Differential neural network identifier for parameter determination of a mixed microbial culture model

Abstract

This paper presents an application of a class of Differential Neural Network (DNN) for the nonparametric identification of mixed microbial culture systems, and its use for the estimation of the interaction parameters between the involved species. The DNN identifier with a projectional operator was implemented. After the identification process; the structure of the reported Lotka-Volterra (LV) model was considered to estimate the unknown interaction parameters in a mixed culture. An optimization problem between the DNN approximation and the LV model was numerically solved to determine the interaction parameters. The approach was assessed considering a reported example of (LV) model for mixed microbial culture, as well as with a set of experimental data. The Automatic and Robotic Intestinal System ARIS was the experimental data source where the growth kinetics of the Lactobacilli and Bifidobacteria were assessed. The parameters estimation for the reported LV model proved average percent errors below 10%. The magnitude of parameters identified for the experimental mixed culture indicated a higher inhibitory competition of genus 1 (Bifidobacteria) exerted over genus 2 (Lactobacilli).