Predictive modeling of biofilters treating benzene vapours using artificial neural networks

Biofiltration is a simple, reliable and cost effective treatment process for the removal of hazardous volatile air contaminants. The variability of feed conditions and performance of biofilters make the application of mathematical models very useful. The lack of information on the micro kinetic and metabolic mechanisms has restricted the use of process based models. Data driven models such as Artificial Neural Networks (ANNs) can predict the performance of biofilters by their ability to map the input-output relationships. This study predicts the performance of a compost based biofilter. operated continuously for 155 days with varying loading rates of benzene vapours in terms of the elimination capacity (EC) and removal efficiency (RE). Two ANN models having three layers were trained using the back propagation algorithm and by changing the network parameters. The best network architecture was selected based on low root mean square value (RMS) in the test data. The two models fitted the data well with RMS values of 0.0845 and 0.0765 for EC and 0.1183 and 0.0874 for RE respectively. The results suggest that ANNs can be a useful predictive tool for modeling biofilters. Copyright © IICAI 2005.