Premium
Start‐up and recovery of a biogas‐reactor using a hierarchical neural network‐based control tool
Author(s) -
Holubar Peter,
Zani Loredana,
Hager Michael,
Fröschl Walter,
Radak Zorana,
Braun Rudolf
Publication year - 2003
Publication title -
journal of chemical technology and biotechnology
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 0.64
H-Index - 117
eISSN - 1097-4660
pISSN - 0268-2575
DOI - 10.1002/jctb.854
Subject(s) - biogas , anaerobic digestion , bioreactor , process engineering , methane , chemical oxygen demand , environmental science , artificial neural network , biogas production , process (computing) , waste management , pulp and paper industry , computer science , engineering , sewage treatment , chemistry , environmental engineering , organic chemistry , operating system , machine learning
Due to its intricate internal biological structure the process of anaerobic digestion is difficult to control. The aim of any applied process control is to maximize methane production and minimize the chemical oxygen demand of the effluent and surplus sludge production. Of special interest is the start‐up and adaptation phase of the bioreactor and the recovery of the biocoenose after a toxic event. It is shown that the anaerobic digestion of surplus sludge can be effectively modeled by means of a hierarchical system of neural networks and a prediction of biogas production and composition can be made several time‐steps in advance. Thus it was possible to optimally control the loading rate during the start‐up of a non‐adapted system and to recover an anaerobic reactor after a period of heavy organic overload. During the controlled period an optimal feeding profile that allowed a minimum loading rate of 6 kg COD m −3 d −1 to be maintained was found. Maximum loading rates higher than 12 kg COD m −3 d −1 were often reached without destabilizing the system. The control strategy resulted simultaneously in a high level of gas production of about 3 m 3 biogas m −3 reactor and a methane content in the biogas of about 70%. To visualize the effects of the control strategy on the reactor's operational space the data were processed using a data‐mining program based on Kohonen Self‐Organizing Maps. Copyright © 2003 Society of Chemical Industry