Accelerating Kinetic Parameter Identification by Extracting Information from Transient Data: A Hydroprocessing Study Case

Ngoc-Yen-Phuong Cao,Joris W Thybaut,Denis Guillaume,Benoit Celse,Isabelle Guibard

doi:10.3390/catal10040361

Ngoc-Yen-Phuong Cao, Joris W Thybaut + Show 3 more

Open Access

https://doi.org/10.3390/catal10040361

Copy DOI

Abstract

Hydroprocessing reactions require several days to reach steady-state, leading to long experimentation times for collecting sufficient data for kinetic modeling purposes. The information contained in the transient data during the evolution toward the steady-state is, at present, not used for kinetic modeling since the stabilization behavior is not well understood. The present work aims at accelerating kinetic model construction by employing these transient data, provided that the stabilization can be adequately accounted for. A comparison between the model obtained against the steady-state data and the one after accounting for the transient information was carried out. It was demonstrated that by accounting for the stabilization, combined with an experimental design algorithm, a more robust and faster manner was obtained to identify kinetic parameters, which saves time and cost. An application was presented in hydrodenitrogenation, but the proposed methodology can be extended to any hydroprocessing reaction.

Highlights

A good prediction of overall process performance based on the input conditions is of great interest in many domains, especially in the chemical industry
A methodology exploiting transient data to determine the parameters in a kinetic model for hydrodenitrogenation has been devised by employing a model considering the stabilization as a function of time on stream
Stabilization is considered to follow a first-order behavior characterized by parameter τ, accounting for the transient phenomena

Summary

Introduction

A good prediction of overall process performance based on the input conditions is of great interest in many domains, especially in the chemical industry. Predictive models are usually trained on the experimental data, which can interpolate well in a similar domain with the experimental data. Extrapolation toward a new domain can be less reliable. Model recalibration is required, which leads to a demand for new training data. Collecting data is usually expensive and time-consuming, acquired data should be exploited in their entirety. For petroleum related conversion processes (e.g., hydrotreating and hydrocracking), it appears that available data are only partially used during modeling, see below. The challenge is, to exploit the non-used data to uncover the underlying information and determine the model more rapidly and/or precisely

Methods

Results

Conclusion