Modelling the removal of volatile pollutants under transient conditions in a two-stage bioreactor using artificial neural networks.

A two-stage biological waste gas treatment system consisting of a first stage biotrickling filter (BTF) and second stage biofilter (BF) was tested for the removal of a gas-phase methanol (M), hydrogen sulphide (HS) and α-pinene (P) mixture. The bioreactors were tested with two types of shock loads, i.e., long-term (66h) low to medium concentration loads, and short-term (12h) low to high concentration loads. M and HS were removed in the BTF, reaching maximum elimination capacities (ECmax) of 684 and 33 gm-3h-1, respectively. P was removed better in the second stage BF with an ECmax of 130 gm-3h-1. The performance was modelled using two multi-layer perceptrons (MLPs) that employed the error backpropagation with momentum algorithm, in order to predict the removal efficiencies (RE, %) of methanol (REM), hydrogen sulphide (REHS) and α-pinene (REP), respectively. It was observed that, a MLP with the topology 3-4-2 was able to predict REM and REHS in the BTF, while a topology of 3-3-1 was able to approximate REP in the BF. The results show that artificial neural network (ANN) based models can effectively be used to model the transient-state performance of bioprocesses treating gas-phase pollutants.