A simulation study of dynamic neural filtering and control of a fed-batch bioreactor under nonideal conditions

Cognitive optimization of microbial PHB production in an optimally dispersed bioreactor by single and mixed cultures

“…While the recurrent neurons and information recycle improve the accuracy of an Elman network, they reduce its speeds of convergence during training and responses to unseen situations. Nevertheless, the Elman network is the best choice, as previous results [8,34,39,40] and the present investigation have shown.…”

“…Sometimes it is helpful to include some bias neurons in these layers so as help the network ride over local minima or shallow troughs in its quest for a global optimum. Obviously the neurons and the flow of signals between them may be arranged in different configurations, and previous studies [8,34,39,40] have demonstrated that an Elman neural network is the most suitable for a reactor with finite dispersion. Such a network has the architecture shown in Fig.…”

“…This approach has been adopted by many earlier studies of the PHB system [1,34,39,40] and other fermentations [42][43][44], which found it useful to explore bioreactor dynamics by generating data by solving a model with laboratory-scale kinetics and nonideal features that mimic real conditions. Finite dispersion in these applications and in the present work was included by employing mass balance equations with dispersion terms expressed by Pe [8,45].…”

Hybrid filtering of feed stream noise from oscillating yeast cultures by combined Kalman and neural network configurations

“…Without noise, they found that oscillations may be periodic, aperiodic or chaotic, depending on the dilution rate and the gas-to-liquid mass transfer coefficient for oxygen. Observations from different production-scale bioreactors indicate [10,22] that the feed streams are major sources of noise; the fluctuations observed may be modeled by Gaussian distributions with mean values equal to the instantaneous deterministic concentrations and different variances [2,3,12]. For S. cerevisiae fermentations, a variance of even 5% suffices to induce stochastic chaos in deterministic oscillations [12,13].…”

“…Understandably, this is more likely with fed-batch and continuous operations than with batch operations. While the effects of noise on fermentations generating monotonic outputs have been investigated in some detail [1][2][3], their effects on oscillating fermentations have not received sufficient attention despite their scientific and industrial importance.…”