Experimental assessment and artificial neural network modeling of dynamic and steady-state methane biofiltration in the presence of volatile organic compounds

This study examined the artificial neural network (ANN) modeling of simultaneous biofiltration of methane (CH4) with two volatile organic compounds (VOCs): xylene and ethylbenzene, using an inorganic packed bed biofilter at an empty bed residence time (EBRT) of 4.5 min. Results showed that the removal efficiency (RE) of CH4 was in the range of 50 to 60% for concentrations of 1000 to 10,000 ppmv (0.6 to 6.5 g m-3), while the VOCs-REs were between 70 and 90% for X and EB concentrations in the range of 200 to 500 ppmv (0.9 to 2.2 g m-3). Artificial neural networks were used to predict and simulate the performances of the biofilter, based on a database containing previous biofiltration works. The ANN1 (architecture of 3 (input layer)-18 (hidden layer)-1 (output layer)) accurately predicted CH4 conversion at the pseudo-steadystate condition, while the ANN2 (4 (input layer)-18 (hidden layer)-2 (output layer)) predicted the simultaneous conversion of CH4 and VOCs with slightly lower accuracy than ANN1. The ANN3 (4 (input layer)-30 (hidden layer)-1 (output layer)) successfully predicted the acclimation period and final phase (CH4 concentration of 10,000 ppmv) of the biofilter but could not accurately predict the transient phases and showed differences (up to 20%) from experimental results once the CH4 concentration was changed. This study developed a decision support and prediction tool to anticipate the performance of biofilters in treating residual gases containing CH4 and VOCs, avoiding costs and delays associated with experimentation.