Efficient Tuning of an Isotope Separation Online System Through Safe Bayesian Optimization with Simulation-Informed Gaussian Process for the Constraints

Santiago Ramos Garces (), Ivan De Boi, João Pedro Ramos, Marc Dierckx, Lucia Popescu and Stijn Derammelaere
Additional contact information
Santiago Ramos Garces: Belgian Nuclear Research Centre (SCK CEN), 2400 Mol, Belgium
Ivan De Boi: Industrial Vision Lab (InViLab), Department of Electromechanics, University of Antwerp, 2020 Antwerp, Belgium
João Pedro Ramos: Belgian Nuclear Research Centre (SCK CEN), 2400 Mol, Belgium
Marc Dierckx: Belgian Nuclear Research Centre (SCK CEN), 2400 Mol, Belgium
Lucia Popescu: Belgian Nuclear Research Centre (SCK CEN), 2400 Mol, Belgium
Stijn Derammelaere: Co-Design for Cyber-Physical Systems (CoSysLab), Department of Electromechanics, University of Antwerp, 2020 Antwerp, Belgium

Mathematics, 2024, vol. 12, issue 23, 1-28

Abstract: Optimizing process outcomes by tuning parameters through an automated system is common in industry. Ideally, this optimization is performed as efficiently as possible, using the minimum number of steps to achieve an optimal configuration. However, care must often be taken to ensure that, in pursuing the optimal solution, the process does not enter an “unsafe” state (for the process itself or its surroundings). Safe Bayesian optimization is a viable method in such contexts, as it guarantees constraint fulfillment during the optimization process, ensuring the system remains safe. This method assumes the constraints are real-valued and continuous functions. However, in some cases, the constraints are binary (true/false) or classification-based (safe/unsafe), limiting the direct application of safe Bayesian optimization. Therefore, a slight modification of safe Bayesian optimization allows for applying the method using a probabilistic classifier for learning classification constraints. However, violation of constraints may occur during the optimization process, as the theoretical guarantees of safe Bayesian optimization do not apply to discontinuous functions. This paper addresses this limitation by introducing an enhanced version of safe Bayesian optimization incorporating a simulation-informed Gaussian process (GP) for handling classification constraints. The simulation-informed GP transforms the classification constraint into a piece-wise function, enabling the application of safe Bayesian optimization. We applied this approach to optimize the parameters of a computational model for the isotope separator online (ISOL) at the MYRRHA facility (Multipurpose Hybrid Research Reactor for High-Tech Applications). The results revealed a significant reduction in constraint violations—approximately 80 % —compared to safe Bayesian optimization methods that directly learn the classification constraints using Laplace approximation and expectation propagation . The sensitivity to the accuracy of the simulation model was analyzed to determine the extent to which it is advantageous to use the proposed method. These findings suggest that incorporating available information into the optimization process is valuable for reducing the number of unsafe outcomes in constrained optimization scenarios.

Keywords: Gaussian process; Bayesian optimization; simulation-informed Gaussian process; safe Bayesian optimization; classification constraints (search for similar items in EconPapers)
JEL-codes: C (search for similar items in EconPapers)
Date: 2024
References: View complete reference list from CitEc
Citations:

Downloads: (external link)
https://www.mdpi.com/2227-7390/12/23/3696/pdf (application/pdf)
https://www.mdpi.com/2227-7390/12/23/3696/ (text/html)

Related works:
This item may be available elsewhere in EconPapers: Search for items with the same title.

Export reference: BibTeX RIS (EndNote, ProCite, RefMan) HTML/Text

Persistent link: https://EconPapers.repec.org/RePEc:gam:jmathe:v:12:y:2024:i:23:p:3696-:d:1529255

Access Statistics for this article

Mathematics is currently edited by Ms. Emma He

More articles in Mathematics from MDPI
Bibliographic data for series maintained by MDPI Indexing Manager ().