Soft-Sensors for Lipid Fermentation Variables Based on PSO Support Vector Machine (PSO-SVM)

Carlos Eduardo Robles-Rodriguez (1) , Carine Bideaux (1) , Gilles Roux (1) , Carole Molina-Jouve (1) , Cesar Arturo Aceves-Lara (1)

1 LISBP - Laboratoire d'Ingénierie des Systèmes Biologiques et des Procédés

On-line monitoring fermentation variables (e.g. biomass) can improve the performance of bio-processes, as well as the quality of the targeted products. However, on-line estimation could be a challenging task when an accurate model is not available. Over the existing methods for state estimation, the support vector machine (SVM) is an attractive method for its fast convergence and generalization of the approximated function. In this paper, a soft-sensor based on SVM and coupled to Particle Swarm Optimization (PSO) algorithm is presented and applied to estimate the concentrations of lipid fermentation variables: lipids, biomass, and citric acid. The soft-sensor was trained with one data set, and validated with an independent data set of fed-batch fermentations. The PSO-SVM was compared with the SVM algorithm. In general, the results show that the PSO-SVM is an efficient alternative for monitoring fermentations.

Mots clés: support vectors machine; soft-sensor; PSO; lipid fermentation; biomass

Mots clés: fermentation; biomasse

Domaines: Sciences du Vivant [q-bio]

Carlos Eduardo Robles-Rodriguez, Carine Bideaux, Gilles Roux, Carole Molina-Jouve, Cesar Arturo Aceves-Lara. Soft-Sensors for Lipid Fermentation Variables Based on PSO Support Vector Machine (PSO-SVM). Distributed Computing and Artificial Intelligence, 13th International Conference, Chapter 19, Editions Springer, 2016, Advances in Intelligent Systems and Computing, 978-3-319-40161-4. ⟨10.1007/978-3-319-40162-1_19⟩. ⟨hal-01601403⟩