IDENTIFICATION AND CONTROL OF ANAEROBIC DIGESTERS USING ADAPTIVE, ONLINE TRAINED NEURAL NETWORKS

This paper introduces an anaerobic digestion identification and contro l scheme, based on adaptive, on-line trained neural networks. Anaerobi c digestion is a complex, nonlinear biochemical process, widely used f or the treatment of organic sludge in municipal wastewater treatment p lants. Conventional control schemes usually fail to overcome the typic al difficulties encountered in systems with complex nonlinear dynamics and difficult-to-measure or time varying parameters. It is shown by s imulation results that, under a predictive control approach, adaptive on-line trained neural networks are successful in tackling such proble ms. in the case of anaerobic digestion. The proposed control scheme fe atures desired tracking, regulation and robustness properties in vario us anaerobic digestion control tasks, including set points or process inputs variations, even in the presence of measurement noise or in cas es of process parameter changes. In addition, the performance of three training algorithms, the back-propagation and two different random op timisation techniques, is examined over the neural controller training task. In all cases the random optimisation techniques converge much f aster than the back-propagation algorithm. Copyright (C) 1996 Elsevier Science Ltd