Machine-learning analysis to predict the fluorescence quantum yield of carbon quantum dots in biochar

Jiao Chen; Mengqian Zhang; Zijun Xu; Ruoxin Ma; Qingdong Shi

doi:10.1016/j.scitotenv.2023.165136

Machine-learning analysis to predict the fluorescence quantum yield of carbon quantum dots in biochar

Sci Total Environ. 2023 Oct 20:896:165136. doi: 10.1016/j.scitotenv.2023.165136. Epub 2023 Jun 26.

Authors

Jiao Chen¹, Mengqian Zhang², Zijun Xu³, Ruoxin Ma¹, Qingdong Shi¹

Affiliations

¹ College of Ecology and Environment, Xin Jiang University, Urumqi 830046, PR China.
² China Energy Conservation and Environmental Protection Group, Beijing 100035, PR China.
³ College of Ecology and Environment, Xin Jiang University, Urumqi 830046, PR China.. Electronic address: xzj@xju.edu.cn.

PMID: 37379935
DOI: 10.1016/j.scitotenv.2023.165136

Abstract

Biochar nanoparticles have recently attracted attention, owing to their environmental behavior and ecological effects. However, biochar has not been shown to contain carbon quantum dots (< 10 nm) with unique photovoltaic properties. Therefore, this study utilized several characterization techniques to demonstrate the generation of carbon quantum dots in biochar produced from 10 types of farm waste. The generated carbon quantum dots had a quasi-spherical morphology and high-resolution lattice stripes with lattice spacings of 0.20-0.23 nm. Moreover, they contained functional groups with good hydrophilic properties, such as amino and hydroxyl groups, and elemental O, C, and N on the surface. A crucial determinant of the photoluminescence properties of carbon quantum dots is their fluorescence quantum yield. Therefore, the relationship between the biochar preparation parameters and the fluorescence quantum yield was investigated using six machine learning analytical models based on 480 samples. Among the models, the gradient-boosting decision-tree regression model exhibited the best predictive performance (R² > 0.9, RMSE <0.02, and MAPE <3), and was used for the analysis of feature importance; compared to the properties of the raw material, the production parameters had a greater effect on the fluorescence quantum yield. Additionally, four key features were identified: pyrolysis temperature, residence time, N content, and C/N ratio, which were independent of farm waste type. These features can be used to accurately predict the fluorescence quantum yield of carbon quantum dots in biochar. The relative error range between the predicted and the experimental value of fluorescence quantum yield is 0.00-4.60 %. Thus, the prediction model has the potential to predict the fluorescence quantum yield of carbon quantum dots in other types of farm waste biochar, and provides fundamental information for the study of biochar nanoparticles.

Keywords: Biochar; Carbon quantum dots; Farm waste; Fluorescence quantum yield; Machine learning; Pyrolysis temperature.