Optimized data exploration applied to the simulation of a chemical process

Abstract

In complex simulation environments, certain parameter space regions may result in non-convergent or unphysical outcomes. All parameters can therefore be labeled with a binary class describing whether or not they lead to valid results. In general, it can be very difficult to determine feasible parameter regions, especially without previous knowledge. We propose a novel algorithm to explore such an unknown parameter space and improve its feasibility classification in an iterative way. Moreover, we include an additional optimization target in the algorithm to guide the exploration towards regions of interest and to improve the classification therein. In our method we make use of well-established concepts from the field of machine learning like kernel support vector machines and kernel ridge regression. From a comparison with a Kriging-based exploration approach based on recently published results we can show the advantages of our algorithm in a binary feasibility classification scenario with a discrete feasibility constraint violation. In this context, we also propose an improvement of the Kriging-based exploration approach. We apply our novel method to a fully realistic, industrially relevant chemical process simulation to demonstrate its practical usability and find a comparably good approximation of the data space topology from relatively few data points.

Modeling, Simulation, and Optimization 4.0 for a Distillation Column

“…Methods which adaptively identify the convergence region could help to increase the number of convergent solutions which then can be used for data‐driven modeling. One example of such a method is given in , in which different targets in sampling could be combined: one is related to identifying the border between convergent and divergent simulation, one is more focused on exploration of the full range and another can be chosen to incorporate an optimization objective to sample where it is of interest. The specific preference of the user can be influenced by weighting these targets accordingly.…”

Some Aspects of Combining Data and Models in Process Engineering

“…Since the standard deviation directly quantifies the uncertainty about a prediction, it can be used to suggest a new simulation to improve the model. Such an approach has already been proposed, e.g., in , and is closely related to, e.g., adaptive sampling and Bayesian optimization .…”

Digitalization in Thermodynamics

“…Machine‐learning methods can help to quantify the feasible domain in the design space, thus, avoiding time‐consuming manual trial‐and‐error calculations. To this end, an adaptive design‐of‐experiments scheme has been developed to run the simulations, making the method computationally efficient .…”

“…Example for the application of machine learning to solving thermodynamic tasks in process simulation : determination of the feasible operating range of a partial evaporator. a) Sketch of the evaporator; b) initial set of design points to start the exploration of the feasible range; c) – e) successive exploration of the feasible domain; f) average distance between the true and the learned boundaries separating the feasible domain (i.e., two‐phase coexistence) from the infeasible (single‐phase) domains.…”

“…In more complex situations corresponding algorithms could be more difficult to devise and their execution could suffer from convergence problems. As an example for machine learning, an adaptive exploration scheme was applied in to learn this feasible range successively from simulations. The procedure commences with a small number of initial design points, as shown in Fig.…”