Abstract:
The anaerobic digestion of chicken manure and rice straw presents a transformative
opportunity for producing carbon-neutral biogas that aligns with global green energy
initiatives. AD was extensively employed to remediate organic waste offering the dual
benefits of generating renewable energy and nutrient-dense digestate side by side the
process also helped to utilize the organic and agricultural waste in a very efficient way that
followed the realm of waste to energy however, the process encounters instability which
negatively impacts biogas production, to control and stabilize AD process machine
learning has gained considerable attention in optimizing the process. Machine learning was used to predict the CH4 (ml) with the application of Linear, Non-Linear and Ensemble
learning regression models using sets of features including Days of digestion, pH, COD,
TAN, FAN, TVFA as independent variables and CH4 (ml) as a dependent variable obtained from the experimentation results of 70 days of digestion period in continuous stirring tank reactors for the process optimization of carbon-nitrogen ratio at different organic loading rate stages. The results demonstrated that C/N 24:1 was optimal for the efficient production of CH4 under a continuous feeding rate. The study aims to propose the appropriate ML model from the comparison of 4 applied models from 70 samples in each reactor dataset; Important attributes are indicated by the pairwise Pearson correlation metrics heatmap. The ensemble learning models outperformed linear and non-linear regression models with a coefficient of determination (R2) on training, testing and validation respectively 0.99, 0.96 and 0.84. Experimental results confirmed operational attributes revealed by Reactor D with the highest specific methane yield of 126.50% of the predicted value. Random Forest feature importance elucidates total volatile fatty acid in Reactor A-B while pH in Reactor C-E is the important feature influencing the process. Ultimately, this research illustrates the efficacy of ML models for optimizing biogas production in AD providing valuable insights into improving the whole mechanism and enhancing methane yield from organic matter.
