Predictive analysis of methane-enhanced conversion of organic waste into sustainable fuel: A machine learning approach

Abstract

Currently, methane-assisted biomass-to-fuel (M-BtF) processes are being introduced, which are of considerable importance due to the production of an environmentally friendly fuel. Due to the lower price, natural gas can address a stronger potential compared to hydrogen and nitrogen under pyrolysis and bio-oil upgrading processes. Based on this, it seems vital to evaluate such a clean fuel production process from the standpoints of the life cycle analysis (LCA) and techno-economic assessment to exhibit the economic feasibility and environmental superiority. This article aims to utilize machine learning (ML) to predict the yield of organic solid waste, evaluate its economic feasibility, and investigate the environmental sustainability of the valorization of a methane-assisted organic solid waste process. The study employs an artificial neural network (ANN) model with a semi-automatic tuning procedure to forecast the yield of bio-oil, based on biomass characterization. This approach eliminates the need for pilot plant trials for new types of biomass to determine process yields before industrial scale-up. The ANN model demonstrated a high prediction accuracy, achieving an R2 (coefficient of determination) value higher than 0.82. The simulation results, derived from the ML predictions, were then employed in the techno-economic analysis and LCA, where they were compared with other commercial processes for economic viability and environmental impact assessments. The minimum selling price for the renewable diesel was calculated as 3.45 US$/gallon, which is notably competitive, being around 26 % lower than the average minimum selling price observed in relevant publications. Based on Monte Carlo simulation, the plant's internal rate of return (IRR) remained above 50 % across various discount rates, suggesting a highly profitable investment. From LCA results, the proposed method was found to emit ∼ 67 % less greenhouse gases (GHG) compared to standard commercial processes. Overall, the study showcases how the integration of machine learning into process simulation can lead to both economically and environmentally beneficial outcomes, particularly in the context of renewable energy production.