RBC effluent quality representation using Artificial Neural Networks

Effluent quality modelling is an important aspect of any water/waste water treatment unit. The process under consideration may be either under a dynamic state or it can be assumed to be under quasi-steady state. Investigators have assumed a quasi-steady state condition for relating substrate flux to the effluent quality. Subsequently, the use of such relationships has been made to optimise RBC disc area. It is not known whether such substrate flux models can be used to represent the effluent quality. The present study has been taken with the expectation that if substrate flux models are successful in representing the effluent quality, one might take advantage of these in understanding the response of the RBC system to different loading patterns. Here, the emphasis has been given to the correct representation of the effluent quality. In this regard, the use of popular Variable Order Models has been examined. It is shown why the use of substrate flux models may fail to represent the effluent quality. The relationships between the input loading and the effluent quality may not always be in one to one correspondence. Also, the solution may not exist in few cases. The effluent quality representation may not necessarily require the dynamics of the individual stages and the problem can be easily handled by treating RBC system as one unit. Contrary to the use of conventional models, Artificial Neural Networks (ANNs) which also have the ability to relate the inputs to the output (effluent quality), has been also assessed. The performance of ANNs in representing the effluent quality has been found to be excellent in different stages as well as for the case of an equivalent single stage representation of the RBC unit. Also, the need for trial and error solutions which is inevitable in conventional models, gets eliminated using ANNs. In the present study, the focus has been on the representation of effluent quality from a RBC pilot plant. The study requires further investigations to test the utility of the suggested approaches in a predictive sense.