A New Bee Colony-WNN Structure for Identification and Modelling of a Yeast Fermentation Bioreactor.

This article focuses on developing an accurate model for identifying and controlling a yeast fermentation bioreactor, a severely nonlinear chemical process important in food and biofuel industries. It proposes a novel hybrid algorithm combining the Artificial Bee Colony (ABC) optimization method with Wavelet Neural Networks (WNN), termed ABC-WNN, to optimize neural network initial weights and improve model accuracy. Using input-output data generated by persistently exciting signals, the ABC-WNN model demonstrated over tenfold improvement in prediction error and enhanced goodness of fit compared to conventional backpropagation-trained neural networks (BP-ANN). The study highlights the hybrid algorithm's ability to escape local minima by leveraging ABC's global search and WNN's local refinement, while noting limitations related to parameter tuning and computational demands. This approach offers a promising tool for modeling complex nonlinear bioprocesses where precise dynamic models are critical for effective control.