Machine learning analysis of biofuels using published data

One of the key objective of this dissertation is to develop a comprehensive understanding of previous machine learning (ML) studies on various biofuel production routes, specifically microalgae biodiesel, lignocellulosic ethanol, and biochar production, through bibliometric analysis and literature review. Microalgae show high potential for biofuel production but are costly, requiring better strains, value-added chemicals and reduced energy consumption for commercial viability. Converting lignocellulosic biomass to ethanol is promising but remains expensive, particularly due to pretreatment step. Biochar, especially in its functionalized and engineered forms, is likely to become more important for environmental remediation. This work also aims to construct comprehensive datasets for lipid production from the oleaginous yeast strain Yarrowia lipolytica and tri-reforming process from biogas, analyzing them with ML tools to estimate unperformed experiments, determine critical variables, and develop optimal heuristics. It was found that the potential of Yarrowia lipolytica for lipid production can be effectively categorized using decision tree (DT) algorithms, and descriptor effects can be determined using association rule mining. Found that, C/N ratio with fermentation time being influential for biomass production, while pH and glucose concentration for lipid content. The tri-reforming models developed using random forest (RF) and artificial neural network (ANN) algorithms also showed high predictive capabilities. Transfer learning methods, using source models from dry-reforming and steam reforming datasets were implemented to enhance the prediction accuracy with smaller datasets. DT algorithms were employed for interpretable knowledge extraction, resulting in the development of hierarchical rules for methane conversion, carbon dioxide conversion, and H2/CO ratios in the produced syngas. While high reaction temperature with high W/F results with high CH4 conversion, moderate reaction temperature with low O2 ratio promote high CO2 conversion. In all cases, not depending on other conditions, increasing H2O, and decreasing CO2 in the feed composition increases H2/CO ratio. The findings align with the literature, indicating that data mining and machine learning methods effectively provide predictive models and detect trends and patterns that are not readily observable.