An integration of machine learning models and life cycle assessment for lignocellulosic bioethanol platforms

Abstract

The availability and accuracy of data play a critical role in the development of Life Cycle Assessment (LCA) models, as they directly impact the effectiveness and reliability of the model. Machine learning (ML) approach has the potential to improve the data quality, leading to enhanced performance of LCA models. In this study, ML algorithms (random forest, extreme gradient boosting, and artificial neural network) were applied to predict the ethanol yield and NaOH consumption, which were key life cycle inventories for lignocellulosic bioethanol process. The prediction models exhibited a decent level of accuracy for ethanol yield (accuracy ∼ 0.85) and NaOH consumption (accuracy > 0.8). In addition, ML algorithms (genetic algorithms, particle swarm optimization, and simulate annealing) were applied to identify optimal feedstock characteristics and conditions for maximum ethanol yield. The optimization models showed an 18% improvement compared to the highest yield from the dataset (0.41 g/g). The ML models (prediction and optimization algorithms) were further integrated with the LCA model (GREET1) to enhance the ethanol yield and to reduce greenhouse gas (GHG) emission for target feedstocks. This resulted in a 6% and 19% improvement in ethanol yield and an approximately 8% and 14% reduction of GHG emissions for grass and corn stover, respectively. This study demonstrates the feasibility of utilizing ML to predict LCA metrics for the development of LCA models and to optimize the process parameters to minimize environmental impact.