Identification by Recurrent Neural Networks applied to a Pressure Swing Adsorption Process for Ethanol Purification

• Published in: 2022 Signal Processing: Algorithms, Architectures, Arrangements, and Applications (SPA) pp. 128 - 134
• Main Authors: Renteria-Vargas, Erasmo Misael, Zuniga Aguilar, Carlos Jesus, Rumbo Morales, Jesse Yoe, De-La-Torre, Miguel, Cervantes, Jose Antonio, Lomeli Huerta, Jose Roberto, Torres, Gerardo Ortiz, Vazquez, Felipe De J. Sorcia, Sanchez, Rene Osorio
• Format: Conference Proceeding
• Language: English
• Published: Division of Signal Processing and Electronic Systems, Poznan University of Technology (DSPES PUT) 21.09.2022
• Subjects: Adsorption; Partial differential equations; Process control; Purification; Recurrent neural networks; Signal processing algorithms; System dynamics
• ISSN: 2326-0262; 2326-0319
• DOI: 10.23919/SPA53010.2022.9927850

The pressure swing adsorption (PSA) process is employed for efficient separation of gas mixtures under pressure, and achieve different levels of purity. When used to produce biofuel, PSA parameters should be optimised to meet international standards of purity. PSA modeling relies on highly nonlinear partial differential equations (PDEs). The objective of this work is to propose a model that captures the dynamics of PSA represented by PDEs, using recurrent neural networks (RNN). The proposed RNN is composed of three layers to process the inputs and previous outputs of the real (or simulated) system. Results demonstrate that the proposed RNN presents a more stable approximation, reaching around 97% of purity, with a fit to the simulated purity curve of 84.51%, compared to the 75.84% obtained with Hammerstein-Wienner approach.