Find in Library

Some microbial systems exhibit sustained oscillations under certain conditions. The maintenance and the suppression of oscillations are both important in different situations. While oscillations are clearly identifiable in small bioreactors, the influx of noise fuzzifies the oscillations in larger vessels. So, noise-filtering devices are employed to recover clear oscillating profiles. Recent work has shown that an auto-associative (AA) neural network is a better than standard algorithmic filters. In this study, nine neural network designs are compared for their ability to filter Gaussian noise in the substrate inflow rate of a continuous fermentation containing Saccharomyces cerevisiae. While the AA network is the best overall, specific performance criteria favor other designs. Thus the choice of a neural filter depends on the evaluation criterion, which is guided by the application.