(474h) Global Deterministic Surrogate-Based Process Design

Surrogate-based process optimization is increasingly popular and is believed to gain more importance as digitalization of industry further progresses because more data becomes available. In the previous literature, there have been many efforts to combine surrogate modeling with (global) optimization. For instance, Gaussian processes (GPs) have been used in the field of Bayesian optimization for optimization of expensive-to-evaluate black-box functions (e.g., [1,2]). In adaptive approaches, GPs emulate black-box objectives, e.g., measurements of a chemical experiment [3], in every iteration to assist a following query point selection. ALAMO is another adaptive sampling approach [4-6] that aims the development of simple surrogate models in the light of small data sets. Artificial neural networks (ANNs) are one of the most commonly used machine-learning tools and are well-established in both academia and industry [7]. In contrast to standard GPs, ANNs can have multiple outputs and can handle large training sets (i.e., GP training requires N by N matrix inversions, where N is the number of training points).

Various works on ANN-based process optimization are available (e.g. [7-10]). In these works, ANNs learn unit operations or sub-process from simulations or experimental data. Then, the ANNs are aggregated to entire (hybrid) process models, which are finally optimized. In the previous literature, ANN-based process optimization has mostly been limited to local or stochastic global solution approaches (e.g., genetic algorithm). These methods can fail to identify global optima and cannot give a guarantee for global optimality. Recent advances in global deterministic optimization with ANNs embedded allow to overcome this limitation [11].

The emerging research field of global deterministic ANN-based process optimization develops new potentials but raises research questions, which are discussed using an organic Rankine cycle process optimization as an illustrative case study. Herein, the complex Helmholtz thermodynamics of the ethanol working fluid is learned via ANNs to eliminate implicit functions and auxiliary variables from the optimization problem. The results show that the solution time of the problem is reduced significantly using the proposed approach. Further, the presentation discusses the influence of the ANN size on its prediction accuracy and the efficiency of global optimization, i.e., tightness of its relaxations.

[1] Bradford, E., Schweidtmann, A.M., Lapkin, A.A.: Efficient multiobjective optimization employing gaussian processes, spectral sampling and a genetic algorithm. Global Optimization Journal (2018).

[2] Forrester, A.I., Keane, A.J.: Recent advances in surrogate-based optimization. Progress in Aerospace Sciences 45(1-3), 50-79 (2009).

[3] Schweidtmann, A.M., Holmes, N., Bradford, E., Bourne, R., Lapkin, A.: Multi-objective self-optimization of continuous flow reactors. In preparation (2018).

[4] Wilson, Z.T., Sahinidis, N.V.: The ALAMO approach to machine learning. Computers & Chemical Engineering 106, 785-795 (2017).

[5] Cozad, A., Sahinidis, N.V., Miller, D.C.: A combined first-principles and data-driven approach to model building. Computers & Chemical Engineering 73, 116-127 (2015).

[6] Cozad, A., Sahinidis, N.V., Miller, D.C.: Learning surrogate models for simulation-based optimization. AIChE Journal 60(6), 2211-2227 (2014).

[7] Meireles, M., Almeida, P., Simoes, M.G.: A comprehensive review for industrial applicability of artificial neural networks. IEEE Transactions on Industrial Electronics 50(3), 585-601 (2003).

[8] Nascimento, C.A.O., Giudici, R., Guardani, R.: Neural network based approach for optimization of industrial chemical processes. Computers & Chemical Engineering 24(9-10), 2303-2314 (2000).

[9] Henao, C.A., Maravelias, C.T.: Surrogate-based superstructure optimization framework. AIChE Journal 57(5), 1216-1232 (2011).

[10] Smith, J.D., Neto, A.A., Cremaschi, S., Crunkleton, D.W.: CFD-based optimization of a flooded bed algae bioreactor. Industrial & Engineering Chemistry Research 52(22), 7181-7188 (2013).

[11] Schweidtmann, A.M., Mitsos, A.: Global Deterministic Optimization with ANNs Embedded. Preprint available at arXiv:1801.07114 (2018).