Transient-state studies and neural modeling of the removal of a gas-phase pollutant mixture in a biotrickling filter

The removal efficiency (RE) of gas-phase hydrogen sulfide (H), methanol (M) and α-pinene (P) in a biotrickling filter (BTF) was modeled using artificial neural networks (ANNs). The inlet concentrations of H, M, P, unit flow and operation time were used as the model inputs, while the outputs were the RE of H, M and P, respectively. After testing and validating the results, an optimal network topology of 5-8-3 was obtained. The model predictions were analyzed using Casual index (CI) values. M removal in the BTF was influenced positively by the inlet concentration of M in mixture (CI = 3.79), while the removal of P and H were influenced more by the time of BTF operation (CI = 25.36, 15.62). The BTF was subjected to different types of short-term shock-loads: 5-h shock-load of HMP mixture simultaneously, and 2.5-h shock-load of either H, M, or P, individually. It was observed that, short-term shock-loads of individual pollutants (M or H) did not significantly affect their own removal, but the removal of P was affected by 50%. The results from this study also show the sensitiveness of the well-acclimated BTF to handle sudden load variations and also revival capability of the BTF when pre-shock conditions were restored.