Hybrid AI for Industrial Thermal Systems

Artificial Intelligence (AI) has disrupted several industries and for very good reasons. AI algorithms, combined with large computational power and scalability of the cloud are capable of finding non-obvious correlations in a large amount of data, predict future trends and help optimise various types of operations. They can be very efficient (i.e. can process a vast amount of data relatively quickly) and easy to apply when physics are not known. However, when deployed to the process industry, they have specific limitations including the fact that i) they rely on large quantities of data and, crucially, on their quality and ii) their extrapolation abilities are limited.

In this presentation we will discuss how deterministic models can be combined with AI ones into an Hybrid-AI approach and how this approach is currently being used in refineries to overcome the traditional limitations of AI methods. Specific case studies for various systems, including crude distillation units will be shown.