Dynamic Surrogate Modeling of an Air Separation Plant for Real-Time Dynamic Market-Driven Optimization Under Uncertainty

Abstract

Cryogenic air separation units (ASUs) are the primary industrial technology for producing high purity oxygen, nitrogen, and argon gases at commercial scale. Cryogenic ASUs are large consumers of electricity, making them ideal candidates for market-driven operation research in today’s volatile and uncertain process manufacturing environment landscape. This thesis focuses on strategies for real-time market-driven operation of ASUs and the development of accurate, computationally efficient reduced-order ASU models.

A dynamic surrogate modeling framework is presented which combines latent variable methods and neural networks for accurate and computationally efficient multistep-ahead simulation and dynamic optimization of an air separation plant. The high-dimensional full-order model (FOM) consisting of ≈ 3800 states states is projected onto a 10-dimensional latent subspace using principal component analysis (PCA). Following order reduction, a ReLU-activated multilayer perceptron (MLP) neural network is trained to compute step-ahead predictions of the latent states in addition to the squared prediction error (SPE) statistic of the step-ahead prediction. The latent variable-based surrogate model (LV-SM) is validated through multistep-ahead simulation case studies, demonstrating high prediction accuracy for restoration of not only the states directly relevant to optimization, but also the entire original state-space. The LV-SM’s performance in dynamic optimization is then validated using a deterministic market-driven dynamic optimization problem, where it achieves a solution nearly identical to the FOM with nearly three orders of magnitude reduction in computation time using a two-tiered optimization approach. Moreover, a trust region is enforced during optimization based on the SPE statistic to provide additional confidence in solution validity.

The LV-SM is then implemented within a two-stage stochastic rolling-horizon scheduling strategy for the real-time market-driven operation of the ASU. The strategy incorporates explicit consideration of market forecast uncertainty in its formulation and performs periodic rescheduling in response to newly available market information. The economic performance of this approach is compared to its deterministic counterpart which does not optimize under uncertainty. Under the parameters of the case studies considered in this work, results demonstrate that the deterministic rolling-horizon strategy is sufficient for handling uncertainties in future market parameters.

The results of this work highlight the potential of the LV-SM as a computationally efficient substitute for high-dimensional and complex first principles-based industrial process models, particularly for use in real-time operations applications. With the rolling-horizon framework in place, further extensions may be considered to identify areas where the two-stage stochastic rolling-horizon scheduling strategy offers improved economic performance and operational robustness compared to deterministic alternatives.