Artificial Neural Network Modelling of Biogas Yield from Co-Digestion of Poultry Droppings and Cattle Dung

Abstract

Mechanistic modeling aimed at predicting biogas yield is marred with complex interactions and hence, a very tedious endeavor. Consequently, an Artificial Neural Network (ANN) modeling approach was used to model the relationship among six physico-chemical properties of a mixture of poultry droppings and cattle dung to predict the volume of biogas produced i.e. pH, Total Dissolved Solids, temperature, mass of the slurry, Biochemical Oxygen Demand and Dissolved Oxygen. Three floating drum anaerobic digesters were loaded with 27 varying ratios of a mixture of poultry droppings and cattle dung using batch method, such that the three digester reactors ran nine different batches of mix ratio for a retention period of 27 days each. Slurry temperature, gas and slurry-gas interfaces were monitored using WZP pt100 and DHT 11 sensors installed on an Arduino microcontroller. The 3-layer Feed-Forward model with Back-Propagation Multi-Layer Perception (MLP) architectures, 6-12-1 (6 nodes each in input layer, 12 nodes in hidden layer and single node in output layer) developed for biogas prediction yielded optimal results. The developed model used the default data separation of 60%, 20%, and 20% in Matlab R2015a software. Correlation Coefficient (R) of developed ANN model for biogas prediction were 0.9653, 0.9245, and 0.9842 for training, validation, and test sets respectively. Statistical analysis showed that mass of slurry and TDS had the best correlation with biogas volume (i.e.), while DO and BOD had the least correlation with biogas volume. The developed ANN modelled biogas production from the co-digestion of poultry droppings and cattle dung efficiently.