Machine learning-aided unveiling the relationship between chemical pretreatment and methane production of lignocellulosic waste
Chao Song,Zhijing Zhang,Xuefeng Wang,Xuejun Hu,Chang Chen,Guangqing Liu
DOI: https://doi.org/10.1016/j.wasman.2024.07.004
2024-10-01
Abstract:Chemical pretreatment is a common method to enhance the cumulative methane yield (CMY) of lignocellulosic waste (LW) but its effectiveness is subject to various factors, and accurate estimation of methane production of pretreated LW remains a challenge. Here, based on 254 LW samples, a machine learning (ML) model to predict the methane production performance of pretreated feedstock was constructed using two automated ML platforms (tree-based pipeline optimization tool and neural network intelligence). Furthermore, the interactive effects of pretreatment conditions, feedstock properties, and digestion conditions on methane production of pretreated LW were studied through model interpretability analysis. The optimal ML model performed well on the validation set, and the digestion time, pretreatment agent, and lignin content (LC) were found to be key factors affecting the methane production of pretreated LW. If the LC in the raw LW was lower than 15%, the maximum CMY might be achieved using the NaOH, KOH, and alkaline hydrogen peroxide (AHP) with concentrations of 3.8%, 4.4%, and 4.5%, respectively. On the other hand, if LC was higher than 15%, only high concentrations of AHP exceeding 4% could significantly increase methane production. This study provides valuable guidance for optimizing pretreatment process, comparing different chemical pretreatment approaches, and regulating the operation of large-scale biogas plants.