, we extend the state-space kriging (SSK) modeling technique presented in a previous work by the authors in order to consider non-autonomous systems. SSK is a data-driven method that computes predictions as linear combinations of past outputs. To model the nonlinear dynamics of the system, we propose the kernel-based state-space kriging (K-SSK), a new version of the SSK where kernel functions are used instead of resorting to considerations about the locality of the data. Also, a Kalman filter can be used to improve the predictions at each time step in the case of noisy measurements. A constrained tracking nonlinear model predictive control (NMPC) scheme using the black-box input-output model obtained by means of the K-SSK prediction method is proposed. Finally, a simulation example and a real experiment are provided in order to assess the performance of the proposed controller.