Artificial Intelligence Application for Crude Distillation Unit: An Overview

conference paper

en

Artificial intelligence (AI) with its efficiency for complex systems is growing in popularity in many engineering fields. The ability of an AI method to be successfully applied is highly dependent on the previous research, which makes knowledge sharing within and across fields extremely valuable. This work focuses on crude distillations units (CDU), whose energy optimization has been a tremendous challenge because of its complexity. The presented overview shows that soft sensors are the most common application of artificial intelligence for a CDU, although a number of recent publications focus on optimization problems. The approaches for optimization are very diverse, which makes them hardly applicable in the current engineering practice. This work provides a guideline for selecting the right method, but also addresses the fact that different methods excel at different problems and with different data set sizes. For neural networks (NN), this further depends on their architecture and hyperparameter adjustment. This urges future research, whose goal could be a workflow that would automatically adapt methods and perform parameter tuning with minimum user input.

Artificial intelligence (AI) with its efficiency for complex systems is growing in popularity in many engineering fields. The ability of an AI method to be successfully applied is highly dependent on the previous research, which makes knowledge sharing within and across fields extremely valuable. This work focuses on crude distillations units (CDU), whose energy optimization has been a tremendous challenge because of its complexity. The presented overview shows that soft sensors are the most common application of artificial intelligence for a CDU, although a number of recent publications focus on optimization problems. The approaches for optimization are very diverse, which makes them hardly applicable in the current engineering practice. This work provides a guideline for selecting the right method, but also addresses the fact that different methods excel at different problems and with different data set sizes. For neural networks (NN), this further depends on their architecture and hyperparameter adjustment. This urges future research, whose goal could be a workflow that would automatically adapt methods and perform parameter tuning with minimum user input.

Crude Distillation Unit (CDU), Artificial Intelligence (AI), Machine Learning (ML), Neural Network (NN), Sustainability

28.02.2022

Springer

Cham, Switzerland

978-3-030-96592-1

1868-4238

Artificial Intelligence for Knowledge Management, Energy, and Sustainability: 9th IFIP WG 12.6 and 1st IFIP WG 12.11 International Workshop, AI4KMES 2021, Held at IJCAI 2021, Montreal, QC, Canada, August 19–20, 2021, Revised Selected Papers

637

1

156–168

13