The real-time adjustment is responsible for two sources of changes during the blending batch execution. The first accounts for deviations in the blend stream properties, measured by the online analyser, from the target properties of the blended gasoline. The second one deals with fluctua - tions in the properties of the blended components streams, also measured by the online analysers. The Modcon.AI optimisation suite, together with the Beacon 3000 NIR analyser, can serve as a basis for the online blending sys- tem shown in Figure 2. As the search for the optimal stream ratios in real-time happens in a small region of the blend model parameter space around the blend recipe found during the blend plan- ning, the model can be simplified to facilitate a real-time optimisation problem solution if its precision can be assured for this small area of the state space. For this purpose, the blend hybrid model can be simplified using the reduced order modelling (ROM) approach. This can be a low-degree polynomial model or even a linear approximation. Conclusion We have shown that two factors are crucial in addressing the objective of octane giveaway reduction. The first is the online analysis of the inbound and outbound gasoline blending station streams. The second is blending model accuracy. In achieving the latter, a hybrid model combining the traditional analytic formulation of the dependency, such as the one by DuPont, can be complemented by a smart bias, implemented using the tools of machine learning.
References 1 Shahnovsky G, Cohen T, McMurray R, Advanced solutions for efficient crude blending, Digital Refining , Apr 2014, www.digitalrefining.com/ article/1000968/advanced-solutions-for-efficient-crude-blending 2 Shahnovsky G, Kigel A, Briskman G, Artificial intelligence for sustain - able development, Digital Refinin g, Jun 2022, www.digitalrefining.com/ article/1002778/artificial-intelligence-for-sustainable-development 3 Shahnovsky G, Kigel A, McMurray R, Intelligent blending, Hydrocarbon Engineering , Mar 2018, www.globalhydrogenreview. com/magazine/hydrocarbon-engineering/march-2018/ Gadi Briskman is AI Business Development Manager at Modcon Systems Ltd. He has several years of experience in computer vision, robotics, deep learning, and process optimisation. He holds an ME degree in systems engineering and a BSc in electrical engineering. Email: gadib@modcon-systems.com Ariel Kigel is R&D Department Manager at Modcon Systems Ltd. He manages market-oriented development and daily operations of online process analytical systems for conventional industry. He leads the definition and development of new applications into new industries, overseeing multi-national teams. He holds a PhD in physical chemistry and an MBA in business administration. Email: arielk@modcon-systems.com Tom Rosenwasser is a Software Engineer at Modcon Systems Ltd. He is an experienced data scientist focusing on deep learning and its applications to modelling and optimising technological processes, as well as medical applications. He holds a BSc in software engineering. Email: tomr@modcon-systems.com
The highest sample throughput. The fewest worries. Get rapid and continuous testing of up to 12 samples with the new AS Vision autosampler • Large sample port numbering for mistake-free setup • Innovative metal distribution block for lowest sample carryover • Fail-safe operation and corrosion-resistant construction • Fewer rinsing cycles for significantly reduced cross-contamination One-button testing delivers reliable vapor pressure with MINIVAP VP Vision and FTIR analysis with MINISCAN IR Vision of gasoline, diesel, jet fuel, solvents, VOCs, and more. Contact us now at www.grabner-instruments.com
GRABNER INSTRUMENTS ® Messtechnik GmbH A-1220 Vienna/Austria Dr. Otto-Neurath-Gasse 1 | +43/1/282 16 27-0 | www.grabner-instruments.com PETROLAB COMPANY (USA) 2001 North Indianwood Avenue, Broken Arrow, OK, 74012, USA | +1-918-459-7170 | www.petrolab.com ©2023 AMETEK, Inc. All rights reserved.
39
PTQ Q3 2024
www.digitalrefining.com
Powered by FlippingBook