(598c) Surrogate-Based Optimization Framework in Process Systems Engineering

Simulations are widely used in engineering and scientific disciplines to describe realistic and complex systems and interactions. Computational advances over the last decades have enabled the use of simulations for process analysis and optimization leading to best economic, environmental friendly and safe solutions. However, there are multiple challenges in the objective of optimizing computationally expensive computer simulations.

In the pursuit of more realistic descriptions of reality, simulations may become computationally expensive leading to interesting decision-making problems where information is limited. Surrogate modeling allows us to build lower fidelity models that approximate the original simulation utilizing limited data. Surrogate modeling has been an active area of research over past two decades and number of techniques exist in the literature [1]. Furthermore, surrogate models can be more efficiently used if one makes use of sequential sampling to gather information.

This work discusses a framework for surrogate-based optimization. First, surrogate models are built from an initial data set obtained using design of experiments techniques. New samples are then generated using adaptive sampling schemes that balance between finding a better optimal solution and reducing the uncertainty in the surrogate model. Surrogate models are iteratively updated using these samples until desired accuracy is achieved or a stopping criteria is met. Specifically, this work focuses on Kriging and radial basis function surrogate models. Adaptive sampling strategies used in this work make use of an infill criteria based on expected improvement function. Finally, this framework is applied to solve problems in three different application domains: integrated scheduling and control, pharmaceutical feasibility analysis, and supply chain optimization. For integrated scheduling and control, adaptive sampling is used to minimize uncertainty in surrogate models that represent the control-level simulation [2]. For feasibility analysis problems, a modified version of expected improvement function is used to understand feasible region boundary [3]. In supply chain simulation optimization, the problem of finding optimal warehouse capacities that minimize the total cost of the supply chain where the supply chain network consists of more than one enterprise is considered[4]. With the help of Kriging surrogate model, adaptive sampling is used to guide the global search in the proposed optimization framework.

References

[1] A. Bhosekar and M. Ierapetritou, “Advances in surrogate based modeling, feasibility analysis, and optimization: A review,” Comput. Chem. Eng., vol. 108, pp. 250–267, 2018.

[2] L. S. Dias and M. G. Ierapetritou, “Integration of scheduling and control under uncertainties: Review and challenges,” Chem. Eng. Res. Des., vol. 116, pp. 98–113, 2016.

[3] Z. Wang, M. S. Escotet-Espinoza, R. Singh, and M. Ierapetritou, “Surrogate-based Optimization for Pharmaceutical Manufacturing Processes,” in 27th European Symposium on Computer Aided Process Engineering, vol. 40, A. Espuña, M. Graells, and L. Puigjaner, Eds. Elsevier, 2017, pp. 2797–2802.

[4] N. Sahay and M. Ierapetritou, “Multienterprise supply chain: Simulation and optimization,” AIChE J., vol. 62, no. 9, pp. 3392–3403, 2016.