(52d) Data-Assisted Modeling and Optimization for Process Systems Engineering

Coupling equation-based first-principle modeling with information that comes in the form of input-output data for enhanced decision-making, is gaining significant interest in process systems engineering [1]. Most data-driven optimization techniques consider the system under study entirely as a “black-box”, however, in order to take advantage of years of obtained physical process knowledge, this work proposes a hybrid modeling and optimization approach [2-3]. This talk will demonstrate the performance of a suite of flexible modeling and optimization tools that are able to incorporate both physics-based and machine-learning components, to enable reliable data-assisted decision making. Specifically, adaptive and tractable surrogate functions will be developed to represent input-output data, and these will be merged with mechanistic equations in the form of constraints or discrepancy modeling [4]. Optimization is performed via a custom-based branch-and-bound framework that incorporates sampling, machine learning and bounding of the uncertainty of the approximations. Finally, the performance of the proposed methods is demonstrated through a set of process design case studies that rely on multifidelity models and data.

References:

1. Venkatasubramanian, V. (2019), AIChE J., 65: 466-478
2. Boukouvala, F., Hasan, MMF, Floudas, C.A. (2017). JOGO, 67(1-2):3-42
3. Kim, S.H. and Boukouvala, F. (2019), Optimization Letters, 10.1007/s11590-019-01428-7
4. Kennedy, M. C. and O'Hagan, A. (2001), Journal of the Royal Statistical Society, 63: 425-464