Data-Driven Model Based Computation and Analysis of Operability Sets Using High-Dimensional Continuation: A Plant-Wide Case Study

Abstract

A framework with a data-driven model based approach to compute and analyze the operability sets is presented. The computationally expensive steady-state simulations are replaced by machine learning models developed through statistical analysis of selected rigorous simulations. The entire solution manifold is computed using a high-dimensional continuation algorithm. The framework is applied to a plant-wide system. The steps of the framework are discussed in depth within the context of the plant-wide system. The operability sets computed by high-dimensional continuation are analyzed in detail. Operability sets of lower dimensional subsystems are computed to gain better insights. Input behavior at cost optimal conditions and the multiplicities in the subsystems are also explored.

Development of a Surrogate-model Based Energy Efficiency Estimator for a Multi-step Chemical Process

“…The ranges for the selected variables were determined based on simulations and earlier findings from the literature. It is well known that the ranges of inputs need to be reduced as the number of inputs is increased (Konge et al, 2020;Tran and Georgakis, 2018). Table 1 lists the selected variables and their feasible ranges applied to this study.…”

“…As indicated by Konge et al (2020), the steam cost in overall cost-efficiency of the TE process is relatively small. On the other hand, it might have more important effect to the energy efficiency.…”

“…The TE process has five main unit comprising an exothermic reactor, a condenser, a compressor, a separator, and a stripper. The operating cost of the TE process is related to the loss of product and reactants (in purge and product streams), steam utilization and the compressor work (Konge et al, 2020).…”

“…Sheta et al (2019) developed dynamic NNARX models with interpretable structures for four TE outputs. Recently, Konge et al (2020) proposed several machine learning based regression modeling techniques for building lower dimensional subsystems and performing process operability analysis to the TE process.…”