Investigation of sawdust microwave-assisted pyrolysis by machine learning, Part I: Optimization insights by large language models

Abstract

Microwave-assisted catalytic pyrolysis shows promise for efficiently converting waste biomass, such as sawdust, into valuable bio-oil. However, current research on pyrolysis characteristics predominantly relies on conventional trial-and-error experimentation. This work pioneers the first use of large language models (LLMs) to gain insights into optimizing bio-oil yield by analyzing the effects of catalyst loading and pretreatment temperature on product distribution. Additionally, we encode the textual LLM outputs into distributed vectors via Word2Vec and concatenate them with artificial neural network (ANN) embeddings, capitalizing on the complementary strengths of data-driven and language models. Our results demonstrate superior accuracy and interpretability, providing better optimization insights on heating dynamics and energy efficiency while avoiding extensive experimental costs. This research establishes the prospects of LLMs as supplementary tools for thermochemical conversion studies, potentially reducing resource-intensive lab trials through reliable data-driven predictions.