Sustainable biodiesel synthesis from multi-feedstock systems: Integrating convolutional neural networks with life cycle assessment for energy-climate trade-off analysis

Abstract

Biodiesel is widely regarded as a viable alternative to fossil fuels and a strategic approach to mitigating climate change. Its production, primarily through the transesterification of plant- or waste-based biomass, is influenced by a variety of experimental factors, including catalyst type, reaction temperature, and feedstock properties. However, conducting the optimization of experimental procedures and comprehensive environmental assessment remains challenging due to the diversity of feedstocks and the variability in reaction conditions, which introduce significant complexity and uncertainty in increasing the production of biodiesel while reducing its environmental impact. To address this challenge, in this work, the integration of machine learning (ML) and life cycle assessment (LCA) in biodiesel production offers a powerful approach to improving both environmental and economic performance. We compared the predictive accuracy of five ML models for biodiesel yield based on transesterification reaction. The results indicated that the Convolutional Neural Network could be a more suitable model for accurate biodiesel yield prediction because of its lowest validation RMSE (5.667) and highest correlation coefficient value (0.9165). In addition, we analyzed the environmental impact of different biomass conversion though LCA. The LCA showed methanol consumption contributed over 90 % of environmental impacts in virgin oils, while waste cooking oil's main burden came from regeneration, contributing 41.0–99.9 % depending on the impact category. This ML-LCA coupling not only improves the accuracy of process optimization but also provides a holistic view of environmental trade-offs across diverse feedstocks, paving the way for smarter and greener biodiesel production strategies.